diff --git a/mini-beatrix-3/arm_refit/README.md b/mini-beatrix-3/arm_refit/README.md index 0ac030876ac06e27fd17ca49e1a39abcf0232fbf..7682e327fe3fdc0b682b24f19890dfb2af152f02 100644 --- a/mini-beatrix-3/arm_refit/README.md +++ b/mini-beatrix-3/arm_refit/README.md @@ -12,8 +12,9 @@ weights and loads with [amoe-lora](https://github.com/AbstractEyes/amoe-lora). ## The runs -Stages 1 to 4 are perspective, concepts, rule chains and arithmetic. A run -name ending in `A` or `B` is one of two seeds of the same recipe. +Stages 1 to 4 are perspective, concepts, rule chains and arithmetic; stages +5 to 8 are cause and effect, try-fail-revise, the mixed stage and register. +A run name ending in `A` or `B` is one of two seeds of the same recipe. | run | what was trained | arm seeds | data seed | |---|---|---|---| @@ -21,6 +22,9 @@ name ending in `A` or `B` is one of two seeds of the same recipe. | solo, `anchor` | one arm alone per stage, continued from the arm that was trained beside the trunk during the run (stages 1 to 4) | 1 to 4 | 20261005 | | solo, `pc_s1` | a repeat of an earlier single-arm control on stage 1 text (400 steps), run first to check the tool against a known result | 1 | 20260930 | | `solo_long/` | one solo arm for 3,200 steps with a reading every 400 (stage 3, stage 4), to see where an arm levels off | 3, 4 | 20261005 | +| `solo_4000/` | one solo arm per stage at the arm length used in the certified ladder, 4,000 steps, kept quiet on web text | 1 to 8, 101 to 108 | 20261005, 20261006 | +| `gXA`, `gXB` | the eight: the four arms of `gRA` / `gRB` switched on and frozen; new arms for stages 5, 6, 7 and 8 trained beside them one after the other, 4,000 steps each, each kept quiet on the stage text of every other arm | 5 to 8, 105 to 108 | 20261011, 20261012 | +| `gEA`, `gEB` | all eight arms from scratch: frozen in turn as in `gS`, every arm kept quiet on every other arm's stage text from its first step, 4,000 steps per stage | 1 to 8, 101 to 108 | 20261013, 20261014 | | `gA`, `gB` | co-trained group: the arms of stages 1 to 4 attached one stage at a time, each new arm over the earlier ones, every attached arm training, 800 steps per stage | 1 to 4, 101 to 104 | 20261005, 20261006 | | `gFA`, `gFB` | the same group with the earlier arms switched on but frozen while the newest trains | 1 to 4, 101 to 104 | 20261005, 20261006 | | `gSA`, `gSB` | frozen in turn as in `gF`, and every arm kept quiet on the stage text of every other arm from its first step | 1 to 4, 101 to 104 | 20261005, 20261006 | @@ -46,7 +50,7 @@ was last updated. | `group/results/routed_.json`, `group/anchors/routed_.dispatch.pt` | a routed reading of a finished group and its trained dispatch keys | | `solo/anchors/__seed_d_step.safetensors` | a solo arm at step 400 and at step 800 | | `solo/results/result___seed_d.json` | the solo run's measurements at steps 0, 200, 400 and 800 | -| `solo_long/` | the same two kinds of file for the 3,200-step solo runs | +| `solo_long/`, `solo_4000/` | the same two kinds of file for the 3,200-step and the 4,000-step solo runs | | `logs/` | the runs' own logs | All losses are bits per byte on held-out text. diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXA_close5_s1_perspective.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXA_close5_s1_perspective.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..107eb0338f3e7babff6374697188b07b912af693 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXA_close5_s1_perspective.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1d30faf8dc9f1f39ad8a661b8444bfeb0de7443ea016dd7f32c925be1479d0aa +size 54819328 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXA_close5_s2_concept.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXA_close5_s2_concept.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..cbb3b7683efbb530ea0324982de1a123d203d408 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXA_close5_s2_concept.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bf5346caf49d4a4ef6bafdb40769ed1a8acb6543f96a1b5b100b06e104aca23e +size 54819320 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXA_close5_s3_rules.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXA_close5_s3_rules.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..835046011e528bb06a29ce4ee71efe0d9a365c47 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXA_close5_s3_rules.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bf6c1d634a100b0a610ae20421818df0482c936193d786754175839f1fcdf643 +size 54819312 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXA_close5_s4_arith.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXA_close5_s4_arith.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2c7cc7572a20074a6a6169a14762a32675ddc5a6 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXA_close5_s4_arith.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:50f9039f8557fa8f66be3674bb9591461e8e6f7d5c317b1358979dc78f3b696c +size 54819312 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXA_close5_s5_causal.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXA_close5_s5_causal.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..464a9ff7c823cb0a9e7f6e35e13515e540816963 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXA_close5_s5_causal.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:340b185ad3dac8202b0c9e5d17ab6d90c7149b844609268c828a0f971d0c15fa +size 54819320 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXA_close6_s1_perspective.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXA_close6_s1_perspective.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c21b10cef805695a43f13f74b66f9a328d39542f --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXA_close6_s1_perspective.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:366182ddee94012f6de384a8d4ffa68c185b31eeb701e4717247725ab5193fd2 +size 54819352 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXA_close6_s2_concept.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXA_close6_s2_concept.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..8832538967bbef196a2d45e1208fad42daa645b1 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXA_close6_s2_concept.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b5bf1b72a4933550091eac32b1d88016764ed0bbd58d4516dd0611451c618f34 +size 54819344 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXA_close6_s3_rules.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXA_close6_s3_rules.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..1cdec708fba860566a8218804f159f295796fe65 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXA_close6_s3_rules.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:190f03efc68b1077ff3411dba966497dad572a9731b06cb53aeb72ccdb106179 +size 54819336 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXA_close6_s4_arith.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXA_close6_s4_arith.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..700628bee5dd005aad76b7f98f57e3e6cc87e3d5 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXA_close6_s4_arith.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1b9c7f3e35f0b1458042ea413a1dc4a38d697aa7544c0cb5b77077a608703ff6 +size 54819336 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXA_close6_s5_causal.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXA_close6_s5_causal.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..da2e9dd34a2ca44273391f7fe324aa5627530d70 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXA_close6_s5_causal.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9b4a3f4c29db6c4a3f44c95b1fff2e15a83e4059845b89db2368c88448610d50 +size 54819336 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXA_close6_s6_tryfail.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXA_close6_s6_tryfail.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..bd8fbd1db6ca7d2e933d4517e717701c277dd3a6 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXA_close6_s6_tryfail.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cbfaad1a44a728aeb23d0dd838e112d461314d8fb7514a411fbbdb7538f551b1 +size 54819344 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXA_close7_s1_perspective.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXA_close7_s1_perspective.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..0a2d341d7a5b509e1bd6a511708ed5ece6f1b4ee --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXA_close7_s1_perspective.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:721d4bff6fdad5430122e452cd257eadedced77ebf1ffe9e5bc0ae0994911e45 +size 54819368 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXA_close7_s2_concept.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXA_close7_s2_concept.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..1a0814f179f657c0eed534070951c1b7149b1ed7 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXA_close7_s2_concept.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7a5d6a9c1d6768e9e4799c078f68571afb7b23f2c536d25a101e47bb399f27e8 +size 54819360 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXA_close7_s3_rules.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXA_close7_s3_rules.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b829523d44acbd57190b4f07e01af611c590f1f1 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXA_close7_s3_rules.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f80c2525685a44e700d5877286cbcf3c4e60898f5297bc65556d14251e6d7feb +size 54819352 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXA_close7_s4_arith.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXA_close7_s4_arith.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f26ce0015d02e74380f1b8395da387d7b8195a6a --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXA_close7_s4_arith.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:debf573ecfee9d03e869c3f9dcd63332788718338050eb499cc8d2185e53aa18 +size 54819352 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXA_close7_s5_causal.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXA_close7_s5_causal.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..6cc037967e298b6f0f1448d40871d587e9921bcc --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXA_close7_s5_causal.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a79faf394275c44e4cca587a5a53f282a02f0c44129524fb21b02ea14befc39b +size 54819360 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXA_close7_s6_tryfail.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXA_close7_s6_tryfail.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..6499d3aa0e952db5c66336fe78f7ad2e0b5f8779 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXA_close7_s6_tryfail.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f3b0754fda7c717d082f9342fd6048d09edb42c178b2624d1f0379638efedf95 +size 54819360 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXA_close7_s7_mixed.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXA_close7_s7_mixed.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..339b782c9a3a4162a7b3a9af800241936fc3ba9e --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXA_close7_s7_mixed.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bf988781ac08734061dba7f9ab35c994d29eb8fa906f3b0f6c572a173295333c +size 54819360 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXA_close8_s1_perspective.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXA_close8_s1_perspective.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2183d65214d8f00864922d18bedae65ba49d8c4d --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXA_close8_s1_perspective.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:65cb5c57cfd64fdbaaaf47d32a37969adc91bb2b1071cc97e089943041ebe044 +size 54819392 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXA_close8_s2_concept.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXA_close8_s2_concept.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ad37e8d4550c3a592bb83dc3406b79f97198c5e6 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXA_close8_s2_concept.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f411b5c9ae9994a26e0cbd25475a754fee5af6b7e58fb2af5dedbe790dd602f5 +size 54819384 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXA_close8_s3_rules.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXA_close8_s3_rules.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..22fec9c2dd0e743de56aea0f3a2fd2d339aceac8 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXA_close8_s3_rules.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bd9f4dd7e15f28ca140e1c86962b0f4ea4b144a8af369219acd270b1368dc5b8 +size 54819384 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXA_close8_s4_arith.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXA_close8_s4_arith.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..88b72f924be362c447b762444db04f043d9c78c6 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXA_close8_s4_arith.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:58552d2cdf11bce31dcad8e1e09fe27b6c760857aef47add6b4c92b0487669b7 +size 54819384 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXA_close8_s5_causal.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXA_close8_s5_causal.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ee9f90c77b142b9c60d81d974dec71d4f14dacc0 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXA_close8_s5_causal.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8d1da9acaf7d528243304d7bebf71780242c7b1622ab3a1f72ef39e875b0f732 +size 54819384 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXA_close8_s6_tryfail.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXA_close8_s6_tryfail.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..a34faf9d1621e5c19d90326b7deb6f57104f241d --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXA_close8_s6_tryfail.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2b7942956f7e2ded57b29168ef5a0916424a39bd3826ad7804fabec92050999b +size 54819384 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXA_close8_s7_mixed.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXA_close8_s7_mixed.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4c09c41307547909f4c134df457d450215bc51e3 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXA_close8_s7_mixed.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b7a4210d7ab9690a5fbddf847abbe60a47755d9d8446dbbb04fe6ba469e8f23c +size 54819384 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXA_close8_s8_register.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXA_close8_s8_register.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ce3df06679f1eba44b568c690e2c4d1b462fe2a7 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXA_close8_s8_register.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:19b02efab5fb482cd50d10508cda56868d8bb1f7fce835a0b7e9babec182aa3b +size 54819384 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXB_close5_s1_perspective.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXB_close5_s1_perspective.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9a14b856df1d3e1eeaeccb83d1a02f02d74ab935 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXB_close5_s1_perspective.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:724da745bbc59aaa175fed285b4fd283274bacadec0add71b73a544824f500f7 +size 54819328 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXB_close5_s2_concept.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXB_close5_s2_concept.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..a15f787b3a2ff8ded4a78d5f74255651cb76ba82 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXB_close5_s2_concept.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ed9a7b08cebcca3d26070904395cc22369ff1c09ae4d81e3322f8c9f5c39c95d +size 54819320 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXB_close5_s3_rules.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXB_close5_s3_rules.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..dc38fec70761d06b9580a5c1050dd1f78ab80a5e --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXB_close5_s3_rules.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2f5b5002e4aea2ff6cf4807958063c700cadd23cc67c3aa61b170a05db63fa9c +size 54819312 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXB_close5_s4_arith.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXB_close5_s4_arith.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..7dad0493d750319ebffa03198cd536434f66ee4b --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXB_close5_s4_arith.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c79a15c43654dee2abb5e22eb8723f3e6b46d83adf19a95d5eb7cfe1df47a8ff +size 54819312 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXB_close5_s5_causal.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXB_close5_s5_causal.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..dd2354acfb018fd66f940a5f26af92412267ca35 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXB_close5_s5_causal.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1e5c9bbdf04af1a7c57a176bf3e7513beda5a3b50c2a0ef60b3e0af59ed7921d +size 54819320 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXB_close6_s1_perspective.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXB_close6_s1_perspective.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c2957bcab73f441bda594075b4116c0d6f841640 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXB_close6_s1_perspective.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:40897bcec458cdf6a5bc93e070c689200a1e90d195b87c7921cc49d60fdb63c7 +size 54819352 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXB_close6_s2_concept.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXB_close6_s2_concept.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9905463eceb785ec98da66df1c3be5a5b62157af --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXB_close6_s2_concept.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3c67e40de9b1e64a1ef484bd28e5b35ce4d1eca493aa59502d209d6c631fd9ab +size 54819344 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXB_close6_s3_rules.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXB_close6_s3_rules.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..a7de540a6dacc0c1beb67794d3586df378537053 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXB_close6_s3_rules.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f72135d94d39333e5f34d8205f3ff52ca0db60e6159e7037af925af46303b44f +size 54819336 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXB_close6_s4_arith.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXB_close6_s4_arith.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..a3c44ff40b5b1be5ba8cb329555265dea48d2eff --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXB_close6_s4_arith.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3755e97b62c903591cf55b2465e231957b0cc772acbe3c11f97ff9f5ee3113cc +size 54819336 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXB_close6_s5_causal.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXB_close6_s5_causal.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9c74c4786ed3cfd2db336b9efd2e023cd1e668f8 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXB_close6_s5_causal.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1fcef1196f31b47260a0e0b9a3c9f1e91978f2058c1c3e87ebb0a137f7ece8b1 +size 54819344 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXB_close6_s6_tryfail.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXB_close6_s6_tryfail.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..855376100851b4b37fe161483b4f311fb4600a4a --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXB_close6_s6_tryfail.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d955e310d38fa625ebd7c72ea995adc072eda01ee246573007eb0b1465d1dd2a +size 54819344 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXB_close7_s1_perspective.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXB_close7_s1_perspective.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..74e5832dd914a792032f03d018eea683b00050a6 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXB_close7_s1_perspective.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:edf78c298fe8a04e9cb072c22a1edfd12c06ce7b85d90bbce6c41a5a13510794 +size 54819368 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXB_close7_s2_concept.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXB_close7_s2_concept.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4ba18beb79989b55132a41fcb9cfd1d5c01cea5c --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXB_close7_s2_concept.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c36fa620e1393ac38b3dfc5dc43144dcdf29944fd3d133d67fba44aa73949848 +size 54819360 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXB_close7_s3_rules.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXB_close7_s3_rules.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c3f3a4a1dc0b56e5c7f20f560d008cbc6971a648 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXB_close7_s3_rules.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bf4b884471a57a7a404b1242dab5d54362481d715c6ebf9577249255c902d3ab +size 54819360 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXB_close7_s4_arith.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXB_close7_s4_arith.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..7be026a7591d64553f144bf8edb4ee54e2a722c2 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXB_close7_s4_arith.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:47ab5819dacd7a4bf266032c6fd38c9179feb4112a99da151e671365409446e9 +size 54819360 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXB_close7_s5_causal.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXB_close7_s5_causal.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..6d823a4c45b2327a3b3faaa758935d5ec3940f7c --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXB_close7_s5_causal.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0cc03db507edcdc5f820b285654036b6c01949ee69c3ae8f55ba886d3a32dce1 +size 54819360 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXB_close7_s6_tryfail.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXB_close7_s6_tryfail.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..52b1457904bc2759b83e6bbaf625bf2094cab1a6 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXB_close7_s6_tryfail.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8feb5e9dafd746edd7a8e27632bc99f2a1705aec4b3ded2a3623a3575129f67b +size 54819360 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXB_close7_s7_mixed.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXB_close7_s7_mixed.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9f98378b2379f1d56c9a9cd480d1d02b8f0c004b --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXB_close7_s7_mixed.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c68798a17ddc470dd53869401a090e975b82deeb2722e39fcf8e38cf40e4164f +size 54819360 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXB_close8_s1_perspective.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXB_close8_s1_perspective.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..135ff35a6860ff1ceaf966d23e28347a0eae1f13 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXB_close8_s1_perspective.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e65fd401bc34494812f51ae16f0ebada5e9480c6d6ef020aadcd9d5350e5b9ec +size 54819392 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXB_close8_s2_concept.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXB_close8_s2_concept.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..8682a1404911f05e6c23d20a61df790c38325135 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXB_close8_s2_concept.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e62782904abbea5c52cf90069f82a82c34e28e35bed25e86f71fc028963e9ade +size 54819384 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXB_close8_s3_rules.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXB_close8_s3_rules.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9392d67051cafe4dba8205e1f63ad87edddb6407 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXB_close8_s3_rules.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f529b803a0a65a3be179e803dcd17a5411259164058f9560d2a2d24acfacb1f5 +size 54819384 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXB_close8_s4_arith.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXB_close8_s4_arith.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4e46fd7e170fe6b99bc09fae9fc9cfd2705c4a17 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXB_close8_s4_arith.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:04e171fcd5bd095c60662773d4a0079150259f5be8653a4f9ed5bfdac6a19a81 +size 54819384 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXB_close8_s5_causal.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXB_close8_s5_causal.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..fb3e33d43eaf4f007f9b7c984f957ea649b55622 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXB_close8_s5_causal.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9f23fc96bc57deda52b5c1d1311358416bd984071f67cf8f9f1feb19e8652302 +size 54819384 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXB_close8_s6_tryfail.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXB_close8_s6_tryfail.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b4942ddbb722bce18b83f3247d6a5e2d1e5b12d8 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXB_close8_s6_tryfail.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:22b21b78b85923d481a692d72aff2aedf1dd88601d1feb8bae6719a0d38c805d +size 54819384 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXB_close8_s7_mixed.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXB_close8_s7_mixed.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..25165b40de111b934616a3bd5d27655810e85673 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXB_close8_s7_mixed.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1e05656147f7f9f4bb39b2ee2c6172135270682a0b4b73cad6cfc60abbc41f9e +size 54819384 diff --git a/mini-beatrix-3/arm_refit/group/anchors/gXB_close8_s8_register.safetensors b/mini-beatrix-3/arm_refit/group/anchors/gXB_close8_s8_register.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2030b20444ee87506df16df062dcc09ed4e6d1e9 --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/anchors/gXB_close8_s8_register.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ae08d9f09ce847c1a146b05ab2b7f7d2dceb39a83e4bb918c76beb24dacd1335 +size 54819392 diff --git a/mini-beatrix-3/arm_refit/group/results/group_gXA.json b/mini-beatrix-3/arm_refit/group/results/group_gXA.json new file mode 100644 index 0000000000000000000000000000000000000000..06f080ad2ccfe4c990da293323caa2d9e266f44d --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/results/group_gXA.json @@ -0,0 +1 @@ +{"run": "gXA", "stages": ["s1_perspective", "s2_concept", "s3_rules", "s4_arith", "s5_causal", "s6_tryfail", "s7_mixed", "s8_register"], "seed_offset": 0, "data_seed": 20261011, "steps_per_stage": 4000, "ctx": 4096, "chunk_rows": 2, "lambda": 2.0, "lr": 0.001, "spec": {"n_slots": 16, "K": 16, "D": 8, "hidden": 256}, "trunk_step": 245674, "smoke": false, "resume": "gRA:4", "no_partner_rows_of": ["s7_mixed"], "freeze_earlier": true, "resume_members": 4, "offdomain": "roster", "options": "the first 4 members loaded from gRA:4 (the later members attach fresh); a member's partner rows come from every other member of the roster, attached or not", "form": "staged group, earlier arms frozen: each stage's arm attaches fresh over the arms already attached, which stay on and stop training; per step the newest member takes one task chunk of its own stage text (all on) and one abstention chunk (one web row + one row of a partner's stage text; KL from this member masked, partners as is, to all on); step loss = task + lambda * KL", "off": {"rows": {"s1_perspective": {"mean": 0.5538333013185592, "se": 0.005481125510495537, "per_batch": [0.5580802623967968, 0.5727959149222103, 0.5683326211670187, 0.5592622991561769, 0.5403289870966665, 0.5581498723747335, 0.5248631032525267, 0.5488533501823444]}, "s2_concept": {"mean": 0.7211839326967642, "se": 0.0036750359411895426, "per_batch": [0.7178970755158839, 0.7206461751942632, 0.7084789615893348, 0.7429370185552426, 0.7197164369836988, 0.7196987227706599, 0.7128483098214926, 0.7272487611435373]}, "s3_rules": {"mean": 0.9296131361815035, "se": 0.006694211546016496, "per_batch": [0.9087232204987897, 0.9265088063377671, 0.9502096494618454, 0.9083145037290143, 0.9115083935381875, 0.9565801448237801, 0.9359078301900705, 0.9391525408725736]}, "s4_arith": {"mean": 1.2338748670942346, "se": 0.013664467393095502, "per_batch": [1.2484452802709944, 1.2007697976426643, 1.2552932854630496, 1.2180985975609502, 1.162178696606884, 1.2806971007959715, 1.2668025364414397, 1.2387136419719225]}, "s5_causal": {"mean": 0.6604376826115352, "se": 0.007017957355969411, "per_batch": [0.625201307410232, 0.6701150926310071, 0.6715778910679207, 0.6723486313177623, 0.6862873952135659, 0.6580370090348883, 0.6384860212848043, 0.6614481129321013]}, "s6_tryfail": {"mean": 0.6728212611401647, "se": 0.004824708124390516, "per_batch": [0.6802316281027101, 0.6945571809803041, 0.6730548350778678, 0.6813304682547353, 0.6768184174180241, 0.6671325694997442, 0.6543597189898025, 0.65508527079813]}, "s7_mixed": {"mean": 0.7829759286650222, "se": 0.08891676616480938, "per_batch": [1.0719286507834895, 1.0487072092986913, 0.9167929044512491, 0.8095931944633957, 0.8840941010166795, 0.6359224908867059, 0.36390978578087546, 0.5328590926390914]}, "s8_register": {"mean": 0.4906504977761471, "se": 0.009077126336788649, "per_batch": [0.4430003073464096, 0.5108862451995572, 0.478870513858552, 0.5101862758105455, 0.4840021751906592, 0.5198495510058996, 0.5052149023090424, 0.4731940114885114]}, "web": {"mean": 0.9505967609111062, "se": 0.038152120989285714, "per_batch": [1.1091405369506755, 0.8791101297860008, 0.8925084382014679, 0.8542297436509799, 1.1341640986425394, 0.9075658632498085, 0.9278510449364259, 0.9002042318709514]}}, "battery_acc": {"s1_perspective": {"in": 0.55, "transfer": 0.4625}, "s2_concept": {"in": 0.2976190476190476, "transfer": 0.6779661016949152}, "s3_rules": {"in": 0.155, "transfer": 0.07, "held-out": 0.05}, "s4_arith": {"in": 0.34375, "transfer": 0.24166666666666667, "held-out": 0.35}}, "probes": {"p0_grounding": {"acc": 0.36666666666666664, "n": 30, "families": {"count": 0.125, "permanence": 0.7142857142857143, "sequence": 0.25, "attribute": 0.42857142857142855}}, "p1_self_other": {"acc": 0.5, "n": 30, "families": {"belief": 0.375, "reality": 0.6666666666666666, "perspective": 0.6, "reference": 0.7142857142857143, "second_order": 0.0}}, "p2_difference": {"acc": 0.5666666666666667, "n": 30, "families": {"category": 0.8333333333333334, "antonym": 0.6666666666666666, "odd_one_out": 0.6666666666666666, "property": 0.5, "analogy": 0.16666666666666666}}, "p3_logic": {"acc": 0.16666666666666666, "n": 30, "families": {"d1": 0.5, "d2": 0.0, "d3": 0.5, "d4": 0.0, "d5": 0.0, "d6": 0.0, "negation": 0.0, "contrapositive": 0.25}}, "p4_arithmetic": {"acc": 0.3, "n": 30, "families": {"add2": 0.0, "add3": 0.0, "sub2": 0.5, "sub3": 0.6, "mul1": 0.25, "mul2": 0.3333333333333333, "word": 0.4}}, "p5_causal": {"acc": 0.5666666666666667, "n": 30, "families": {"effect": 0.7142857142857143, "cause": 0.42857142857142855, "counterfactual": 0.5, "prevention": 0.4, "chain": 0.8}}, "p6_learning": {"acc": 0.36666666666666664, "n": 30, "families": {"error_detect": 0.25, "correction": 0.25, "strategy": 0.7142857142857143, "check": 0.2857142857142857}}, "p7_composite": {"acc": 0.3333333333333333, "n": 30, "families": {"logic_arith": 0.25, "causal_category": 0.375, "belief_causal": 0.2857142857142857, "three_step": 0.42857142857142855}}, "p8_articulation": {"acc": 0.5333333333333333, "n": 30, "families": {"definition": 0.5, "register": 0.5714285714285714, "grammar": 0.875, "paraphrase": 0.14285714285714285}}}, "probes_mean": 0.41111111111111115}, "closes": [{"close": 4, "attached": ["s1_perspective", "s2_concept", "s3_rules", "s4_arith"], "rows": {"s1_perspective": {"off": 0.5538333013185592, "all_on": 0.04815859600255551, "all_on_minus_off": {"mean": -0.5056747053160037, "se": 0.004772479263522749, "per_batch": [-0.5063001880783965, -0.5250790758408056, -0.5167261418368331, -0.5114098033844463, -0.49384486053761023, -0.50753373364171, -0.4813020959422336, -0.5032017432659943]}}, "s2_concept": {"off": 0.7211839326967642, "all_on": 0.1122730718885469, "all_on_minus_off": {"mean": -0.6089108608082172, "se": 0.003752184076790614, "per_batch": [-0.6082019832553367, -0.6086282207576689, -0.595556442902696, -0.6314853277455297, -0.607369307753353, -0.6071645624075008, -0.6000273683880197, -0.6128536732556329]}}, "s3_rules": {"off": 0.9296131361815035, "all_on": 0.22363961516270922, "all_on_minus_off": {"mean": -0.7059735210187943, "se": 0.005221307822158275, "per_batch": [-0.6867047971072099, -0.7063797628507347, -0.7201309151722782, -0.6889754840465516, -0.6934461730557302, -0.7262080941231664, -0.7120921880969743, -0.7138507536977091]}}, "s4_arith": {"off": 1.2338748670942346, "all_on": 0.33837794280053096, "all_on_minus_off": {"mean": -0.8954969242937036, "se": 0.005946450836102067, "per_batch": [-0.910657939329671, -0.8882506852118887, -0.9108384351217513, -0.8697349685265048, -0.8740700922111406, -0.9133315816199843, -0.8973221466077924, -0.8997695457208957]}}, "s5_causal": {"off": 0.6604376826115352, "all_on": 0.4785927403795772, "all_on_minus_off": {"mean": -0.18184494223195802, "se": 0.0033497617991287668, "per_batch": [-0.17095922510318684, -0.19072291239672545, -0.19174147964646038, -0.18731713996168925, -0.18458098197778738, -0.16505252395496628, -0.1795989885475936, -0.18478628626725485]}}, "s6_tryfail": {"off": 0.6728212611401647, "all_on": 0.4878734132957978, "all_on_minus_off": {"mean": -0.184947847844367, "se": 0.003676818534227089, "per_batch": [-0.1884349412009721, -0.20595171815665092, -0.18054029259142845, -0.1902336217550208, -0.18704897601333142, -0.17570147471807784, -0.17538352179229527, -0.17628823652715908]}}, "s7_mixed": {"off": 0.7829759286650222, "all_on": 0.7748060486950619, "all_on_minus_off": {"mean": -0.008169879969960388, "se": 0.008071949593130186, "per_batch": [0.0004919563727929788, 0.0011674182340568606, 0.00042866675727548653, 0.0013318336482815996, 0.0007409872512423865, -0.06092782076887737, 0.010358472065804258, -0.0189505533202593]}}, "s8_register": {"off": 0.4906504977761471, "all_on": 0.38211507915591003, "all_on_minus_off": {"mean": -0.10853541862023713, "se": 0.0019516197366443405, "per_batch": [-0.09871494590753666, -0.11137953333876027, -0.10322927551437178, -0.1126224089608831, -0.106098161109077, -0.11508425461665872, -0.1124128481008067, -0.10874192141380279]}}, "web": {"off": 0.9505967609111062, "all_on": 0.9523115031826181, "all_on_minus_off": {"mean": 0.001714742271511971, "se": 0.00034702970440909263, "per_batch": [0.0009441847532367653, 0.0009109921016202671, 0.0004626333308209718, 0.001167074268755064, 0.0028966177971569973, 0.002886212846779701, 0.0026413095519510765, 0.0018089135217749241]}}}, "member_in_group": {"s1_perspective": {"s1_perspective": {"masked": 0.5249068459648821, "all_on_minus_masked": {"mean": -0.4767482499623266, "se": 0.0046404523438755485, "per_batch": [-0.48082775166778047, -0.4936192354230517, -0.4881338111525776, -0.48067822350177103, -0.4611429184196625, -0.4802470522471423, -0.45437442833134195, -0.4749625789552856]}}, "s2_concept": {"masked": 0.35281911638962526, "all_on_minus_masked": {"mean": -0.24054604450107833, "se": 0.0014807298058492127, "per_batch": [-0.23822385413226108, -0.2408589293131183, -0.23448836945768958, -0.2473656026877894, -0.2378669686338894, -0.241816141752158, -0.2385748277270094, -0.24517366230471155]}}, "s3_rules": {"masked": 0.4084850715235476, "all_on_minus_masked": {"mean": -0.18484545636083843, "se": 0.002956439251782064, "per_batch": [-0.18119343970374266, -0.1794106030514047, -0.18685682839665493, -0.17461738207836416, -0.1871887549128202, -0.2029063353664977, -0.18414444701423488, -0.18244586036298835]}}, "s4_arith": {"masked": 0.4958662425011527, "all_on_minus_masked": {"mean": -0.15748829970062178, "se": 0.006339441968984009, "per_batch": [-0.161703935751342, -0.18804284225483625, -0.12821710781051915, -0.1616860065599896, -0.1697660524671633, -0.15242112917535916, -0.15629533935568296, -0.14177398423008175]}}, "web": {"masked": 0.951965818054384, "all_on_minus_masked": {"mean": 0.00034568512823422715, "se": 0.00043580641666973355, "per_batch": [0.0007402133293135105, -0.0011999229550698765, -0.0012829905754365711, -0.0003994297066288732, 0.001545608083304062, 0.001611305455933687, 0.0016756269673563473, 7.507042710153122e-05]}}}, "s2_concept": {"s1_perspective": {"masked": 0.052446636749289, "all_on_minus_masked": {"mean": -0.004288040746733487, "se": 0.0003889620581606781, "per_batch": [-0.004534354836744983, -0.004600208068651919, -0.005045755998001229, -0.002841895067435443, -0.005946902092859627, -0.004756002852502272, -0.002704722779998836, -0.003874484277673583]}}, "s2_concept": {"masked": 0.2405273978196043, "all_on_minus_masked": {"mean": -0.1282543259310574, "se": 0.0014059423490730538, "per_batch": [-0.13360332496373462, -0.129027210589577, -0.12432546288896414, -0.13343453474082848, -0.12880887861430668, -0.12865435220250646, -0.12266383100982174, -0.12551701243872004]}}, "s3_rules": {"masked": 0.33001281952617856, "all_on_minus_masked": {"mean": -0.10637320436346935, "se": 0.0010514704692854385, "per_batch": [-0.10446492786515835, -0.10497683422545148, -0.1058582453111234, -0.10532797980285061, -0.10214397899610203, -0.10692161404140813, -0.11100856676084919, -0.1102834879048116]}}, "s4_arith": {"masked": 0.4588231411824152, "all_on_minus_masked": {"mean": -0.1204451983818842, "se": 0.0038520304482725204, "per_batch": [-0.13326828126202256, -0.12483405858322816, -0.10634659004824837, -0.11504800927298942, -0.10546328715341696, -0.13021889987523677, -0.11822074521610648, -0.1301617156438249]}}, "web": {"masked": 0.9524113176136126, "all_on_minus_masked": {"mean": -9.98144309945076e-05, "se": 0.00030423668601581294, "per_batch": [-0.0008894942702624142, 0.001381708617031796, -0.0001835914797960747, 0.0006292845195070562, -0.00018410742774865874, -0.0014308096638531165, -0.00020999081670347586, 8.848507386882698e-05]}}}, "s3_rules": {"s1_perspective": {"masked": 0.04808155885978826, "all_on_minus_masked": {"mean": 7.703714276724623e-05, "se": 5.0944941012207696e-05, "per_batch": [0.00021867594057205925, 3.234886173572593e-05, 6.291877792659328e-05, -8.1498278677615e-05, -1.1237991342311071e-05, 0.00029280046309393404, -9.422230761253758e-05, 0.00019651167644212097]}}, "s2_concept": {"masked": 0.11159634165523312, "all_on_minus_masked": {"mean": 0.000676730233313794, "se": 6.847269411209852e-05, "per_batch": [0.0006824916521177005, 0.0008053195115807904, 0.0004454135679033133, 0.0005459481762482982, 0.0009728521113525207, 0.00043202041896737986, 0.0006866084868227262, 0.000843187941517623]}}, "s3_rules": {"masked": 0.3188603089357986, "all_on_minus_masked": {"mean": -0.09522069377308937, "se": 0.0009429880452452454, "per_batch": [-0.09182523492264708, -0.09329274138462809, -0.0963564188292668, -0.0972481273761591, -0.09265679253740014, -0.09730316182443519, -0.09925729319703247, -0.09382578011314599]}}, "s4_arith": {"masked": 0.33836632590991095, "all_on_minus_masked": {"mean": 1.1616890620000375e-05, "se": 0.00012757624626812272, "per_batch": [-0.00033128158122447404, -7.227570902507141e-05, -0.0002141184003242036, -0.0003463730588377234, 0.00033177603134570965, -6.793314709097054e-06, 0.0007221551509729585, 9.846006761904391e-06]}}, "web": {"masked": 0.9519085370827309, "all_on_minus_masked": {"mean": 0.0004029660998872281, "se": 6.482262636932689e-05, "per_batch": [0.00035101659041103517, 0.0005267828595927337, 0.00013672620743598873, 0.00024137765048592996, 0.0003450831889562078, 0.0007316141967703693, 0.0003940982444521346, 0.0004970298609934254]}}}, "s4_arith": {"s1_perspective": {"masked": 0.04824758157281072, "all_on_minus_masked": {"mean": -8.898557025521434e-05, "se": 5.347876859202058e-05, "per_batch": [-0.00021872968515045804, 1.0485567244790384e-05, -0.00016130360313579284, 0.00019391043884781772, -0.00020162816030528524, 3.384833547293714e-05, -0.00015069442336069355, -0.00021777303165503026]}}, "s2_concept": {"masked": 0.1129011726837148, "all_on_minus_masked": {"mean": -0.0006281007951678755, "se": 4.737390225833419e-05, "per_batch": [-0.0006924236502050263, -0.000494020164603412, -0.000547882981070516, -0.0007538429543944153, -0.00048002507638945924, -0.0006830290979016329, -0.0008440908504346589, -0.0005294915863438832]}}, "s3_rules": {"masked": 0.22361192864314922, "all_on_minus_masked": {"mean": 2.768651955999471e-05, "se": 8.589008508571182e-05, "per_batch": [0.00034469622799179755, 0.000345104686787584, -0.00020048877524331266, -4.217874512410513e-05, -3.91905465653708e-05, -0.00034095560533550784, 0.0001466797033546352, 7.825210614237355e-06]}}, "s4_arith": {"masked": 0.4860708147619841, "all_on_minus_masked": {"mean": -0.14769287196145317, "se": 0.0022419971765959474, "per_batch": [-0.155302784481891, -0.13937248298692756, -0.1500984604797943, -0.14428733868979082, -0.1390868842973385, -0.15540167450613712, -0.14880399006241263, -0.14918936018733342]}}, "web": {"masked": 0.952207862137642, "all_on_minus_masked": {"mean": 0.00010364104497621873, "se": 9.211283472483794e-05, "per_batch": [0.00021601020948369687, 0.00016888696314731888, -0.00018264557521630032, 0.00013208267586262146, -0.00021704210538886493, -0.00014463740937553649, 0.0004161980150880762, 0.0004402755862087382]}}}}, "member_alone": {"s1_perspective": {"s1_perspective": {"alone": 0.05251879489204885, "alone_minus_off": {"mean": -0.5013145064265103, "se": 0.004723722359706749, "per_batch": [-0.5020161539899823, -0.5206146212027206, -0.511447510580648, -0.5087860038161363, -0.4873552833133265, -0.5025627202288822, -0.47857692335015567, -0.4991568349302314]}}, "s2_concept": {"alone": 0.23945795595196098, "alone_minus_off": {"mean": -0.48172597674480316, "se": 0.0029625557017275653, "per_batch": [-0.4753027767620657, -0.4804484116845556, -0.4714891474705084, -0.49853255940543073, -0.48017119714919604, -0.4800707377832608, -0.47906046869643004, -0.4887325150069781]}}, "s3_rules": {"alone": 0.508758760689348, "alone_minus_off": {"mean": -0.42085437549215554, "se": 0.004040788518956882, "per_batch": [-0.4068824458170247, -0.41901358606093464, -0.4391522815246798, -0.40577522151077006, -0.4185499638298712, -0.4321406348403284, -0.4199813324373397, -0.42533953791629586]}}, "s4_arith": {"alone": 0.7225499640202783, "alone_minus_off": {"mean": -0.5113249030739563, "se": 0.003602454892240459, "per_batch": [-0.5097063582233129, -0.5105177293744209, -0.5238398924712184, -0.49785761349211977, -0.5006369821170508, -0.5244675431555423, -0.5047830108770632, -0.5187900948809221]}}, "web": {"alone": 0.9520279682848395, "alone_minus_off": {"mean": 0.0014312073737332287, "se": 0.0004766934609748917, "per_batch": [0.0014245322970964924, -0.0004134462926742577, 0.00041473616255571777, 1.4618525323362164e-05, 0.0028610173884282553, 0.0033452345419324736, 0.002158726233630004, 0.0016442401335737822]}}}, "s2_concept": {"s1_perspective": {"alone": 0.5265382088974727, "alone_minus_off": {"mean": -0.02729509242108652, "se": 0.0008478035966868174, "per_batch": [-0.023477093695305706, -0.029719504978010347, -0.026994955823041678, -0.02971271166330125, -0.0301825252697957, -0.02641580424626122, -0.025215322347576063, -0.02664282134540019]}}, "s2_concept": {"alone": 0.35026574371684355, "alone_minus_off": {"mean": -0.3709181889799206, "se": 0.0024118108231179045, "per_batch": [-0.37237346048863007, -0.3698774117831641, -0.3638556972371791, -0.385719013225945, -0.3722404964016817, -0.36824759669443125, -0.36423210276642376, -0.37079973324191]}}, "s3_rules": {"alone": 0.6188593181726778, "alone_minus_off": {"mean": -0.31075381800882573, "se": 0.002804570723181844, "per_batch": [-0.3059784617336435, -0.31155671901437876, -0.3166699782028437, -0.29686763780852365, -0.3055774841831569, -0.3180009949335336, -0.3101597899327456, -0.3212194782617802]}}, "s4_arith": {"alone": 0.8801855050550988, "alone_minus_off": {"mean": -0.3536893620391357, "se": 0.004028817126679558, "per_batch": [-0.3418542146862763, -0.3484509532254211, -0.36770071336251886, -0.35094263787112157, -0.33855532946875133, -0.36934916707103904, -0.36129874517547633, -0.3513631354524811]}}, "web": {"alone": 0.9516445652114878, "alone_minus_off": {"mean": 0.0010478043003816634, "se": 0.000252314831290859, "per_batch": [9.347257074376891e-05, 0.0012871181590572434, 0.0019100393204823929, 0.0013596088463959655, 0.0010667223919766933, -7.928400204770814e-05, 0.0010015409672995412, 0.0017432161491454101]}}}, "s3_rules": {"s1_perspective": {"alone": 0.55350150916384, "alone_minus_off": {"mean": -0.00033179215471922097, "se": 0.00018341479131836675, "per_batch": [0.0003735463176740561, -0.00032229548771689487, -0.0008996842423260043, -0.0003734173306858546, -0.0009771194308770825, -0.0004045461904920167, 0.00045029357562154093, -0.0005011144489515118]}}, "s2_concept": {"alone": 0.7242528502377963, "alone_minus_off": {"mean": 0.0030689175410321956, "se": 0.00019465440878027398, "per_batch": [0.002732804322210014, 0.003277301394841614, 0.0030090514551583958, 0.0019815841032413406, 0.0035649423784097634, 0.003502254702170249, 0.0028635111368657817, 0.0036198908353604065]}}, "s3_rules": {"alone": 0.8236434891082838, "alone_minus_off": {"mean": -0.10596964707321978, "se": 0.009346493007512462, "per_batch": [-0.11229495399364031, -0.10826746427573508, -0.09548760547277069, -0.1291266380596071, -0.08108225264517621, -0.11603213699688908, -0.14432027336743114, -0.0611458517745086]}}, "s4_arith": {"alone": 1.2339533341786908, "alone_minus_off": {"mean": 7.84670844561075e-05, "se": 0.00040067832550554967, "per_batch": [0.0005466468557673299, -0.00020706711163875902, -8.02299066273715e-05, 0.0007894003674602246, -7.782214951546074e-05, 0.0008427149892276109, -0.002416270253297048, 0.0012303638842723341]}}, "web": {"alone": 0.9511259085321484, "alone_minus_off": {"mean": 0.0005291476210421558, "se": 0.0001432382824627441, "per_batch": [-0.00010852105269454349, 0.0007161357581927374, 0.0011528857000588921, 0.0004110385355621249, 0.0007747818421370845, 0.0008065986325466179, 0.0002391418760246955, 0.00024111967650963795]}}}, "s4_arith": {"s1_perspective": {"alone": 0.5529388195514218, "alone_minus_off": {"mean": -0.0008944817671374533, "se": 0.00022219662593559158, "per_batch": [-0.0008394043231986226, -0.0012983400270260015, 0.0003624964323560853, -0.0017084756536711598, -0.0014122355375599804, -0.0008876454567656733, -0.0007099873784243149, -0.0006622621928099592]}}, "s2_concept": {"alone": 0.7207903557747224, "alone_minus_off": {"mean": -0.0003935769220416857, "se": 0.00036656930724860673, "per_batch": [-0.000407598882544824, -0.00045631296840176194, 0.001248465058275916, -0.00226862314753129, 0.0005794525464195566, -0.0009882123118576391, -0.0004823253443447806, -0.00037346032634866244]}}, "s3_rules": {"alone": 0.9280330993212799, "alone_minus_off": {"mean": -0.001580036860223663, "se": 0.0002472548294138001, "per_batch": [-0.0006835450458542169, -0.0017735710870226962, -0.0018863057146633055, -0.0008067706151975162, -0.0008262906460703157, -0.0023852273848162797, -0.0022952804584150943, -0.00198330392974988]}}, "s4_arith": {"alone": 1.126693075548818, "alone_minus_off": {"mean": -0.10718179154541654, "se": 0.019733005800031626, "per_batch": [-0.12218541827078422, -0.000754745863311479, -0.15389351564505827, -0.15261577054047337, -0.042793669092263675, -0.11302235461546473, -0.12726140021967436, -0.14492745811630225]}}, "web": {"alone": 0.9507103339541697, "alone_minus_off": {"mean": 0.00011357304306353067, "se": 0.00010289644243726181, "per_batch": [-0.0003639152892258579, 4.480148054974986e-05, -4.5575402478625904e-05, 0.0002953802028569097, 5.933401454827347e-06, 0.0003312385855618327, 3.989997500020159e-05, 0.0006008213907892079]}}}}, "battery": {"s1_perspective": {"acc_off": {"in": 0.55, "transfer": 0.4625}, "acc_all_on": {"in": 0.9625, "transfer": 0.475}, "acc_member_masked": {"in": 0.55, "transfer": 0.4625}, "acc_member_alone": {"in": 0.8375, "transfer": 0.4625}, "margin_all_on_minus_off": {"in": {"mean": 2.3936356898397206, "se": 0.16972763326741638, "per_batch": null}, "transfer": {"mean": -0.12855345383286476, "se": 0.11015434333558147, "per_batch": null}}}, "s2_concept": {"acc_off": {"in": 0.2976190476190476, "transfer": 0.6779661016949152}, "acc_all_on": {"in": 0.9761904761904762, "transfer": 0.7796610169491526}, "acc_member_masked": {"in": 0.6904761904761905, "transfer": 0.711864406779661}, "acc_member_alone": {"in": 0.9642857142857143, "transfer": 0.7203389830508474}, "margin_all_on_minus_off": {"in": {"mean": 4.172865661748109, "se": 0.322182336483479, "per_batch": null}, "transfer": {"mean": 0.26194089600595377, "se": 0.19708812594069705, "per_batch": null}}}, "s3_rules": {"acc_off": {"in": 0.155, "transfer": 0.07, "held-out": 0.05}, "acc_all_on": {"in": 0.905, "transfer": 0.3, "held-out": 0.4583333333333333}, "acc_member_masked": {"in": 0.805, "transfer": 0.22, "held-out": 0.43333333333333335}, "acc_member_alone": {"in": 0.32, "transfer": 0.08, "held-out": 0.10833333333333334}, "margin_all_on_minus_off": {"in": {"mean": 2.839122497625649, "se": 0.12183926783746772, "per_batch": null}, "transfer": {"mean": 1.0348015177249907, "se": 0.14924431340601638, "per_batch": null}, "held-out": {"mean": 1.5638080275307098, "se": 0.18997366321172568, "per_batch": null}}}, "s4_arith": {"acc_off": {"in": 0.34375, "transfer": 0.24166666666666667, "held-out": 0.35}, "acc_all_on": {"in": 0.6125, "transfer": 0.2375, "held-out": 0.35}, "acc_member_masked": {"in": 0.46875, "transfer": 0.25, "held-out": 0.3}, "acc_member_alone": {"in": 0.38125, "transfer": 0.24583333333333332, "held-out": 0.3875}, "margin_all_on_minus_off": {"in": {"mean": 1.4841932335766614, "se": 0.21805308784719607, "per_batch": null}, "transfer": {"mean": -0.2683388772110144, "se": 0.06709485076421261, "per_batch": null}, "held-out": {"mean": -0.11913171410560608, "se": 0.10964168667654946, "per_batch": null}}}}, "probes_all_on": {"p0_grounding": {"acc": 0.36666666666666664, "n": 30, "families": {"count": 0.125, "permanence": 0.5714285714285714, "sequence": 0.25, "attribute": 0.5714285714285714}}, "p1_self_other": {"acc": 0.4666666666666667, "n": 30, "families": {"belief": 0.125, "reality": 0.5, "perspective": 0.8, "reference": 0.8571428571428571, "second_order": 0.0}}, "p2_difference": {"acc": 0.5333333333333333, "n": 30, "families": {"category": 1.0, "antonym": 0.5, "odd_one_out": 0.3333333333333333, "property": 0.5, "analogy": 0.3333333333333333}}, "p3_logic": {"acc": 0.26666666666666666, "n": 30, "families": {"d1": 0.5, "d2": 0.25, "d3": 0.5, "d4": 0.0, "d5": 0.0, "d6": 0.0, "negation": 0.5, "contrapositive": 0.25}}, "p4_arithmetic": {"acc": 0.36666666666666664, "n": 30, "families": {"add2": 0.4, "add3": 0.0, "sub2": 0.75, "sub3": 0.4, "mul1": 0.25, "mul2": 0.3333333333333333, "word": 0.4}}, "p5_causal": {"acc": 0.5666666666666667, "n": 30, "families": {"effect": 0.7142857142857143, "cause": 0.42857142857142855, "counterfactual": 0.5, "prevention": 0.4, "chain": 0.8}}, "p6_learning": {"acc": 0.36666666666666664, "n": 30, "families": {"error_detect": 0.25, "correction": 0.25, "strategy": 0.7142857142857143, "check": 0.2857142857142857}}, "p7_composite": {"acc": 0.36666666666666664, "n": 30, "families": {"logic_arith": 0.375, "causal_category": 0.375, "belief_causal": 0.2857142857142857, "three_step": 0.42857142857142855}}, "p8_articulation": {"acc": 0.4666666666666667, "n": 30, "families": {"definition": 0.375, "register": 0.42857142857142855, "grammar": 0.875, "paraphrase": 0.14285714285714285}}}, "probes_mean_off": 0.41111111111111115, "probes_mean_all_on": 0.4185185185185185, "gates": {"s1_perspective": {"mean": 0.2099039478571495, "max": 0.5712027580899203}, "s2_concept": {"mean": 0.27674460361666475, "max": 0.4977012170865557}, "s3_rules": {"mean": 0.12033936123137423, "max": 0.1738736117222015}, "s4_arith": {"mean": 0.10767731599767934, "max": 0.15932411779931022}}, "chain_proof": "compiled chain: 8 cache entries (limit 256), 0 limit warnings", "resumed_read": true}, {"close": 5, "attached": ["s1_perspective", "s2_concept", "s3_rules", "s4_arith", "s5_causal"], "rows": {"s1_perspective": {"off": 0.5538333013185592, "all_on": 0.04885903095397748, "all_on_minus_off": {"mean": -0.5049742703645816, "se": 0.004796770503449524, "per_batch": [-0.5055043383615289, -0.5247208413535162, -0.5155674517223077, -0.511398194555513, -0.49295205560131866, -0.5064249614915974, -0.4806327824607383, -0.5025935373701337]}}, "s2_concept": {"off": 0.7211839326967642, "all_on": 0.1116441286471127, "all_on_minus_off": {"mean": -0.6095398040496514, "se": 0.003755981907762475, "per_batch": [-0.6093713792898783, -0.6091496936529163, -0.5955959344189005, -0.6315846907220657, -0.6082681105845934, -0.6070219135475258, -0.6007921752364067, -0.6145345349449248]}}, "s3_rules": {"off": 0.9296131361815035, "all_on": 0.2241419791731084, "all_on_minus_off": {"mean": -0.7054711570083951, "se": 0.005260435633683978, "per_batch": [-0.6862704334246245, -0.704988874659713, -0.7205644404394407, -0.6887581194736924, -0.692788038446541, -0.7258577654633589, -0.7107437581228844, -0.7137978260369061]}}, "s4_arith": {"off": 1.2338748670942346, "all_on": 0.3385110090021782, "all_on_minus_off": {"mean": -0.8953638580920564, "se": 0.00594196505841796, "per_batch": [-0.9100197977036445, -0.8881827305669665, -0.9108780771227752, -0.8693016797354872, -0.8740893972637, -0.9130636756456028, -0.8974808865945387, -0.8998946201037357]}}, "s5_causal": {"off": 0.6604376826115352, "all_on": 0.04318483153912111, "all_on_minus_off": {"mean": -0.6172528510724141, "se": 0.00699243098168417, "per_batch": [-0.5823933691324045, -0.6267433495407596, -0.6286143595484117, -0.6287963306291692, -0.6429463322158949, -0.6154171821558997, -0.5947638286094479, -0.6183480567473254]}}, "s6_tryfail": {"off": 0.6728212611401647, "all_on": 0.4832413561899892, "all_on_minus_off": {"mean": -0.18957990495017557, "se": 0.003691925622744965, "per_batch": [-0.19367761733421124, -0.2087966981628862, -0.1841603554094397, -0.198054231833368, -0.1917027405543536, -0.17953208729640785, -0.17987076413161956, -0.18084474487911834]}}, "s7_mixed": {"off": 0.7829759286650222, "all_on": 0.7749160799704091, "all_on_minus_off": {"mean": -0.00805984869461307, "se": 0.007906394154199452, "per_batch": [0.00030784894504432003, 0.0006773536704232086, 0.0005281587207996985, 0.0020627599144485975, 0.0006473426978476082, -0.05921719533207048, 0.010474947316101102, -0.01996000548949861]}}, "s8_register": {"off": 0.4906504977761471, "all_on": 0.37807989052714974, "all_on_minus_off": {"mean": -0.1125706072489974, "se": 0.0020992661495918075, "per_batch": [-0.1019466074064056, -0.11548867182016265, -0.10684822044515241, -0.11700383897426397, -0.11052082396300222, -0.11987173566872672, -0.11650255254266156, -0.11238240717160403]}}, "web": {"off": 0.9505967609111062, "all_on": 0.9527085788767133, "all_on_minus_off": {"mean": 0.0021118179656071517, "se": 0.00041727264938435114, "per_batch": [0.001138009200759127, 0.0009627588795300124, 0.0005842250649810143, 0.0021441936996321465, 0.003989567543390082, 0.0033302720513074258, 0.002356334299471352, 0.0023891829857860536]}}}, "member_in_group": {"s1_perspective": {"s1_perspective": {"masked": 0.5074123837581528, "all_on_minus_masked": {"mean": -0.4585533528041753, "se": 0.004268047171637093, "per_batch": [-0.46227722999342813, -0.4771562875349926, -0.4651781258590251, -0.46404492142357445, -0.44770513998582234, -0.45943147606526397, -0.43714602328549745, -0.45548761828579853]}}, "s2_concept": {"masked": 0.34467560888428317, "all_on_minus_masked": {"mean": -0.2330314802371704, "se": 0.001496618473584831, "per_batch": [-0.2307930212342631, -0.23336145313791096, -0.22593599469775244, -0.23942013320491576, -0.2318011620357895, -0.23382102302776347, -0.23121697996647125, -0.23790207459249677]}}, "s3_rules": {"masked": 0.40602267229861455, "all_on_minus_masked": {"mean": -0.18188069312550612, "se": 0.003157668615993281, "per_batch": [-0.18031677963880086, -0.17529912130638434, -0.18433745454416567, -0.170739065812251, -0.18309982439289418, -0.20117898461689435, -0.18096633661327832, -0.17910797807938023]}}, "s4_arith": {"masked": 0.49240091032433964, "all_on_minus_masked": {"mean": -0.15388990132216138, "se": 0.006222562782504418, "per_batch": [-0.15933074715208612, -0.18311031533459682, -0.12280773749293117, -0.15702375037558103, -0.16443536421234517, -0.1511548209170564, -0.1531408335735573, -0.14011564151913708]}}, "s5_causal": {"masked": 0.05903609491348296, "all_on_minus_masked": {"mean": -0.015851263374361848, "se": 0.0008952882360751147, "per_batch": [-0.015819388808530455, -0.013116571273835712, -0.018361698159849643, -0.015768928023875743, -0.01597884359816948, -0.0204531442430163, -0.014018864815442837, -0.013292668072174628]}}, "web": {"masked": 0.9521777329269938, "all_on_minus_masked": {"mean": 0.000530845949719444, "se": 0.0005123261616510245, "per_batch": [0.00014274560021609872, -0.0010074743687542576, -0.0016562789191342242, 9.312860544230528e-05, 0.002056052591064894, 0.002469670866389939, 0.0015893776679486438, 0.0005595455545821526]}}}, "s2_concept": {"s1_perspective": {"masked": 0.05456580764502775, "all_on_minus_masked": {"mean": -0.005706776691050273, "se": 0.0004876333503072643, "per_batch": [-0.005184234278692829, -0.006210266650366822, -0.006966592604283062, -0.0033933628118848827, -0.007413064940380749, -0.006060206413031184, -0.004134591913730071, -0.006291893916032583]}}, "s2_concept": {"masked": 0.23556453745931896, "all_on_minus_masked": {"mean": -0.12392040881220626, "se": 0.0013149292452526342, "per_batch": [-0.1288169188032345, -0.1249334430575955, -0.11956102751211176, -0.1285368128095574, -0.12496809756174437, -0.12415822050991572, -0.11905686037110741, -0.12133188987238336]}}, "s3_rules": {"masked": 0.33123365180947284, "all_on_minus_masked": {"mean": -0.10709167263636443, "se": 0.0010726425713160075, "per_batch": [-0.10496191473048913, -0.10507821799813513, -0.10879938511702236, -0.10854665661157947, -0.1017688633367389, -0.10676347599393979, -0.11084320544204462, -0.10997166186096596]}}, "s4_arith": {"masked": 0.45778693496205103, "all_on_minus_masked": {"mean": -0.11927592595987282, "se": 0.0038334816670067408, "per_batch": [-0.1311973951673207, -0.12242593600801577, -0.1051499347635449, -0.11493839183089571, -0.10362732935979374, -0.13010930393097447, -0.11805835059802927, -0.12870076602040803]}}, "s5_causal": {"masked": 0.04523111470944831, "all_on_minus_masked": {"mean": -0.0020462831703271945, "se": 0.0004705857954826123, "per_batch": [-0.002424116961742648, -0.0016683150174657546, -0.004344682157903582, -0.0024048065347254607, -0.0030922211902241065, -0.0001255742074190505, -0.0015636608871868185, -0.0007468884059501357]}}, "web": {"masked": 0.9529648867710773, "all_on_minus_masked": {"mean": -0.00025630789436396906, "se": 0.0003637713034777748, "per_batch": [-0.0012993289272684727, 0.0011768772798542715, 9.373054472028297e-05, 0.0010347336189161505, -0.00021833197527043602, -0.0013287379605660954, -0.001399250847419764, -0.00011015488787768923]}}}, "s3_rules": {"s1_perspective": {"masked": 0.04892689826395975, "all_on_minus_masked": {"mean": -6.786730998227606e-05, "se": 0.00013440126420771972, "per_batch": [0.00029793307033061656, -0.00035129606221864784, -0.0002405607328943571, 8.763590953028005e-05, -0.0006902953649005963, -0.00010823083197109701, -6.5546887810089e-05, 0.0005274224200756822]}}, "s2_concept": {"masked": 0.11080526848777031, "all_on_minus_masked": {"mean": 0.0008388601593424094, "se": 5.022706159360183e-05, "per_batch": [0.0007420836406416692, 0.0008775199781961285, 0.0009969404313889818, 0.0006345514881896941, 0.0009353276467173877, 0.0007116749581859844, 0.0007858102296235264, 0.0010269729017959028]}}, "s3_rules": {"masked": 0.3161543554049545, "all_on_minus_masked": {"mean": -0.0920123762318461, "se": 0.0010444746580264203, "per_batch": [-0.08824659842566243, -0.08900439698102666, -0.09440280340462204, -0.09465127533945611, -0.089279741205058, -0.09366007482903968, -0.09577227975561126, -0.09108183991429258]}}, "s4_arith": {"masked": 0.33878855944115266, "all_on_minus_masked": {"mean": -0.0002775504389744485, "se": 8.900308431882457e-05, "per_batch": [-6.320362419209857e-06, -0.00010462994521864832, -0.00020977583838988068, -0.0005862028654659013, -0.0005586856413278274, -5.856009261878681e-05, -0.00012081781223111099, -0.0005754109541242225]}}, "s5_causal": {"masked": 0.04308878829801184, "all_on_minus_masked": {"mean": 9.604324110927369e-05, "se": 0.00013766005603753692, "per_batch": [0.0006387274419309488, -2.6020437629760418e-05, -0.00013437756936008488, -0.00014436599925472804, 0.0001460938874501133, 0.00037927548973088615, -0.0005544774408395362, 0.0004634905568463507]}}, "web": {"masked": 0.9522174394215118, "all_on_minus_masked": {"mean": 0.0004911394552014614, "se": 0.00014879655293862913, "per_batch": [0.00040329931627325344, 0.0003462870675122742, 3.8868079095033536e-05, 0.0003526504255941809, 0.0012245164741431225, 0.000944012770585978, 1.0834907004375616e-05, 0.000608646601403473]}}}, "s4_arith": {"s1_perspective": {"masked": 0.04878236867964454, "all_on_minus_masked": {"mean": 7.666227433294256e-05, "se": 6.0210448307982555e-05, "per_batch": [0.0001882995048634109, -1.3876850141494868e-05, 0.0003004106953946109, 0.000294735267916138, -0.00016946203013609717, 6.3483095999739e-05, 1.5207028456760974e-05, -6.54985176895273e-05]}}, "s2_concept": {"masked": 0.1121614806768117, "all_on_minus_masked": {"mean": -0.000517352029698985, "se": 3.1534122363225956e-05, "per_batch": [-0.0006758380733124375, -0.0005382197058751564, -0.00047842348795330836, -0.00043696492017968047, -0.0004838731882025188, -0.00042214216545843897, -0.0006210078544342384, -0.00048234684217610124]}}, "s3_rules": {"masked": 0.22398825355552626, "all_on_minus_masked": {"mean": 0.00015372561758215247, "se": 7.909666270378593e-05, "per_batch": [0.00039631252108196513, 0.00023922786735017243, -0.00016944590676257199, 9.49129254449177e-05, 0.00019588823933272548, 1.4726014480132e-05, 0.0004972663371384523, -3.908305740857321e-05]}}, "s4_arith": {"masked": 0.4841386864197815, "all_on_minus_masked": {"mean": -0.14562767741760335, "se": 0.0022445357001523525, "per_batch": [-0.15290822703826534, -0.13661355879961773, -0.14777540482326484, -0.14193760422454632, -0.13749393798938447, -0.15308463824238788, -0.14777531883193945, -0.1474327293914207]}}, "s5_causal": {"masked": 0.043118578246008746, "all_on_minus_masked": {"mean": 6.625329311236213e-05, "se": 0.00011283360529400487, "per_batch": [0.000709355879628254, 1.710421207409135e-05, -0.0002184851473187588, -0.0002053123511542368, -2.8618987995145118e-05, 4.463487235673719e-05, -0.00014884829709284342, 0.00036019616440079866]}}, "web": {"masked": 0.9524866030190278, "all_on_minus_masked": {"mean": 0.00022197585768549133, "se": 0.00013938063682036297, "per_batch": [-7.704822758647367e-05, 0.000589556527157753, -0.00037767390129472833, 0.0004974598176207268, 0.0005825912297976465, 0.00034654504148856624, -0.00025057872230715983, 0.0004649550966075999]}}}, "s5_causal": {"s1_perspective": {"masked": 0.04815859600255551, "all_on_minus_masked": {"mean": 0.0007004349514219731, "se": 0.00013521838025848463, "per_batch": [0.0007958497168676429, 0.00035823448728938834, 0.0011586901145254555, 1.160882893323778e-05, 0.0008928049362915552, 0.0011087721501125275, 0.0006693134814953344, 0.0006082058958606432]}}, "s2_concept": {"masked": 0.1122730718885469, "all_on_minus_masked": {"mean": -0.000628943241434211, "se": 0.00022058030607753198, "per_batch": [-0.0011693960345416643, -0.000521472895247399, -3.949151620437352e-05, -9.93629765359827e-05, -0.0008988028312404001, 0.00014264885997501697, -0.0007648068483869092, -0.0016808616892919764]}}, "s3_rules": {"masked": 0.22363961516270922, "all_on_minus_masked": {"mean": 0.0005023640103991867, "se": 0.00022024388097213884, "per_batch": [0.0004343636825853703, 0.001390888191021622, -0.00043352526716242124, 0.00021736457285922994, 0.0006581346091892593, 0.0003503286598075528, 0.0013484299740898487, 5.292766080303157e-05]}}, "s4_arith": {"masked": 0.33837794280053096, "all_on_minus_masked": {"mean": 0.00013306620164726252, "se": 0.00010110890538820489, "per_batch": [0.0006381416260264894, 6.795464492215242e-05, -3.964200102385407e-05, 0.0004332887910175054, -1.930505255937076e-05, 0.00026790597438153263, -0.00015873998674637013, -0.00012507438283998473]}}, "s5_causal": {"masked": 0.4785927403795772, "all_on_minus_masked": {"mean": -0.4354079088404561, "se": 0.005634811634513396, "per_batch": [-0.41143414402921763, -0.43602043714403416, -0.43687287990195134, -0.4414791906674799, -0.45836535023810754, -0.45036465820093347, -0.4151648400618542, -0.4335617704800705]}}, "web": {"masked": 0.9523115031826181, "all_on_minus_masked": {"mean": 0.00039707569409518084, "se": 0.0001668025491916777, "per_batch": [0.0001938244475223616, 5.1766777909745265e-05, 0.00012159173416004254, 0.0009771194308770825, 0.001092949746233085, 0.00044405920452772474, -0.00028497525247972444, 0.0005802694640111294]}}}}, "member_alone": {"s1_perspective": {"s1_perspective": {"alone": 0.05251879489204885, "alone_minus_off": {"mean": -0.5013145064265103, "se": 0.004723722359706749, "per_batch": [-0.5020161539899823, -0.5206146212027206, -0.511447510580648, -0.5087860038161363, -0.4873552833133265, -0.5025627202288822, -0.47857692335015567, -0.4991568349302314]}}, "s2_concept": {"alone": 0.23945795595196098, "alone_minus_off": {"mean": -0.48172597674480316, "se": 0.0029625557017275653, "per_batch": [-0.4753027767620657, -0.4804484116845556, -0.4714891474705084, -0.49853255940543073, -0.48017119714919604, -0.4800707377832608, -0.47906046869643004, -0.4887325150069781]}}, "s3_rules": {"alone": 0.508758760689348, "alone_minus_off": {"mean": -0.42085437549215554, "se": 0.004040788518956882, "per_batch": [-0.4068824458170247, -0.41901358606093464, -0.4391522815246798, -0.40577522151077006, -0.4185499638298712, -0.4321406348403284, -0.4199813324373397, -0.42533953791629586]}}, "s4_arith": {"alone": 0.7225499640202783, "alone_minus_off": {"mean": -0.5113249030739563, "se": 0.003602454892240459, "per_batch": [-0.5097063582233129, -0.5105177293744209, -0.5238398924712184, -0.49785761349211977, -0.5006369821170508, -0.5244675431555423, -0.5047830108770632, -0.5187900948809221]}}, "s5_causal": {"alone": 0.4626770842822263, "alone_minus_off": {"mean": -0.19776059832930892, "se": 0.003074781006523783, "per_batch": [-0.18829842997184976, -0.20876677318163606, -0.20486710956898452, -0.20499300086941613, -0.19802718756151977, -0.18360145478979822, -0.19417430322990303, -0.19935652746136384]}}, "web": {"alone": 0.9520279682848395, "alone_minus_off": {"mean": 0.0014312073737332287, "se": 0.0004766934609748917, "per_batch": [0.0014245322970964924, -0.0004134462926742577, 0.00041473616255571777, 1.4618525323362164e-05, 0.0028610173884282553, 0.0033452345419324736, 0.002158726233630004, 0.0016442401335737822]}}}, "s2_concept": {"s1_perspective": {"alone": 0.5265382088974727, "alone_minus_off": {"mean": -0.02729509242108652, "se": 0.0008478035966868174, "per_batch": [-0.023477093695305706, -0.029719504978010347, -0.026994955823041678, -0.02971271166330125, -0.0301825252697957, -0.02641580424626122, -0.025215322347576063, -0.02664282134540019]}}, "s2_concept": {"alone": 0.35026574371684355, "alone_minus_off": {"mean": -0.3709181889799206, "se": 0.0024118108231179045, "per_batch": [-0.37237346048863007, -0.3698774117831641, -0.3638556972371791, -0.385719013225945, -0.3722404964016817, -0.36824759669443125, -0.36423210276642376, -0.37079973324191]}}, "s3_rules": {"alone": 0.6188593181726778, "alone_minus_off": {"mean": -0.31075381800882573, "se": 0.002804570723181844, "per_batch": [-0.3059784617336435, -0.31155671901437876, -0.3166699782028437, -0.29686763780852365, -0.3055774841831569, -0.3180009949335336, -0.3101597899327456, -0.3212194782617802]}}, "s4_arith": {"alone": 0.8801855050550988, "alone_minus_off": {"mean": -0.3536893620391357, "se": 0.004028817126679558, "per_batch": [-0.3418542146862763, -0.3484509532254211, -0.36770071336251886, -0.35094263787112157, -0.33855532946875133, -0.36934916707103904, -0.36129874517547633, -0.3513631354524811]}}, "s5_causal": {"alone": 0.6589420516116804, "alone_minus_off": {"mean": -0.001495630999854905, "se": 0.000777079044912105, "per_batch": [0.0004168429500286397, -0.0005169798484936372, -0.005893587471092143, -0.0007514781929449654, -0.0015341712370217087, -0.001300575801487236, -0.003323435740935632, 0.0009383373431074427]}}, "web": {"alone": 0.9516445652114878, "alone_minus_off": {"mean": 0.0010478043003816634, "se": 0.000252314831290859, "per_batch": [9.347257074376891e-05, 0.0012871181590572434, 0.0019100393204823929, 0.0013596088463959655, 0.0010667223919766933, -7.928400204770814e-05, 0.0010015409672995412, 0.0017432161491454101]}}}, "s3_rules": {"s1_perspective": {"alone": 0.55350150916384, "alone_minus_off": {"mean": -0.00033179215471922097, "se": 0.00018341479131836675, "per_batch": [0.0003735463176740561, -0.00032229548771689487, -0.0008996842423260043, -0.0003734173306858546, -0.0009771194308770825, -0.0004045461904920167, 0.00045029357562154093, -0.0005011144489515118]}}, "s2_concept": {"alone": 0.7242528502377963, "alone_minus_off": {"mean": 0.0030689175410321956, "se": 0.00019465440878027398, "per_batch": [0.002732804322210014, 0.003277301394841614, 0.0030090514551583958, 0.0019815841032413406, 0.0035649423784097634, 0.003502254702170249, 0.0028635111368657817, 0.0036198908353604065]}}, "s3_rules": {"alone": 0.8236434891082838, "alone_minus_off": {"mean": -0.10596964707321978, "se": 0.009346493007512462, "per_batch": [-0.11229495399364031, -0.10826746427573508, -0.09548760547277069, -0.1291266380596071, -0.08108225264517621, -0.11603213699688908, -0.14432027336743114, -0.0611458517745086]}}, "s4_arith": {"alone": 1.2339533341786908, "alone_minus_off": {"mean": 7.84670844561075e-05, "se": 0.00040067832550554967, "per_batch": [0.0005466468557673299, -0.00020706711163875902, -8.02299066273715e-05, 0.0007894003674602246, -7.782214951546074e-05, 0.0008427149892276109, -0.002416270253297048, 0.0012303638842723341]}}, "s5_causal": {"alone": 0.6675733556594478, "alone_minus_off": {"mean": 0.007135673047912558, "se": 0.0005647387818478271, "per_batch": [0.007655162768217094, 0.008555921902111074, 0.003638808926955339, 0.00856619786550028, 0.007419245566896104, 0.007850750037911025, 0.006295510926158543, 0.007103786389551003]}}, "web": {"alone": 0.9511259085321484, "alone_minus_off": {"mean": 0.0005291476210421558, "se": 0.0001432382824627441, "per_batch": [-0.00010852105269454349, 0.0007161357581927374, 0.0011528857000588921, 0.0004110385355621249, 0.0007747818421370845, 0.0008065986325466179, 0.0002391418760246955, 0.00024111967650963795]}}}, "s4_arith": {"s1_perspective": {"alone": 0.5529388195514218, "alone_minus_off": {"mean": -0.0008944817671374533, "se": 0.00022219662593559158, "per_batch": [-0.0008394043231986226, -0.0012983400270260015, 0.0003624964323560853, -0.0017084756536711598, -0.0014122355375599804, -0.0008876454567656733, -0.0007099873784243149, -0.0006622621928099592]}}, "s2_concept": {"alone": 0.7207903557747224, "alone_minus_off": {"mean": -0.0003935769220416857, "se": 0.00036656930724860673, "per_batch": [-0.000407598882544824, -0.00045631296840176194, 0.001248465058275916, -0.00226862314753129, 0.0005794525464195566, -0.0009882123118576391, -0.0004823253443447806, -0.00037346032634866244]}}, "s3_rules": {"alone": 0.9280330993212799, "alone_minus_off": {"mean": -0.001580036860223663, "se": 0.0002472548294138001, "per_batch": [-0.0006835450458542169, -0.0017735710870226962, -0.0018863057146633055, -0.0008067706151975162, -0.0008262906460703157, -0.0023852273848162797, -0.0022952804584150943, -0.00198330392974988]}}, "s4_arith": {"alone": 1.126693075548818, "alone_minus_off": {"mean": -0.10718179154541654, "se": 0.019733005800031626, "per_batch": [-0.12218541827078422, -0.000754745863311479, -0.15389351564505827, -0.15261577054047337, -0.042793669092263675, -0.11302235461546473, -0.12726140021967436, -0.14492745811630225]}}, "s5_causal": {"alone": 0.6571682010528499, "alone_minus_off": {"mean": -0.0032694815586852416, "se": 0.00036918803111562825, "per_batch": [-0.004515404498403042, -0.0033061944801866527, -0.003756875016772576, -0.0038970838728884516, -0.0014912615656313966, -0.0025041533878878974, -0.002423407533307831, -0.004261472114404086]}}, "web": {"alone": 0.9507103339541697, "alone_minus_off": {"mean": 0.00011357304306353067, "se": 0.00010289644243726181, "per_batch": [-0.0003639152892258579, 4.480148054974986e-05, -4.5575402478625904e-05, 0.0002953802028569097, 5.933401454827347e-06, 0.0003312385855618327, 3.989997500020159e-05, 0.0006008213907892079]}}}, "s5_causal": {"s1_perspective": {"alone": 0.5355407643694168, "alone_minus_off": {"mean": -0.0182925369491425, "se": 0.0010792101601506997, "per_batch": [-0.017861086222706102, -0.017265596294093633, -0.023324072131700513, -0.01829319263299889, -0.012617335197899093, -0.02079786196883937, -0.017867363589462615, -0.018313787555439776]}}, "s2_concept": {"alone": 0.7102577029047991, "alone_minus_off": {"mean": -0.010926229791965086, "se": 0.0004629922489753803, "per_batch": [-0.012461389929229183, -0.010230344990911466, -0.010227593268497648, -0.013156930764981145, -0.009059831069139102, -0.010504141371085018, -0.010754161149776897, -0.011015445792100231]}}, "s3_rules": {"alone": 0.9173749367374315, "alone_minus_off": {"mean": -0.01223819944407202, "se": 0.0003759474969709794, "per_batch": [-0.01179603204870372, -0.013579406146825623, -0.013483955775596801, -0.010735930988785447, -0.01172250946545983, -0.011976441849458785, -0.01136538749094329, -0.013245931786802667]}}, "s4_arith": {"alone": 1.2247629037840526, "alone_minus_off": {"mean": -0.009111963310181992, "se": 0.0003133156648215143, "per_batch": [-0.00816814402007826, -0.009321803642065918, -0.009664909030537183, -0.009344677334630846, -0.007534817908275926, -0.010003456878760808, -0.010028738328437425, -0.008829159338669568]}}, "s5_causal": {"alone": 0.09109720145841224, "alone_minus_off": {"mean": -0.5693404811531231, "se": 0.007627785122552896, "per_batch": [-0.5395056362790345, -0.5729805168000802, -0.5723726226227741, -0.5818498663344731, -0.6069291611541324, -0.5535619986868231, -0.5485124375726715, -0.579011609774995]}}, "web": {"alone": 0.9508772001211694, "alone_minus_off": {"mean": 0.0002804392100631825, "se": 0.00013042933220109836, "per_batch": [0.0004127583620707753, -5.262669116401497e-05, 0.00043288033222166344, 0.0006003914341621286, 0.0007997193265121272, -0.0001322546585135198, -0.00021454835695133845, 0.00039719393216763876]}}}}, "battery": {"s1_perspective": {"acc_off": {"in": 0.55, "transfer": 0.4625}, "acc_all_on": {"in": 0.9125, "transfer": 0.4375}, "acc_member_masked": {"in": 0.5625, "transfer": 0.425}, "acc_member_alone": {"in": 0.8375, "transfer": 0.4625}, "margin_all_on_minus_off": {"in": {"mean": 2.0437853633076886, "se": 0.18067712575053008, "per_batch": null}, "transfer": {"mean": -0.18364578820765018, "se": 0.20333139382759688, "per_batch": null}}}, "s2_concept": {"acc_off": {"in": 0.2976190476190476, "transfer": 0.6779661016949152}, "acc_all_on": {"in": 0.9761904761904762, "transfer": 0.7457627118644068}, "acc_member_masked": {"in": 0.7023809523809523, "transfer": 0.711864406779661}, "acc_member_alone": {"in": 0.9642857142857143, "transfer": 0.7203389830508474}, "margin_all_on_minus_off": {"in": {"mean": 4.17145732204829, "se": 0.31922697958698865, "per_batch": null}, "transfer": {"mean": 0.20284840919203678, "se": 0.2149416410288776, "per_batch": null}}}, "s3_rules": {"acc_off": {"in": 0.155, "transfer": 0.07, "held-out": 0.05}, "acc_all_on": {"in": 0.88, "transfer": 0.3, "held-out": 0.5083333333333333}, "acc_member_masked": {"in": 0.825, "transfer": 0.22, "held-out": 0.425}, "acc_member_alone": {"in": 0.32, "transfer": 0.08, "held-out": 0.10833333333333334}, "margin_all_on_minus_off": {"in": {"mean": 2.828108463473618, "se": 0.12233268908423411, "per_batch": null}, "transfer": {"mean": 1.058280459046364, "se": 0.15093378928294338, "per_batch": null}, "held-out": {"mean": 1.6545117162168026, "se": 0.19927686095053043, "per_batch": null}}}, "s4_arith": {"acc_off": {"in": 0.34375, "transfer": 0.24166666666666667, "held-out": 0.35}, "acc_all_on": {"in": 0.60625, "transfer": 0.22916666666666666, "held-out": 0.3375}, "acc_member_masked": {"in": 0.475, "transfer": 0.25833333333333336, "held-out": 0.2875}, "acc_member_alone": {"in": 0.38125, "transfer": 0.24583333333333332, "held-out": 0.3875}, "margin_all_on_minus_off": {"in": {"mean": 1.5874161858824665, "se": 0.23010487651149647, "per_batch": null}, "transfer": {"mean": -0.3282564729452133, "se": 0.07549495223635017, "per_batch": null}, "held-out": {"mean": -0.1319847047328949, "se": 0.11231124201537564, "per_batch": null}}}}, "probes_all_on": {"p0_grounding": {"acc": 0.43333333333333335, "n": 30, "families": {"count": 0.125, "permanence": 0.5714285714285714, "sequence": 0.625, "attribute": 0.42857142857142855}}, "p1_self_other": {"acc": 0.5333333333333333, "n": 30, "families": {"belief": 0.25, "reality": 0.5, "perspective": 0.8, "reference": 1.0, "second_order": 0.0}}, "p2_difference": {"acc": 0.5666666666666667, "n": 30, "families": {"category": 1.0, "antonym": 0.6666666666666666, "odd_one_out": 0.3333333333333333, "property": 0.5, "analogy": 0.3333333333333333}}, "p3_logic": {"acc": 0.26666666666666666, "n": 30, "families": {"d1": 0.5, "d2": 0.5, "d3": 0.5, "d4": 0.0, "d5": 0.0, "d6": 0.0, "negation": 0.25, "contrapositive": 0.25}}, "p4_arithmetic": {"acc": 0.4, "n": 30, "families": {"add2": 0.4, "add3": 0.0, "sub2": 0.75, "sub3": 0.6, "mul1": 0.25, "mul2": 0.3333333333333333, "word": 0.4}}, "p5_causal": {"acc": 0.5, "n": 30, "families": {"effect": 0.5714285714285714, "cause": 0.42857142857142855, "counterfactual": 0.5, "prevention": 0.4, "chain": 0.6}}, "p6_learning": {"acc": 0.3333333333333333, "n": 30, "families": {"error_detect": 0.125, "correction": 0.25, "strategy": 0.7142857142857143, "check": 0.2857142857142857}}, "p7_composite": {"acc": 0.36666666666666664, "n": 30, "families": {"logic_arith": 0.375, "causal_category": 0.375, "belief_causal": 0.2857142857142857, "three_step": 0.42857142857142855}}, "p8_articulation": {"acc": 0.4666666666666667, "n": 30, "families": {"definition": 0.375, "register": 0.42857142857142855, "grammar": 0.875, "paraphrase": 0.14285714285714285}}}, "probes_mean_off": 0.41111111111111115, "probes_mean_all_on": 0.42962962962962964, "gates": {"s1_perspective": {"mean": 0.2099039478571495, "max": 0.5712027580899203}, "s2_concept": {"mean": 0.27674460361666475, "max": 0.4977012170865557}, "s3_rules": {"mean": 0.12033936123137423, "max": 0.1738736117222015}, "s4_arith": {"mean": 0.10767731599767934, "max": 0.15932411779931022}, "s5_causal": {"mean": 0.18518384792916404, "max": 0.3681551185947585}}, "chain_proof": "compiled chain: 13 cache entries (limit 256), 0 limit warnings", "curve": [{"step": 1, "train_bpb": {"s5_causal": 0.48256010066317456}, "quiet_kl": {"s5_causal": 0.00012739913654513657}}, {"step": 50, "train_bpb": {"s5_causal": 0.46273601822213245}, "quiet_kl": {"s5_causal": 0.00022650838174740785}}, {"step": 100, "train_bpb": {"s5_causal": 0.4352985604974664}, "quiet_kl": {"s5_causal": 0.0005840317354886792}}, {"step": 150, "train_bpb": {"s5_causal": 0.4017903972488909}, "quiet_kl": {"s5_causal": 0.0006464591546682641}}, {"step": 200, "train_bpb": {"s5_causal": 0.3647996167071769}, "quiet_kl": {"s5_causal": 0.0006799186731223017}}, {"step": 250, "train_bpb": {"s5_causal": 0.3277244537377306}, "quiet_kl": {"s5_causal": 0.0007688259473070502}}, {"step": 300, "train_bpb": {"s5_causal": 0.289507785160233}, "quiet_kl": {"s5_causal": 0.0007398774055764079}}, {"step": 350, "train_bpb": {"s5_causal": 0.24717576479811398}, "quiet_kl": {"s5_causal": 0.0008841762389056384}}, {"step": 400, "train_bpb": {"s5_causal": 0.20819544423573003}, "quiet_kl": {"s5_causal": 0.0008365676627727225}}, {"step": 450, "train_bpb": {"s5_causal": 0.17481709779159182}, "quiet_kl": {"s5_causal": 0.0008536538394400849}}, {"step": 500, "train_bpb": {"s5_causal": 0.14292952706209638}, "quiet_kl": {"s5_causal": 0.0008562629256630317}}, {"step": 550, "train_bpb": {"s5_causal": 0.11564968520257513}, "quiet_kl": {"s5_causal": 0.0007896375481504947}}, {"step": 600, "train_bpb": {"s5_causal": 0.09516689759989946}, "quiet_kl": {"s5_causal": 0.0007412045559613034}}, {"step": 650, "train_bpb": {"s5_causal": 0.08069618083103995}, "quiet_kl": {"s5_causal": 0.0007247306162025779}}, {"step": 700, "train_bpb": {"s5_causal": 0.07193030659334344}, "quiet_kl": {"s5_causal": 0.0007330642815213651}}, {"step": 750, "train_bpb": {"s5_causal": 0.06309123922666865}, "quiet_kl": {"s5_causal": 0.0006985399028053507}}, {"step": 800, "train_bpb": {"s5_causal": 0.058628970165608696}, "quiet_kl": {"s5_causal": 0.0006497892597690225}}, {"step": 850, "train_bpb": {"s5_causal": 0.055995267191327025}, "quiet_kl": {"s5_causal": 0.0006124301016097888}}, {"step": 900, "train_bpb": {"s5_causal": 0.05335847325885273}, "quiet_kl": {"s5_causal": 0.0005771298077888787}}, {"step": 950, "train_bpb": {"s5_causal": 0.05191219515291353}, "quiet_kl": {"s5_causal": 0.0005546556925401092}}, {"step": 1000, "train_bpb": {"s5_causal": 0.05049779521369247}, "quiet_kl": {"s5_causal": 0.0005248704273253679}}, {"step": 1050, "train_bpb": {"s5_causal": 0.04961225364219264}, "quiet_kl": {"s5_causal": 0.0004938355169724673}}, {"step": 1100, "train_bpb": {"s5_causal": 0.04967114738115409}, "quiet_kl": {"s5_causal": 0.00047423944459296764}}, {"step": 1150, "train_bpb": {"s5_causal": 0.049504874876767385}, "quiet_kl": {"s5_causal": 0.00048552474298048766}}, {"step": 1200, "train_bpb": {"s5_causal": 0.047943663559974654}, "quiet_kl": {"s5_causal": 0.00047749927209224553}}, {"step": 1250, "train_bpb": {"s5_causal": 0.04756512867053404}, "quiet_kl": {"s5_causal": 0.0004536332900170237}}, {"step": 1300, "train_bpb": {"s5_causal": 0.047109847168080844}, "quiet_kl": {"s5_causal": 0.000413103605969809}}, {"step": 1350, "train_bpb": {"s5_causal": 0.04689733121610607}, "quiet_kl": {"s5_causal": 0.00044183028629049656}}, {"step": 1400, "train_bpb": {"s5_causal": 0.047028037762038906}, "quiet_kl": {"s5_causal": 0.0004168206168105826}}, {"step": 1450, "train_bpb": {"s5_causal": 0.04602538014363404}, "quiet_kl": {"s5_causal": 0.00038557536085136233}}, {"step": 1500, "train_bpb": {"s5_causal": 0.046036768243583805}, "quiet_kl": {"s5_causal": 0.0004239655626588501}}, {"step": 1550, "train_bpb": {"s5_causal": 0.046101995996087386}, "quiet_kl": {"s5_causal": 0.0003920695025590248}}, {"step": 1600, "train_bpb": {"s5_causal": 0.04566146958393122}, "quiet_kl": {"s5_causal": 0.0003752727838582359}}, {"step": 1650, "train_bpb": {"s5_causal": 0.04521774434821361}, "quiet_kl": {"s5_causal": 0.00039212245523231106}}, {"step": 1700, "train_bpb": {"s5_causal": 0.04532218951957174}, "quiet_kl": {"s5_causal": 0.0003712935996009037}}, {"step": 1750, "train_bpb": {"s5_causal": 0.0451356418192411}, "quiet_kl": {"s5_causal": 0.00037527971726376564}}, {"step": 1800, "train_bpb": {"s5_causal": 0.04487873698384488}, "quiet_kl": {"s5_causal": 0.0003534854407189414}}, {"step": 1850, "train_bpb": {"s5_causal": 0.044952003474172705}, "quiet_kl": {"s5_causal": 0.00041100562986684964}}, {"step": 1900, "train_bpb": {"s5_causal": 0.0449542574143014}, "quiet_kl": {"s5_causal": 0.00038041080988477917}}, {"step": 1950, "train_bpb": {"s5_causal": 0.04468983091766972}, "quiet_kl": {"s5_causal": 0.00037928345816908406}}, {"step": 2000, "train_bpb": {"s5_causal": 0.04461915781822778}, "quiet_kl": {"s5_causal": 0.00035743289219681174}}, {"step": 2050, "train_bpb": {"s5_causal": 0.04467334224225197}, "quiet_kl": {"s5_causal": 0.0003355463469051756}}, {"step": 2100, "train_bpb": {"s5_causal": 0.044146638387189435}, "quiet_kl": {"s5_causal": 0.00033191879891091957}}, {"step": 2150, "train_bpb": {"s5_causal": 0.04457681364729136}, "quiet_kl": {"s5_causal": 0.0003520575995207764}}, {"step": 2200, "train_bpb": {"s5_causal": 0.043977206346528075}, "quiet_kl": {"s5_causal": 0.0003253212900017388}}, {"step": 2250, "train_bpb": {"s5_causal": 0.044047933351782156}, "quiet_kl": {"s5_causal": 0.0003176975224050693}}, {"step": 2300, "train_bpb": {"s5_causal": 0.043850737453112486}, "quiet_kl": {"s5_causal": 0.0003083313937531784}}, {"step": 2350, "train_bpb": {"s5_causal": 0.04383906402319601}, "quiet_kl": {"s5_causal": 0.0003397114170365967}}, {"step": 2400, "train_bpb": {"s5_causal": 0.04400233414045528}, "quiet_kl": {"s5_causal": 0.00034711130370851606}}, {"step": 2450, "train_bpb": {"s5_causal": 0.043887612092050256}, "quiet_kl": {"s5_causal": 0.00034979217598447577}}, {"step": 2500, "train_bpb": {"s5_causal": 0.043973335715736674}, "quiet_kl": {"s5_causal": 0.0003528247619397007}}, {"step": 2550, "train_bpb": {"s5_causal": 0.043765208277310695}, "quiet_kl": {"s5_causal": 0.00034165946184657516}}, {"step": 2600, "train_bpb": {"s5_causal": 0.04434015186858122}, "quiet_kl": {"s5_causal": 0.0003676405193982646}}, {"step": 2650, "train_bpb": {"s5_causal": 0.0440447830058304}, "quiet_kl": {"s5_causal": 0.00031815161288250237}}, {"step": 2700, "train_bpb": {"s5_causal": 0.043656463002235205}, "quiet_kl": {"s5_causal": 0.00032517564250156283}}, {"step": 2750, "train_bpb": {"s5_causal": 0.043677581235635865}, "quiet_kl": {"s5_causal": 0.0003214837648556568}}, {"step": 2800, "train_bpb": {"s5_causal": 0.04402955087865552}, "quiet_kl": {"s5_causal": 0.0003464984821039252}}, {"step": 2850, "train_bpb": {"s5_causal": 0.0433482386513107}, "quiet_kl": {"s5_causal": 0.0003146877489052713}}, {"step": 2900, "train_bpb": {"s5_causal": 0.04329225727730786}, "quiet_kl": {"s5_causal": 0.0002953191005508415}}, {"step": 2950, "train_bpb": {"s5_causal": 0.04321632381976639}, "quiet_kl": {"s5_causal": 0.0003004885598784313}}, {"step": 3000, "train_bpb": {"s5_causal": 0.04321832456918628}, "quiet_kl": {"s5_causal": 0.0003083940441138111}}, {"step": 3050, "train_bpb": {"s5_causal": 0.04333039894468126}, "quiet_kl": {"s5_causal": 0.000308277118310798}}, {"step": 3100, "train_bpb": {"s5_causal": 0.04352012337955166}, "quiet_kl": {"s5_causal": 0.0003145809468696825}}, {"step": 3150, "train_bpb": {"s5_causal": 0.04318608932348928}, "quiet_kl": {"s5_causal": 0.00031198142707580703}}, {"step": 3200, "train_bpb": {"s5_causal": 0.043479236976559636}, "quiet_kl": {"s5_causal": 0.00028890647372463717}}, {"step": 3250, "train_bpb": {"s5_causal": 0.04331085720099868}, "quiet_kl": {"s5_causal": 0.0003204650233965367}}, {"step": 3300, "train_bpb": {"s5_causal": 0.04336379405212461}, "quiet_kl": {"s5_causal": 0.0003430244661285542}}, {"step": 3350, "train_bpb": {"s5_causal": 0.04312640097093626}, "quiet_kl": {"s5_causal": 0.0003073601922369562}}, {"step": 3400, "train_bpb": {"s5_causal": 0.0431224802501978}, "quiet_kl": {"s5_causal": 0.00032204276038100944}}, {"step": 3450, "train_bpb": {"s5_causal": 0.04299334761424323}, "quiet_kl": {"s5_causal": 0.0003684659893042408}}, {"step": 3500, "train_bpb": {"s5_causal": 0.042920722941670625}, "quiet_kl": {"s5_causal": 0.00030332700203871354}}, {"step": 3550, "train_bpb": {"s5_causal": 0.043113903529143835}, "quiet_kl": {"s5_causal": 0.00032717513822717593}}, {"step": 3600, "train_bpb": {"s5_causal": 0.04299616915086437}, "quiet_kl": {"s5_causal": 0.0003400566938216798}}, {"step": 3650, "train_bpb": {"s5_causal": 0.04292773333073183}, "quiet_kl": {"s5_causal": 0.00029363154899328945}}, {"step": 3700, "train_bpb": {"s5_causal": 0.04273112695634259}, "quiet_kl": {"s5_causal": 0.0003106381944962777}}, {"step": 3750, "train_bpb": {"s5_causal": 0.042633735008474644}, "quiet_kl": {"s5_causal": 0.00030504668393405154}}, {"step": 3800, "train_bpb": {"s5_causal": 0.042748863365892335}, "quiet_kl": {"s5_causal": 0.000303326983412262}}, {"step": 3850, "train_bpb": {"s5_causal": 0.04270073799814719}, "quiet_kl": {"s5_causal": 0.00030906730418791997}}, {"step": 3900, "train_bpb": {"s5_causal": 0.042642227350540504}, "quiet_kl": {"s5_causal": 0.00029756936884950847}}, {"step": 3950, "train_bpb": {"s5_causal": 0.04300612941859994}, "quiet_kl": {"s5_causal": 0.00040992689115228133}}, {"step": 4000, "train_bpb": {"s5_causal": 0.04305106628174267}, "quiet_kl": {"s5_causal": 0.0003536857929429971}}], "steps_done": {"s1_perspective": 3200, "s2_concept": 2400, "s3_rules": 800, "s4_arith": 800, "s5_causal": 4000, "s6_tryfail": 0, "s7_mixed": 0, "s8_register": 0}, "stage_seconds": 2690.410659313202, "inert_at_birth": true}, {"close": 6, "attached": ["s1_perspective", "s2_concept", "s3_rules", "s4_arith", "s5_causal", "s6_tryfail"], "rows": {"s1_perspective": {"off": 0.5538333013185592, "all_on": 0.048797346957739504, "all_on_minus_off": {"mean": -0.5050359543608197, "se": 0.004832520573614138, "per_batch": [-0.5054571183749513, -0.5246725142286237, -0.5162597732578137, -0.511185344527239, -0.492715020512767, -0.5068229830902724, -0.4805988696317713, -0.5025760112631192]}}, "s2_concept": {"off": 0.7211839326967642, "all_on": 0.11188156547638788, "all_on_minus_off": {"mean": -0.6093023672203762, "se": 0.0037988458477961136, "per_batch": [-0.6095218641093832, -0.6083938836469526, -0.595040247725047, -0.6315069330660443, -0.607986274015492, -0.6070693485124169, -0.6004772105091828, -0.6144231761784912]}}, "s3_rules": {"off": 0.9296131361815035, "all_on": 0.22408305361735653, "all_on_minus_off": {"mean": -0.7055300825641471, "se": 0.005269007632384245, "per_batch": [-0.6858040809689786, -0.7053346457792727, -0.7201927859309262, -0.6892778725424311, -0.6927485254325052, -0.726144933494637, -0.7108631140825832, -0.7138747022818418]}}, "s4_arith": {"off": 1.2338748670942346, "all_on": 0.3379026848682451, "all_on_minus_off": {"mean": -0.8959721822259895, "se": 0.005824882361550658, "per_batch": [-0.9103632900530803, -0.8886026262090481, -0.9110276590333632, -0.8703108739307501, -0.8755902898577798, -0.9137369877237306, -0.8978049019087644, -0.9003408290913992]}}, "s5_causal": {"off": 0.6604376826115352, "all_on": 0.04301686327236076, "all_on_minus_off": {"mean": -0.6174208193391744, "se": 0.006909360217757065, "per_batch": [-0.5834439600891759, -0.6266083780938083, -0.6285370721574514, -0.6288753190360358, -0.6429261511267077, -0.61573802654, -0.5945137900201536, -0.6187238576500631]}}, "s6_tryfail": {"off": 0.6728212611401647, "all_on": 0.05490541224602387, "all_on_minus_off": {"mean": -0.6179158488941409, "se": 0.0036178670880528717, "per_batch": [-0.6279002526480535, -0.6231004748364488, -0.6213122959730184, -0.6268040244825382, -0.6223293153788956, -0.6171316987001825, -0.6048951438402094, -0.5998535852937807]}}, "s7_mixed": {"off": 0.7829759286650222, "all_on": 0.7747712329571779, "all_on_minus_off": {"mean": -0.008204695707844448, "se": 0.008102894349188948, "per_batch": [0.00046418117467861286, 0.0010812549259744397, 0.001072827776082197, 0.00244352950345883, 0.0007637749524816995, -0.06006769253624977, 0.010629817693203036, -0.022025259152384624]}}, "s8_register": {"off": 0.4906504977761471, "all_on": 0.3778845907909504, "all_on_minus_off": {"mean": -0.11276590698519673, "se": 0.002027868532052983, "per_batch": [-0.10262952901515043, -0.11564801374618705, -0.10744095865135112, -0.11726232889851079, -0.11029161408506483, -0.11956440267163493, -0.11701746859934481, -0.11227294021432982]}}, "web": {"off": 0.9505967609111062, "all_on": 0.9527351179495246, "all_on_minus_off": {"mean": 0.0021383570384184286, "se": 0.000426921051071179, "per_batch": [0.0012652763623977048, 0.0007972255780744897, 0.0009640487494114724, 0.0016616963726364675, 0.003811909465048835, 0.003478865061652958, 0.0032986272435486796, 0.0018292074745768216]}}}, "member_in_group": {"s1_perspective": {"s1_perspective": {"masked": 0.5047585517194358, "all_on_minus_masked": {"mean": -0.4559612047616963, "se": 0.004398367409453014, "per_batch": [-0.4594003364565421, -0.47396312865208556, -0.4650013760640588, -0.46061479194560984, -0.44232100812224856, -0.4575301756132285, -0.4346929594938314, -0.45416586174596596]}}, "s2_concept": {"masked": 0.3453099131474122, "all_on_minus_masked": {"mean": -0.23342834767102433, "se": 0.0014877771201188097, "per_batch": [-0.2304771321002908, -0.23250959082144046, -0.22703370621363134, -0.24039697316398517, -0.23301757382751082, -0.23334051425016844, -0.2323615245079632, -0.23828976648320427]}}, "s3_rules": {"masked": 0.4047210538523064, "all_on_minus_masked": {"mean": -0.18063800023494983, "se": 0.00306595006463749, "per_batch": [-0.17810312644605153, -0.1754844756083517, -0.18225151996751335, -0.17046864459160063, -0.18113049405352685, -0.20005989350773, -0.1800701350196322, -0.17753571268519244]}}, "s4_arith": {"masked": 0.48673440444588045, "all_on_minus_masked": {"mean": -0.14883171957763533, "se": 0.005854247214307735, "per_batch": [-0.15774682993330236, -0.17521992329566205, -0.11878957783383526, -0.15129900687197345, -0.15729365564827885, -0.1416875199523604, -0.1502772364453661, -0.13834000664030405]}}, "s5_causal": {"masked": 0.060734746058223564, "all_on_minus_masked": {"mean": -0.017717882785862796, "se": 0.0008508729437896051, "per_batch": [-0.01842629376862215, -0.017156521672241756, -0.01691086595331559, -0.018973470008486773, -0.016679436424290484, -0.02250807894685057, -0.014125706350062439, -0.01696268916303261]}}, "s6_tryfail": {"masked": 0.09343194656074245, "all_on_minus_masked": {"mean": -0.038526534314718586, "se": 0.0024161175051779903, "per_batch": [-0.048683816910338684, -0.04017644786034924, -0.03884284064098063, -0.03340896817010261, -0.04049713101071431, -0.034686449926253714, -0.026933240923671206, -0.04498337907533826]}}, "web": {"masked": 0.9522361317858775, "all_on_minus_masked": {"mean": 0.0004989861636470749, "se": 0.0004983142441016544, "per_batch": [5.572237887951914e-05, -0.0011473822552312551, -0.0011931296403607794, -0.0001833335058197827, 0.0016889556227981384, 0.002088041364125437, 0.002416012279320756, 0.0002670030654645661]}}}, "s2_concept": {"s1_perspective": {"masked": 0.05432210957236961, "all_on_minus_masked": {"mean": -0.005524762614630105, "se": 0.00046447413244369944, "per_batch": [-0.004871376338942017, -0.00621498004989203, -0.006862973057138219, -0.0036330098869465227, -0.006931733870736305, -0.006052445695910999, -0.003687721867752257, -0.0059438601497224935]}}, "s2_concept": {"masked": 0.23704136323119357, "all_on_minus_masked": {"mean": -0.12515979775480568, "se": 0.0012693768084306473, "per_batch": [-0.12968532370093488, -0.1253157712393792, -0.12126011936239517, -0.12959302276199997, -0.12700263082362, -0.12565966129869505, -0.12022713781673476, -0.12253471503468649]}}, "s3_rules": {"masked": 0.33074329433788696, "all_on_minus_masked": {"mean": -0.10666024072053046, "se": 0.0010683837407400906, "per_batch": [-0.10422548501949458, -0.10521240746147084, -0.10710256128794712, -0.10708555650334309, -0.10152987194553367, -0.10730494187234996, -0.11098242539791808, -0.10983867627618629]}}, "s4_arith": {"masked": 0.45664216469332974, "all_on_minus_masked": {"mean": -0.1187394798250847, "se": 0.003774068246105638, "per_batch": [-0.13123527434617327, -0.11816308803240477, -0.1033464816909348, -0.1167175523540715, -0.1043608353657236, -0.1293107884830184, -0.11901130646646013, -0.12777051186189115]}}, "s5_causal": {"masked": 0.0446971357867121, "all_on_minus_masked": {"mean": -0.0016802725143513306, "se": 0.00035774277822330083, "per_batch": [-0.0019981884907363037, -0.0009383373431074843, -0.003451885283298789, -0.0022498716641294256, -0.002184932090055193, -0.0003769537239442372, -0.0016075245248435913, -0.0006344869946956211]}}, "s6_tryfail": {"masked": 0.058203156063065184, "all_on_minus_masked": {"mean": -0.0032977438170413067, "se": 0.0004098946524451417, "per_batch": [-0.0036957513077644083, -0.0057195732751660205, -0.002643314224725135, -0.003687028562690965, -0.0024689829357864845, -0.002504604842346443, -0.002105293373790035, -0.003557402014060962]}}, "web": {"masked": 0.9528742949097353, "all_on_minus_masked": {"mean": -0.00013917696021070736, "se": 0.0003464823175465939, "per_batch": [-0.001541910456310358, 0.0014918635049093742, 0.0003750941615318082, 0.000666776737395125, -0.0006920581870719644, -0.0010637986869118654, -8.065986325467289e-05, -0.0002687228919731055]}}}, "s3_rules": {"s1_perspective": {"masked": 0.04879591096978553, "all_on_minus_masked": {"mean": 1.435987953981796e-06, "se": 0.0001024534354613978, "per_batch": [0.0003765667629797992, -0.0001797971125613862, -0.000217692414787439, 0.00011628714427245473, -0.0004112212671286558, -4.654280488974871e-05, -4.930258899030898e-05, 0.00042319018473713915]}}, "s2_concept": {"masked": 0.11108403086698755, "all_on_minus_masked": {"mean": 0.0007975346094003175, "se": 6.440667848003038e-05, "per_batch": [0.0007085685215547782, 0.0006977873591288014, 0.0011211549009746413, 0.0005778509579833502, 0.0009143457633121188, 0.0006106673975511262, 0.0008960941044893062, 0.0008538078702084173]}}, "s3_rules": {"masked": 0.31661550001095395, "all_on_minus_masked": {"mean": -0.0925324463935974, "se": 0.0010165731385940189, "per_batch": [-0.08885874917357722, -0.09021139272261322, -0.0944380168523862, -0.09509851622302487, -0.08947169534125224, -0.09436750396553245, -0.09638260318786068, -0.09143109368253224]}}, "s4_arith": {"masked": 0.33799874792766765, "all_on_minus_masked": {"mean": -9.606305942255755e-05, "se": 0.00011384707229067808, "per_batch": [7.457597698040663e-05, -9.261265748966574e-05, -0.00038309135479691614, -0.0006497504549596855, -0.0002357667165015509, 0.00013780109900385362, 0.0003768139880404031, 3.525644342694534e-06]}}, "s5_causal": {"masked": 0.043013404136933844, "all_on_minus_masked": {"mean": 3.4591354269239427e-06, "se": 0.00010545483336404358, "per_batch": [5.805758081079554e-05, 0.00019159673474793137, -0.0002336760524020015, 0.00014985600793773757, 0.00034935857016751837, -0.00015480050915004512, -0.0005564552413244578, 0.0002237359926279131]}}, "s6_tryfail": {"masked": 0.05507706571134402, "all_on_minus_masked": {"mean": -0.00017165346532013997, "se": 8.612867556519337e-05, "per_batch": [-0.0004352880893337685, 0.00036830084682271114, -0.00037155776827341863, -0.00013642523779692356, -0.00016522695735860654, -0.0002667880871509293, -0.00022673762733125408, -0.0001395048021389303]}}, "web": {"masked": 0.9524110058950579, "all_on_minus_masked": {"mean": 0.00032411205446664093, "se": 0.00018886526787357188, "per_batch": [0.000555503962286874, 0.0007632590045291154, -3.3536616918294904e-05, 0.0009190752862108242, 4.093187090536965e-05, 0.0006540500212313116, 0.00040562108205988157, -0.0007120081745719542]}}}, "s4_arith": {"s1_perspective": {"masked": 0.04871582986845945, "all_on_minus_masked": {"mean": 8.151708928005093e-05, "se": 4.25771852100445e-05, "per_batch": [3.509520969169344e-05, 6.463322997737886e-05, -5.174528007834828e-05, 0.0003193825315679147, 0.00012861077609834654, 0.000156665446020339, 4.1587554761760015e-05, -4.2092753798676774e-05]}}, "s2_concept": {"masked": 0.1124235459591362, "all_on_minus_masked": {"mean": -0.0005419804827483214, "se": 6.957609266820099e-05, "per_batch": [-0.0006488152992956131, -0.0004547866223753444, -0.00019751132560025253, -0.0007305500541181786, -0.0005510324133644467, -0.000388498059383402, -0.0005503767295080148, -0.0008142733583413192]}}, "s3_rules": {"masked": 0.2239882105598635, "all_on_minus_masked": {"mean": 9.484305749300065e-05, "se": 9.418867707871503e-05, "per_batch": [0.00047288779637866263, 0.00014186418913045284, 0.0004395231621112661, -0.00012451543922467612, 0.00010445796256780548, -0.00031907081301321694, -1.4704516648783583e-05, 5.8302118642494793e-05]}}, "s4_arith": {"masked": 0.48612123792543394, "all_on_minus_masked": {"mean": -0.14821855305718884, "se": 0.0022416608827440457, "per_batch": [-0.15587746450974904, -0.1380309538167036, -0.14976915669905472, -0.14556110669557487, -0.14098561575852664, -0.15526172362399748, -0.1497763369747282, -0.15048606637917633]}}, "s5_causal": {"masked": 0.043072992094614425, "all_on_minus_masked": {"mean": -5.6128822253659165e-05, "se": 8.958540289447796e-05, "per_batch": [-5.702031044778344e-05, 0.00047265669469155613, -0.000410264613633228, -7.493069119779427e-05, -5.198713068112204e-05, -0.0002515219396579324, -8.894996447202791e-05, 1.298737736905864e-05]}}, "s6_tryfail": {"masked": 0.05934209638864507, "all_on_minus_masked": {"mean": -0.004436684142621199, "se": 0.0004425414351223769, "per_batch": [-0.005146043381285603, -0.007088780903263245, -0.003594415905201341, -0.0032612102680703303, -0.004202245589013227, -0.003528025227510459, -0.0038681370429651807, -0.004804614823660208]}}, "web": {"masked": 0.9524081789302343, "all_on_minus_masked": {"mean": 0.0003269390192901789, "se": 0.0001333752396286293, "per_batch": [-8.796912591635397e-05, 0.00025530824520580975, 0.00010946695727420686, 0.00023028476950526233, 0.00011101480113184792, 0.0010944975900909482, 0.0006699584164360228, 0.00023295050059368716]}}}, "s5_causal": {"s1_perspective": {"masked": 0.048220451309637105, "all_on_minus_masked": {"mean": 0.0005768956481023929, "se": 0.00013398849293798476, "per_batch": [0.0005622220345861961, 0.0004078138608584331, 0.000752402599693433, -1.7574477135046118e-05, 0.0011174680228967782, 0.001016073501297475, 0.0005228971265748492, 0.0002538625160470251]}}, "s2_concept": {"masked": 0.11257789367021329, "all_on_minus_masked": {"mean": -0.0006963281938253945, "se": 0.00019369888484346758, "per_batch": [-0.001111190656140279, -0.0005853751989586126, -0.00019781229523926913, -0.00030476400624457334, -0.0008394795656084114, -2.9989474744196443e-05, -0.0007846493467301985, -0.0017173650069376156]}}, "s3_rules": {"masked": 0.22376450950121157, "all_on_minus_masked": {"mean": 0.00031854411614496217, "se": 0.0001791901905905551, "per_batch": [0.0007564656898200739, 0.0007175546150623435, -0.00019150268173573348, -0.00047142594384633196, 0.000651792748938701, -2.373360581905959e-05, 0.000888720348333566, 0.00022048175840613804]}}, "s4_arith": {"masked": 0.33766619797439307, "all_on_minus_masked": {"mean": 0.00023648689385202815, "se": 9.859099822910126e-05, "per_batch": [0.0007909267134866838, 0.00031915680433863836, -9.237618134472214e-05, 0.0004575813404518736, 0.00013378100453992348, 7.881104975787645e-05, 7.481245312529472e-05, 0.00012920196646065696]}}, "s5_causal": {"masked": 0.4749034814168829, "all_on_minus_masked": {"mean": -0.4318866181445221, "se": 0.005505612180170685, "per_batch": [-0.4094035368087937, -0.4329518286343277, -0.4327736922292387, -0.4372093397733041, -0.4545420737947284, -0.44679736180999, -0.4113220838960359, -0.43009302820975825]}}, "s6_tryfail": {"masked": 0.05567234872332967, "all_on_minus_masked": {"mean": -0.0007669364773057958, "se": 0.00015284018385378523, "per_batch": [-0.000560405467836457, -0.0007100035017979095, -0.0014800074509155742, -0.0005454483516692393, -0.001382031084502175, -0.0007094123114355644, -0.00028912433393173814, -0.00045905931635770864]}}, "web": {"masked": 0.9525273414094511, "all_on_minus_masked": {"mean": 0.00020777654007363144, "se": 0.0001524944673956985, "per_batch": [0.0003728583870705737, -4.5231437176940226e-05, -0.000276978059214561, 0.0006238670660048129, 0.0004861949539891608, -0.00020250957139089643, 0.0008596552803377122, -0.00015564429903081045]}}}, "s6_tryfail": {"s1_perspective": {"masked": 0.04885903095397748, "all_on_minus_masked": {"mean": -6.168399623797394e-05, "se": 0.00011297947258806259, "per_batch": [4.721998657752219e-05, 4.8327124892451334e-05, -0.0006923215355060727, 0.000212850028274085, 0.00023703508855166255, -0.0003980215986749344, 3.3912828967010145e-05, 1.7526107014484427e-05]}}, "s2_concept": {"masked": 0.1116441286471127, "all_on_minus_masked": {"mean": 0.0002374368292751665, "se": 0.00010781961618136669, "per_batch": [-0.0001504848195049563, 0.0007558100059636419, 0.0005556866938534188, 7.77576560213461e-05, 0.0002818365691014263, -4.7434964891096554e-05, 0.00031496472722387925, 0.00011135876643367237]}}, "s3_rules": {"masked": 0.2241419791731084, "all_on_minus_masked": {"mean": -5.8925555751869474e-05, "se": 0.00012113432595847113, "per_batch": [0.00046635245564588534, -0.0003457711195596902, 0.0003716545085145351, -0.0005197530687387764, 3.9513014035735816e-05, -0.0002871680312781788, -0.00011935595969880808, -7.687624493565859e-05]}}, "s4_arith": {"masked": 0.3385110090021782, "all_on_minus_masked": {"mean": -0.000608324133933115, "se": 0.00015725304441320477, "per_batch": [-0.0003434923494357589, -0.00041989564208155805, -0.00014958191058794812, -0.0010091941952629635, -0.0015008925940797058, -0.0006733120781278745, -0.0003240153142255453, -0.00044620898766356554]}}, "s5_causal": {"masked": 0.04318483153912111, "all_on_minus_masked": {"mean": -0.00016796826676034483, "se": 0.0001473309236681627, "per_batch": [-0.001050590956771423, 0.00013497144695134855, 7.728739096039133e-05, -7.898840686658415e-05, 2.0181089187186474e-05, -0.00032084438410024535, 0.0002500385892942464, -0.0003758009027376788]}}, "s6_tryfail": {"masked": 0.4832413561899892, "all_on_minus_masked": {"mean": -0.4283359439439653, "se": 0.00298350154811217, "per_batch": [-0.4342226353138422, -0.41430377667356255, -0.43715194056357864, -0.42874979264917024, -0.430626574824542, -0.43759961140377457, -0.4250243797085898, -0.4190088404146624]}}, "web": {"masked": 0.9527085788767133, "all_on_minus_masked": {"mean": 2.6539072811276876e-05, "se": 0.00017269730787175826, "per_batch": [0.00012726716163857787, -0.0001655333014555227, 0.0003798236844304581, -0.00048249732699567893, -0.00017765807834124736, 0.0001485930103455324, 0.0009422929440773276, -0.000559975511209232]}}}}, "member_alone": {"s1_perspective": {"s1_perspective": {"alone": 0.05251879489204885, "alone_minus_off": {"mean": -0.5013145064265103, "se": 0.004723722359706749, "per_batch": [-0.5020161539899823, -0.5206146212027206, -0.511447510580648, -0.5087860038161363, -0.4873552833133265, -0.5025627202288822, -0.47857692335015567, -0.4991568349302314]}}, "s2_concept": {"alone": 0.23945795595196098, "alone_minus_off": {"mean": -0.48172597674480316, "se": 0.0029625557017275653, "per_batch": [-0.4753027767620657, -0.4804484116845556, -0.4714891474705084, -0.49853255940543073, -0.48017119714919604, -0.4800707377832608, -0.47906046869643004, -0.4887325150069781]}}, "s3_rules": {"alone": 0.508758760689348, "alone_minus_off": {"mean": -0.42085437549215554, "se": 0.004040788518956882, "per_batch": [-0.4068824458170247, -0.41901358606093464, -0.4391522815246798, -0.40577522151077006, -0.4185499638298712, -0.4321406348403284, -0.4199813324373397, -0.42533953791629586]}}, "s4_arith": {"alone": 0.7225499640202783, "alone_minus_off": {"mean": -0.5113249030739563, "se": 0.003602454892240459, "per_batch": [-0.5097063582233129, -0.5105177293744209, -0.5238398924712184, -0.49785761349211977, -0.5006369821170508, -0.5244675431555423, -0.5047830108770632, -0.5187900948809221]}}, "s5_causal": {"alone": 0.4626770842822263, "alone_minus_off": {"mean": -0.19776059832930892, "se": 0.003074781006523783, "per_batch": [-0.18829842997184976, -0.20876677318163606, -0.20486710956898452, -0.20499300086941613, -0.19802718756151977, -0.18360145478979822, -0.19417430322990303, -0.19935652746136384]}}, "s6_tryfail": {"alone": 0.49460838826503134, "alone_minus_off": {"mean": -0.1782128728751334, "se": 0.002204184674331319, "per_batch": [-0.1800673403015557, -0.18936055182791578, -0.1759819784216351, -0.1815247642806299, -0.17975192411987334, -0.17801537229844877, -0.17237782399882867, -0.16862322775217997]}}, "web": {"alone": 0.9520279682848395, "alone_minus_off": {"mean": 0.0014312073737332287, "se": 0.0004766934609748917, "per_batch": [0.0014245322970964924, -0.0004134462926742577, 0.00041473616255571777, 1.4618525323362164e-05, 0.0028610173884282553, 0.0033452345419324736, 0.002158726233630004, 0.0016442401335737822]}}}, "s2_concept": {"s1_perspective": {"alone": 0.5265382088974727, "alone_minus_off": {"mean": -0.02729509242108652, "se": 0.0008478035966868174, "per_batch": [-0.023477093695305706, -0.029719504978010347, -0.026994955823041678, -0.02971271166330125, -0.0301825252697957, -0.02641580424626122, -0.025215322347576063, -0.02664282134540019]}}, "s2_concept": {"alone": 0.35026574371684355, "alone_minus_off": {"mean": -0.3709181889799206, "se": 0.0024118108231179045, "per_batch": [-0.37237346048863007, -0.3698774117831641, -0.3638556972371791, -0.385719013225945, -0.3722404964016817, -0.36824759669443125, -0.36423210276642376, -0.37079973324191]}}, "s3_rules": {"alone": 0.6188593181726778, "alone_minus_off": {"mean": -0.31075381800882573, "se": 0.002804570723181844, "per_batch": [-0.3059784617336435, -0.31155671901437876, -0.3166699782028437, -0.29686763780852365, -0.3055774841831569, -0.3180009949335336, -0.3101597899327456, -0.3212194782617802]}}, "s4_arith": {"alone": 0.8801855050550988, "alone_minus_off": {"mean": -0.3536893620391357, "se": 0.004028817126679558, "per_batch": [-0.3418542146862763, -0.3484509532254211, -0.36770071336251886, -0.35094263787112157, -0.33855532946875133, -0.36934916707103904, -0.36129874517547633, -0.3513631354524811]}}, "s5_causal": {"alone": 0.6589420516116804, "alone_minus_off": {"mean": -0.001495630999854905, "se": 0.000777079044912105, "per_batch": [0.0004168429500286397, -0.0005169798484936372, -0.005893587471092143, -0.0007514781929449654, -0.0015341712370217087, -0.001300575801487236, -0.003323435740935632, 0.0009383373431074427]}}, "s6_tryfail": {"alone": 0.6732533568015419, "alone_minus_off": {"mean": 0.0004320956613771132, "se": 0.001552406797931269, "per_batch": [-0.001823876012400083, -0.007744722733654008, 0.0013104218082491403, -0.002184394644271337, 0.0010933367071975786, 0.004876267095575981, 0.006227835753043975, 0.001701897317275658]}}, "web": {"alone": 0.9516445652114878, "alone_minus_off": {"mean": 0.0010478043003816634, "se": 0.000252314831290859, "per_batch": [9.347257074376891e-05, 0.0012871181590572434, 0.0019100393204823929, 0.0013596088463959655, 0.0010667223919766933, -7.928400204770814e-05, 0.0010015409672995412, 0.0017432161491454101]}}}, "s3_rules": {"s1_perspective": {"alone": 0.55350150916384, "alone_minus_off": {"mean": -0.00033179215471922097, "se": 0.00018341479131836675, "per_batch": [0.0003735463176740561, -0.00032229548771689487, -0.0008996842423260043, -0.0003734173306858546, -0.0009771194308770825, -0.0004045461904920167, 0.00045029357562154093, -0.0005011144489515118]}}, "s2_concept": {"alone": 0.7242528502377963, "alone_minus_off": {"mean": 0.0030689175410321956, "se": 0.00019465440878027398, "per_batch": [0.002732804322210014, 0.003277301394841614, 0.0030090514551583958, 0.0019815841032413406, 0.0035649423784097634, 0.003502254702170249, 0.0028635111368657817, 0.0036198908353604065]}}, "s3_rules": {"alone": 0.8236434891082838, "alone_minus_off": {"mean": -0.10596964707321978, "se": 0.009346493007512462, "per_batch": [-0.11229495399364031, -0.10826746427573508, -0.09548760547277069, -0.1291266380596071, -0.08108225264517621, -0.11603213699688908, -0.14432027336743114, -0.0611458517745086]}}, "s4_arith": {"alone": 1.2339533341786908, "alone_minus_off": {"mean": 7.84670844561075e-05, "se": 0.00040067832550554967, "per_batch": [0.0005466468557673299, -0.00020706711163875902, -8.02299066273715e-05, 0.0007894003674602246, -7.782214951546074e-05, 0.0008427149892276109, -0.002416270253297048, 0.0012303638842723341]}}, "s5_causal": {"alone": 0.6675733556594478, "alone_minus_off": {"mean": 0.007135673047912558, "se": 0.0005647387818478271, "per_batch": [0.007655162768217094, 0.008555921902111074, 0.003638808926955339, 0.00856619786550028, 0.007419245566896104, 0.007850750037911025, 0.006295510926158543, 0.007103786389551003]}}, "s6_tryfail": {"alone": 0.6738381569335117, "alone_minus_off": {"mean": 0.0010168957933469058, "se": 0.00017347574800339347, "per_batch": [0.0017297155110527207, 0.0012240865175160431, 0.0009007161382312834, 0.0013980039732010008, 0.00020771204657954456, 0.0009364025382853081, 0.0012371142033188454, 0.0005014154185905006]}}, "web": {"alone": 0.9511259085321484, "alone_minus_off": {"mean": 0.0005291476210421558, "se": 0.0001432382824627441, "per_batch": [-0.00010852105269454349, 0.0007161357581927374, 0.0011528857000588921, 0.0004110385355621249, 0.0007747818421370845, 0.0008065986325466179, 0.0002391418760246955, 0.00024111967650963795]}}}, "s4_arith": {"s1_perspective": {"alone": 0.5529388195514218, "alone_minus_off": {"mean": -0.0008944817671374533, "se": 0.00022219662593559158, "per_batch": [-0.0008394043231986226, -0.0012983400270260015, 0.0003624964323560853, -0.0017084756536711598, -0.0014122355375599804, -0.0008876454567656733, -0.0007099873784243149, -0.0006622621928099592]}}, "s2_concept": {"alone": 0.7207903557747224, "alone_minus_off": {"mean": -0.0003935769220416857, "se": 0.00036656930724860673, "per_batch": [-0.000407598882544824, -0.00045631296840176194, 0.001248465058275916, -0.00226862314753129, 0.0005794525464195566, -0.0009882123118576391, -0.0004823253443447806, -0.00037346032634866244]}}, "s3_rules": {"alone": 0.9280330993212799, "alone_minus_off": {"mean": -0.001580036860223663, "se": 0.0002472548294138001, "per_batch": [-0.0006835450458542169, -0.0017735710870226962, -0.0018863057146633055, -0.0008067706151975162, -0.0008262906460703157, -0.0023852273848162797, -0.0022952804584150943, -0.00198330392974988]}}, "s4_arith": {"alone": 1.126693075548818, "alone_minus_off": {"mean": -0.10718179154541654, "se": 0.019733005800031626, "per_batch": [-0.12218541827078422, -0.000754745863311479, -0.15389351564505827, -0.15261577054047337, -0.042793669092263675, -0.11302235461546473, -0.12726140021967436, -0.14492745811630225]}}, "s5_causal": {"alone": 0.6571682010528499, "alone_minus_off": {"mean": -0.0032694815586852416, "se": 0.00036918803111562825, "per_batch": [-0.004515404498403042, -0.0033061944801866527, -0.003756875016772576, -0.0038970838728884516, -0.0014912615656313966, -0.0025041533878878974, -0.002423407533307831, -0.004261472114404086]}}, "s6_tryfail": {"alone": 0.6593261748623824, "alone_minus_off": {"mean": -0.013495086277782314, "se": 0.0007129585002580727, "per_batch": [-0.01444473685464187, -0.015171234567549097, -0.014263424144969727, -0.016344715190048764, -0.014004375277107628, -0.011502156694041865, -0.010623712309097422, -0.011606335184802141]}}, "web": {"alone": 0.9507103339541697, "alone_minus_off": {"mean": 0.00011357304306353067, "se": 0.00010289644243726181, "per_batch": [-0.0003639152892258579, 4.480148054974986e-05, -4.5575402478625904e-05, 0.0002953802028569097, 5.933401454827347e-06, 0.0003312385855618327, 3.989997500020159e-05, 0.0006008213907892079]}}}, "s5_causal": {"s1_perspective": {"alone": 0.5355407643694168, "alone_minus_off": {"mean": -0.0182925369491425, "se": 0.0010792101601506997, "per_batch": [-0.017861086222706102, -0.017265596294093633, -0.023324072131700513, -0.01829319263299889, -0.012617335197899093, -0.02079786196883937, -0.017867363589462615, -0.018313787555439776]}}, "s2_concept": {"alone": 0.7102577029047991, "alone_minus_off": {"mean": -0.010926229791965086, "se": 0.0004629922489753803, "per_batch": [-0.012461389929229183, -0.010230344990911466, -0.010227593268497648, -0.013156930764981145, -0.009059831069139102, -0.010504141371085018, -0.010754161149776897, -0.011015445792100231]}}, "s3_rules": {"alone": 0.9173749367374315, "alone_minus_off": {"mean": -0.01223819944407202, "se": 0.0003759474969709794, "per_batch": [-0.01179603204870372, -0.013579406146825623, -0.013483955775596801, -0.010735930988785447, -0.01172250946545983, -0.011976441849458785, -0.01136538749094329, -0.013245931786802667]}}, "s4_arith": {"alone": 1.2247629037840526, "alone_minus_off": {"mean": -0.009111963310181992, "se": 0.0003133156648215143, "per_batch": [-0.00816814402007826, -0.009321803642065918, -0.009664909030537183, -0.009344677334630846, -0.007534817908275926, -0.010003456878760808, -0.010028738328437425, -0.008829159338669568]}}, "s5_causal": {"alone": 0.09109720145841224, "alone_minus_off": {"mean": -0.5693404811531231, "se": 0.007627785122552896, "per_batch": [-0.5395056362790345, -0.5729805168000802, -0.5723726226227741, -0.5818498663344731, -0.6069291611541324, -0.5535619986868231, -0.5485124375726715, -0.579011609774995]}}, "s6_tryfail": {"alone": 0.6662963307322131, "alone_minus_off": {"mean": -0.00652493040795174, "se": 0.00038463576812808813, "per_batch": [-0.006364690947468166, -0.005736481319529085, -0.006751049972631429, -0.00707196959914147, -0.008866049617279548, -0.006194643101427477, -0.005688197190299338, -0.005526361515837408]}}, "web": {"alone": 0.9508772001211694, "alone_minus_off": {"mean": 0.0002804392100631825, "se": 0.00013042933220109836, "per_batch": [0.0004127583620707753, -5.262669116401497e-05, 0.00043288033222166344, 0.0006003914341621286, 0.0007997193265121272, -0.0001322546585135198, -0.00021454835695133845, 0.00039719393216763876]}}}, "s6_tryfail": {"s1_perspective": {"alone": 0.5518238130302154, "alone_minus_off": {"mean": -0.0020094882883437554, "se": 0.0003256805712077641, "per_batch": [-0.0010533937365346802, -0.002643029378459616, -0.0009457755927573253, -0.001739604513477322, -0.0036941873405331727, -0.0021753225594381975, -0.002429684900064233, -0.0013949082854854966]}}, "s2_concept": {"alone": 0.7223676892826529, "alone_minus_off": {"mean": 0.0011837565858887206, "se": 0.00030040574308138647, "per_batch": [0.0005063599198028568, 0.00019365246487146326, 0.0026010656116491893, 0.0005574817627718165, 0.0017863837945120142, 0.0005542570880681108, 0.0017871577164408903, 0.0014836943289934235]}}, "s3_rules": {"alone": 0.9301771102893455, "alone_minus_off": {"mean": 0.0005639741078418553, "se": 0.0001864979651283735, "per_batch": [0.0006568877349705238, -0.0003355381518332923, 0.0004389857163272781, 0.0014670980031851188, 0.0005839670910047223, 0.0010127198396056025, 0.00033837586557250443, 0.00034929676390238473]}}, "s4_arith": {"alone": 1.2374947041850164, "alone_minus_off": {"mean": 0.00361983709078198, "se": 0.000311374566884852, "per_batch": [0.004592366734664122, 0.0031974154535159283, 0.002824987023072545, 0.004043054147808256, 0.0025683029166598814, 0.005088364699752379, 0.0034874641941959883, 0.0031567415565867396]}}, "s5_causal": {"alone": 0.6506552986815429, "alone_minus_off": {"mean": -0.009782383929992228, "se": 0.00038481175604609925, "per_batch": [-0.008596939764341949, -0.009804085990748002, -0.010320377908638045, -0.011653931383428295, -0.00987790954363088, -0.010220327001498664, -0.008083184590552017, -0.00970231525709997]}}, "s6_tryfail": {"alone": 0.16377145340459293, "alone_minus_off": {"mean": -0.5090498077355718, "se": 0.004677470835012531, "per_batch": [-0.5175160259054989, -0.49329147411171215, -0.5123662633993884, -0.5249514593394076, -0.5133465000158124, -0.5087508763596975, -0.5168361354909756, -0.485339727262082]}}, "web": {"alone": 0.9507240925662387, "alone_minus_off": {"mean": 0.00012733165513255373, "se": 0.00013788984803610982, "per_batch": [-0.0002956381768333127, -0.00018548328895551247, 0.0008409951627191825, 0.0002792998250011891, 0.0002971860206910648, 0.00027371038884815846, -0.00032573514073419574, 0.00013431845032385592]}}}}, "battery": {"s1_perspective": {"acc_off": {"in": 0.55, "transfer": 0.4625}, "acc_all_on": {"in": 0.8875, "transfer": 0.425}, "acc_member_masked": {"in": 0.55, "transfer": 0.425}, "acc_member_alone": {"in": 0.8375, "transfer": 0.4625}, "margin_all_on_minus_off": {"in": {"mean": 2.0089860945940017, "se": 0.1758094229338807, "per_batch": null}, "transfer": {"mean": -0.2828065149486065, "se": 0.1867353549375036, "per_batch": null}}}, "s2_concept": {"acc_off": {"in": 0.2976190476190476, "transfer": 0.6779661016949152}, "acc_all_on": {"in": 0.9761904761904762, "transfer": 0.711864406779661}, "acc_member_masked": {"in": 0.75, "transfer": 0.7203389830508474}, "acc_member_alone": {"in": 0.9642857142857143, "transfer": 0.7203389830508474}, "margin_all_on_minus_off": {"in": {"mean": 4.217070121229405, "se": 0.32258911924235306, "per_batch": null}, "transfer": {"mean": 0.17814018867783626, "se": 0.22260892817968353, "per_batch": null}}}, "s3_rules": {"acc_off": {"in": 0.155, "transfer": 0.07, "held-out": 0.05}, "acc_all_on": {"in": 0.87, "transfer": 0.32, "held-out": 0.475}, "acc_member_masked": {"in": 0.8, "transfer": 0.23, "held-out": 0.425}, "acc_member_alone": {"in": 0.32, "transfer": 0.08, "held-out": 0.10833333333333334}, "margin_all_on_minus_off": {"in": {"mean": 2.798402752056718, "se": 0.12091775875425413, "per_batch": null}, "transfer": {"mean": 1.0553787994384765, "se": 0.1577693968171716, "per_batch": null}, "held-out": {"mean": 1.63976268209517, "se": 0.20477018279855197, "per_batch": null}}}, "s4_arith": {"acc_off": {"in": 0.34375, "transfer": 0.24166666666666667, "held-out": 0.35}, "acc_all_on": {"in": 0.6375, "transfer": 0.24166666666666667, "held-out": 0.325}, "acc_member_masked": {"in": 0.5125, "transfer": 0.24166666666666667, "held-out": 0.2875}, "acc_member_alone": {"in": 0.38125, "transfer": 0.24583333333333332, "held-out": 0.3875}, "margin_all_on_minus_off": {"in": {"mean": 1.6223410779362895, "se": 0.22531791211737523, "per_batch": null}, "transfer": {"mean": -0.393949360648791, "se": 0.07830538002094956, "per_batch": null}, "held-out": {"mean": -0.13549973964691162, "se": 0.11009789062608755, "per_batch": null}}}}, "probes_all_on": {"p0_grounding": {"acc": 0.4666666666666667, "n": 30, "families": {"count": 0.125, "permanence": 0.7142857142857143, "sequence": 0.625, "attribute": 0.42857142857142855}}, "p1_self_other": {"acc": 0.5333333333333333, "n": 30, "families": {"belief": 0.25, "reality": 0.5, "perspective": 0.8, "reference": 1.0, "second_order": 0.0}}, "p2_difference": {"acc": 0.5666666666666667, "n": 30, "families": {"category": 1.0, "antonym": 0.6666666666666666, "odd_one_out": 0.3333333333333333, "property": 0.5, "analogy": 0.3333333333333333}}, "p3_logic": {"acc": 0.3, "n": 30, "families": {"d1": 0.5, "d2": 0.5, "d3": 0.25, "d4": 0.25, "d5": 0.0, "d6": 0.0, "negation": 0.5, "contrapositive": 0.25}}, "p4_arithmetic": {"acc": 0.3333333333333333, "n": 30, "families": {"add2": 0.4, "add3": 0.0, "sub2": 0.75, "sub3": 0.4, "mul1": 0.0, "mul2": 0.3333333333333333, "word": 0.4}}, "p5_causal": {"acc": 0.5, "n": 30, "families": {"effect": 0.5714285714285714, "cause": 0.42857142857142855, "counterfactual": 0.5, "prevention": 0.4, "chain": 0.6}}, "p6_learning": {"acc": 0.36666666666666664, "n": 30, "families": {"error_detect": 0.25, "correction": 0.25, "strategy": 0.7142857142857143, "check": 0.2857142857142857}}, "p7_composite": {"acc": 0.36666666666666664, "n": 30, "families": {"logic_arith": 0.375, "causal_category": 0.375, "belief_causal": 0.2857142857142857, "three_step": 0.42857142857142855}}, "p8_articulation": {"acc": 0.4666666666666667, "n": 30, "families": {"definition": 0.375, "register": 0.42857142857142855, "grammar": 0.875, "paraphrase": 0.14285714285714285}}}, "probes_mean_off": 0.41111111111111115, "probes_mean_all_on": 0.43333333333333335, "gates": {"s1_perspective": {"mean": 0.2099039478571495, "max": 0.5712027580899203}, "s2_concept": {"mean": 0.27674460361666475, "max": 0.4977012170865557}, "s3_rules": {"mean": 0.12033936123137423, "max": 0.1738736117222015}, "s4_arith": {"mean": 0.10767731599767934, "max": 0.15932411779931022}, "s5_causal": {"mean": 0.18518384792916404, "max": 0.3681551185947585}, "s6_tryfail": {"mean": 0.23037076111374397, "max": 0.5373235543789295}}, "chain_proof": "compiled chain: 18 cache entries (limit 256), 0 limit warnings", "curve": [{"step": 1, "train_bpb": {"s6_tryfail": 0.5002930598109879}, "quiet_kl": {"s6_tryfail": 0.00012237469491083175}}, {"step": 50, "train_bpb": {"s6_tryfail": 0.47839155593451116}, "quiet_kl": {"s6_tryfail": 0.00018462371721398086}}, {"step": 100, "train_bpb": {"s6_tryfail": 0.44747906750139327}, "quiet_kl": {"s6_tryfail": 0.0004046990547794849}}, {"step": 150, "train_bpb": {"s6_tryfail": 0.4024823247291481}, "quiet_kl": {"s6_tryfail": 0.0005105614266358316}}, {"step": 200, "train_bpb": {"s6_tryfail": 0.3530191907941692}, "quiet_kl": {"s6_tryfail": 0.0006100823951419443}}, {"step": 250, "train_bpb": {"s6_tryfail": 0.31166694011557694}, "quiet_kl": {"s6_tryfail": 0.0007389484351733699}}, {"step": 300, "train_bpb": {"s6_tryfail": 0.2727909356300738}, "quiet_kl": {"s6_tryfail": 0.0007169208495179191}}, {"step": 350, "train_bpb": {"s6_tryfail": 0.2477709240900802}, "quiet_kl": {"s6_tryfail": 0.0006833641446428373}}, {"step": 400, "train_bpb": {"s6_tryfail": 0.2153718111735548}, "quiet_kl": {"s6_tryfail": 0.0007316118670860305}}, {"step": 450, "train_bpb": {"s6_tryfail": 0.19394364869769834}, "quiet_kl": {"s6_tryfail": 0.0007027628214564174}}, {"step": 500, "train_bpb": {"s6_tryfail": 0.17313787939652886}, "quiet_kl": {"s6_tryfail": 0.0006750317133264616}}, {"step": 550, "train_bpb": {"s6_tryfail": 0.1560838669084423}, "quiet_kl": {"s6_tryfail": 0.0007547189411707222}}, {"step": 600, "train_bpb": {"s6_tryfail": 0.1364431856913887}, "quiet_kl": {"s6_tryfail": 0.0006635376933263615}}, {"step": 650, "train_bpb": {"s6_tryfail": 0.12259003045961842}, "quiet_kl": {"s6_tryfail": 0.0006801160908071325}}, {"step": 700, "train_bpb": {"s6_tryfail": 0.1138594775887995}, "quiet_kl": {"s6_tryfail": 0.0006852408364647999}}, {"step": 750, "train_bpb": {"s6_tryfail": 0.1028377311047958}, "quiet_kl": {"s6_tryfail": 0.0007006171671673655}}, {"step": 800, "train_bpb": {"s6_tryfail": 0.0953177240199447}, "quiet_kl": {"s6_tryfail": 0.0007095109054353088}}, {"step": 850, "train_bpb": {"s6_tryfail": 0.09068869563610697}, "quiet_kl": {"s6_tryfail": 0.0006043417408363894}}, {"step": 900, "train_bpb": {"s6_tryfail": 0.09183736475152328}, "quiet_kl": {"s6_tryfail": 0.000590550106135197}}, {"step": 950, "train_bpb": {"s6_tryfail": 0.09079498338659081}, "quiet_kl": {"s6_tryfail": 0.0005882581893820316}}, {"step": 1000, "train_bpb": {"s6_tryfail": 0.08567725866859625}, "quiet_kl": {"s6_tryfail": 0.000641582629759796}}, {"step": 1050, "train_bpb": {"s6_tryfail": 0.08564678839485372}, "quiet_kl": {"s6_tryfail": 0.0005746858497150243}}, {"step": 1100, "train_bpb": {"s6_tryfail": 0.08149941741774047}, "quiet_kl": {"s6_tryfail": 0.0005571391090052202}}, {"step": 1150, "train_bpb": {"s6_tryfail": 0.08134344538427057}, "quiet_kl": {"s6_tryfail": 0.0006022596196271479}}, {"step": 1200, "train_bpb": {"s6_tryfail": 0.08207715518964173}, "quiet_kl": {"s6_tryfail": 0.0006434066989459097}}, {"step": 1250, "train_bpb": {"s6_tryfail": 0.07932030135098084}, "quiet_kl": {"s6_tryfail": 0.0005737369495909661}}, {"step": 1300, "train_bpb": {"s6_tryfail": 0.07830241923713122}, "quiet_kl": {"s6_tryfail": 0.0005626186321023852}}, {"step": 1350, "train_bpb": {"s6_tryfail": 0.07677186791244274}, "quiet_kl": {"s6_tryfail": 0.0005909824330592528}}, {"step": 1400, "train_bpb": {"s6_tryfail": 0.07580492963951845}, "quiet_kl": {"s6_tryfail": 0.000562069175648503}}, {"step": 1450, "train_bpb": {"s6_tryfail": 0.07456815327461934}, "quiet_kl": {"s6_tryfail": 0.0005464612611103803}}, {"step": 1500, "train_bpb": {"s6_tryfail": 0.07668068905281551}, "quiet_kl": {"s6_tryfail": 0.0005684417032171041}}, {"step": 1550, "train_bpb": {"s6_tryfail": 0.07425913946958279}, "quiet_kl": {"s6_tryfail": 0.0005336960422573611}}, {"step": 1600, "train_bpb": {"s6_tryfail": 0.07372694979791433}, "quiet_kl": {"s6_tryfail": 0.00048604847921524197}}, {"step": 1650, "train_bpb": {"s6_tryfail": 0.06763844879916321}, "quiet_kl": {"s6_tryfail": 0.0005251623078947887}}, {"step": 1700, "train_bpb": {"s6_tryfail": 0.06929512166605713}, "quiet_kl": {"s6_tryfail": 0.0005285109201213345}}, {"step": 1750, "train_bpb": {"s6_tryfail": 0.07033784699629134}, "quiet_kl": {"s6_tryfail": 0.0005398944951593876}}, {"step": 1800, "train_bpb": {"s6_tryfail": 0.07151695016160432}, "quiet_kl": {"s6_tryfail": 0.0005307430145330727}}, {"step": 1850, "train_bpb": {"s6_tryfail": 0.06844207836622096}, "quiet_kl": {"s6_tryfail": 0.000534223651047796}}, {"step": 1900, "train_bpb": {"s6_tryfail": 0.06820508734822357}, "quiet_kl": {"s6_tryfail": 0.0005472498812014237}}, {"step": 1950, "train_bpb": {"s6_tryfail": 0.07023501179752359}, "quiet_kl": {"s6_tryfail": 0.000540313171222806}}, {"step": 2000, "train_bpb": {"s6_tryfail": 0.06963154719462433}, "quiet_kl": {"s6_tryfail": 0.00048818277369719}}, {"step": 2050, "train_bpb": {"s6_tryfail": 0.06743725607743208}, "quiet_kl": {"s6_tryfail": 0.0004848940373631194}}, {"step": 2100, "train_bpb": {"s6_tryfail": 0.06917213794731669}, "quiet_kl": {"s6_tryfail": 0.000477516264654696}}, {"step": 2150, "train_bpb": {"s6_tryfail": 0.06847405983001877}, "quiet_kl": {"s6_tryfail": 0.0005366378434700891}}, {"step": 2200, "train_bpb": {"s6_tryfail": 0.06547037595339378}, "quiet_kl": {"s6_tryfail": 0.0004501626710407436}}, {"step": 2250, "train_bpb": {"s6_tryfail": 0.06596386877619399}, "quiet_kl": {"s6_tryfail": 0.0004669543949421495}}, {"step": 2300, "train_bpb": {"s6_tryfail": 0.06444915909594394}, "quiet_kl": {"s6_tryfail": 0.000513481542293448}}, {"step": 2350, "train_bpb": {"s6_tryfail": 0.06580571052076416}, "quiet_kl": {"s6_tryfail": 0.00046714137453818694}}, {"step": 2400, "train_bpb": {"s6_tryfail": 0.061946769658863504}, "quiet_kl": {"s6_tryfail": 0.0004179173830198124}}, {"step": 2450, "train_bpb": {"s6_tryfail": 0.06244454995540515}, "quiet_kl": {"s6_tryfail": 0.0004658200714038685}}, {"step": 2500, "train_bpb": {"s6_tryfail": 0.06399826809558898}, "quiet_kl": {"s6_tryfail": 0.000493889493518509}}, {"step": 2550, "train_bpb": {"s6_tryfail": 0.06387798525689117}, "quiet_kl": {"s6_tryfail": 0.00040577747917268424}}, {"step": 2600, "train_bpb": {"s6_tryfail": 0.0631643421722444}, "quiet_kl": {"s6_tryfail": 0.00045037508418317886}}, {"step": 2650, "train_bpb": {"s6_tryfail": 0.06319595817290911}, "quiet_kl": {"s6_tryfail": 0.0004189631945337169}}, {"step": 2700, "train_bpb": {"s6_tryfail": 0.06367222962045435}, "quiet_kl": {"s6_tryfail": 0.000452824610110838}}, {"step": 2750, "train_bpb": {"s6_tryfail": 0.06232281128356781}, "quiet_kl": {"s6_tryfail": 0.00041039727453608067}}, {"step": 2800, "train_bpb": {"s6_tryfail": 0.062274626044362326}, "quiet_kl": {"s6_tryfail": 0.0004436369045288302}}, {"step": 2850, "train_bpb": {"s6_tryfail": 0.060220913384751586}, "quiet_kl": {"s6_tryfail": 0.0004329192460863851}}, {"step": 2900, "train_bpb": {"s6_tryfail": 0.06334506648627224}, "quiet_kl": {"s6_tryfail": 0.00042569231474772097}}, {"step": 2950, "train_bpb": {"s6_tryfail": 0.06271330375051087}, "quiet_kl": {"s6_tryfail": 0.0004766579263377935}}, {"step": 3000, "train_bpb": {"s6_tryfail": 0.059077178774056015}, "quiet_kl": {"s6_tryfail": 0.0004107370838755742}}, {"step": 3050, "train_bpb": {"s6_tryfail": 0.061683986133993327}, "quiet_kl": {"s6_tryfail": 0.0004684449202613905}}, {"step": 3100, "train_bpb": {"s6_tryfail": 0.06074693586604925}, "quiet_kl": {"s6_tryfail": 0.00042863285896601157}}, {"step": 3150, "train_bpb": {"s6_tryfail": 0.061948542208803546}, "quiet_kl": {"s6_tryfail": 0.0004699334979522973}}, {"step": 3200, "train_bpb": {"s6_tryfail": 0.06080482098050797}, "quiet_kl": {"s6_tryfail": 0.00042934722674544903}}, {"step": 3250, "train_bpb": {"s6_tryfail": 0.05876442692610295}, "quiet_kl": {"s6_tryfail": 0.00042889808100881054}}, {"step": 3300, "train_bpb": {"s6_tryfail": 0.06250578776415329}, "quiet_kl": {"s6_tryfail": 0.00041181570268236103}}, {"step": 3350, "train_bpb": {"s6_tryfail": 0.061606059128805464}, "quiet_kl": {"s6_tryfail": 0.000444968756637536}}, {"step": 3400, "train_bpb": {"s6_tryfail": 0.06231180213664031}, "quiet_kl": {"s6_tryfail": 0.00038978537428192796}}, {"step": 3450, "train_bpb": {"s6_tryfail": 0.05854130222430969}, "quiet_kl": {"s6_tryfail": 0.00042884805734502154}}, {"step": 3500, "train_bpb": {"s6_tryfail": 0.06241828739844973}, "quiet_kl": {"s6_tryfail": 0.0004357010981766507}}, {"step": 3550, "train_bpb": {"s6_tryfail": 0.06092014658052837}, "quiet_kl": {"s6_tryfail": 0.00042723089864011855}}, {"step": 3600, "train_bpb": {"s6_tryfail": 0.06102536115927082}, "quiet_kl": {"s6_tryfail": 0.00038741940719773995}}, {"step": 3650, "train_bpb": {"s6_tryfail": 0.058422374501411564}, "quiet_kl": {"s6_tryfail": 0.00038738875213311985}}, {"step": 3700, "train_bpb": {"s6_tryfail": 0.059848828597142446}, "quiet_kl": {"s6_tryfail": 0.00045668031787499786}}, {"step": 3750, "train_bpb": {"s6_tryfail": 0.0603000119680039}, "quiet_kl": {"s6_tryfail": 0.00038121772813610734}}, {"step": 3800, "train_bpb": {"s6_tryfail": 0.05960034327990838}, "quiet_kl": {"s6_tryfail": 0.0004771452586282976}}, {"step": 3850, "train_bpb": {"s6_tryfail": 0.06164454922228062}, "quiet_kl": {"s6_tryfail": 0.00040386825829045846}}, {"step": 3900, "train_bpb": {"s6_tryfail": 0.05932488721340938}, "quiet_kl": {"s6_tryfail": 0.0004747083969414234}}, {"step": 3950, "train_bpb": {"s6_tryfail": 0.05927295404228496}, "quiet_kl": {"s6_tryfail": 0.0004449613447650336}}, {"step": 4000, "train_bpb": {"s6_tryfail": 0.05889391658729179}, "quiet_kl": {"s6_tryfail": 0.0004013206050149165}}], "steps_done": {"s1_perspective": 3200, "s2_concept": 2400, "s3_rules": 800, "s4_arith": 800, "s5_causal": 4000, "s6_tryfail": 4000, "s7_mixed": 0, "s8_register": 0}, "stage_seconds": 2877.2282288074493, "inert_at_birth": true}, {"close": 7, "attached": ["s1_perspective", "s2_concept", "s3_rules", "s4_arith", "s5_causal", "s6_tryfail", "s7_mixed"], "rows": {"s1_perspective": {"off": 0.5538333013185592, "all_on": 0.04849286779776042, "all_on_minus_off": {"mean": -0.5053404335207988, "se": 0.00483881029631658, "per_batch": [-0.5060997530477316, -0.5249557696545948, -0.5160955244517819, -0.5115133745614706, -0.4932762052769902, -0.5072051607872367, -0.48048044882773655, -0.5030972315588481]}}, "s2_concept": {"off": 0.7211839326967642, "all_on": 0.11124442349951956, "all_on_minus_off": {"mean": -0.6099395091972446, "se": 0.003811817979267732, "per_batch": [-0.6102310345702161, -0.6090908755863211, -0.5958348075720332, -0.6325691624112671, -0.6079126546920071, -0.6079173304703274, -0.6014244694523038, -0.6145357388234809]}}, "s3_rules": {"off": 0.9296131361815035, "all_on": 0.22322506441894902, "all_on_minus_off": {"mean": -0.7063880717625545, "se": 0.005409308242481403, "per_batch": [-0.6852324536331733, -0.7064374630300991, -0.7214196456686873, -0.6896619098018077, -0.6943694189212245, -0.7274904397636622, -0.7123858699711538, -0.7141073733106278]}}, "s4_arith": {"off": 1.2338748670942346, "all_on": 0.32787814936583803, "all_on_minus_off": {"mean": -0.9059967177283965, "se": 0.006200536389578348, "per_batch": [-0.9218238269456134, -0.8980296617356107, -0.9206243769428333, -0.8798600171394254, -0.8831496797872977, -0.9259406186549792, -0.9086669391762949, -0.9098786214451176]}}, "s5_causal": {"off": 0.6604376826115352, "all_on": 0.04288413431794293, "all_on_minus_off": {"mean": -0.6175535482935923, "se": 0.0069105663783547416, "per_batch": [-0.5835154054444648, -0.6267226336930174, -0.6282974949503416, -0.6290403525129121, -0.6433255029036979, -0.6158088135242036, -0.594852738265486, -0.6188654450546148]}}, "s6_tryfail": {"off": 0.6728212611401647, "all_on": 0.05458588796313044, "all_on_minus_off": {"mean": -0.6182353731770343, "se": 0.0035839227605414317, "per_batch": [-0.6282119389559953, -0.623646025302817, -0.6224395401347251, -0.6266066743906732, -0.6221839792900008, -0.6171622524929998, -0.6051066932496862, -0.600525881599377]}}, "s7_mixed": {"off": 0.7829759286650222, "all_on": 0.7187657867576811, "all_on_minus_off": {"mean": -0.06421014190734109, "se": 0.03917954272712242, "per_batch": [-0.0001537524898713727, 0.0009671444371270876, -0.0343675510958914, 0.0013758612069024734, -3.8524113793236836e-05, -0.24151424770639973, 0.0038079538640788946, -0.24375801936088137]}}, "s8_register": {"off": 0.4906504977761471, "all_on": 0.37574392004843205, "all_on_minus_off": {"mean": -0.11490657772771504, "se": 0.0020452969828823253, "per_batch": [-0.10449659417074847, -0.11788297128981196, -0.10957294158277192, -0.11922555385377248, -0.11240721566899098, -0.1216326230352483, -0.11920096033469907, -0.11483376188567707]}}, "web": {"off": 0.9505967609111062, "all_on": 0.9520328482925577, "all_on_minus_off": {"mean": 0.0014360873814515118, "se": 0.0003002439726702609, "per_batch": [0.0004877427978469129, 0.0003888527736007896, 0.0012231406129362687, 0.0009423789354028322, 0.002805552983525139, 0.0015689977238214636, 0.00230998497506385, 0.0017620482494148382]}}}, "member_in_group": {"s1_perspective": {"s1_perspective": {"masked": 0.4462977145880287, "all_on_minus_masked": {"mean": -0.3978048467902683, "se": 0.005724543280066031, "per_batch": [-0.4002544924267484, -0.4214188171778295, -0.4142323071332134, -0.40115675103437953, -0.3837858153666324, -0.39609611316262083, -0.3697566853367162, -0.39573779268400605]}}, "s2_concept": {"masked": 0.3338837351638468, "all_on_minus_masked": {"mean": -0.2226393116643272, "se": 0.0013820222080270902, "per_batch": [-0.21836826360966766, -0.2208292923478517, -0.2168886216218851, -0.22769680682186427, -0.22303566852468631, -0.2223435780908674, -0.224513397205478, -0.2274388650923171]}}, "s3_rules": {"masked": 0.4053504727332594, "all_on_minus_masked": {"mean": -0.18212540831431037, "se": 0.0028038956762696725, "per_batch": [-0.17880352579169034, -0.18204262554020914, -0.18215166254085616, -0.17408200008622823, -0.18062632691252253, -0.19974262851255087, -0.18429843598025114, -0.17525606115017459]}}, "s4_arith": {"masked": 0.4546856793081988, "all_on_minus_masked": {"mean": -0.12680752994236077, "se": 0.00514065374958129, "per_batch": [-0.13742634880759536, -0.14704703679902953, -0.09934964788054584, -0.12844494182724964, -0.13693217815817244, -0.12167011525313642, -0.12508951731325352, -0.11850045349990351]}}, "s5_causal": {"masked": 0.05776489522673896, "all_on_minus_masked": {"mean": -0.014880760908796028, "se": 0.0006072445257869698, "per_batch": [-0.013907000849972262, -0.0163959982667529, -0.014292043132964893, -0.017021160577097044, -0.015515089692888992, -0.016481221044713265, -0.012444476510408455, -0.012989097195570412]}}, "s6_tryfail": {"masked": 0.08900458441524381, "all_on_minus_masked": {"mean": -0.03441869645211336, "se": 0.002640085088238837, "per_batch": [-0.043791146969436714, -0.0358554267530839, -0.03386891427200383, -0.031799054698749364, -0.038883600529235125, -0.026411278952760416, -0.022350016141791967, -0.04239013329984556]}}, "s7_mixed": {"masked": 0.7225226778978273, "all_on_minus_masked": {"mean": -0.003756891140146136, "se": 0.003227249179183778, "per_batch": [-0.00036830084682271114, -0.0007640329264579915, -0.0023596879611631483, 0.0006994534410590392, -0.00021343046972077673, -0.025742535163796043, 0.002938882533606413, -0.00424547772787387]}}, "web": {"masked": 0.9517336629735504, "all_on_minus_masked": {"mean": 0.0002991853190072169, "se": 0.00031906115825222153, "per_batch": [0.00021755805334144895, -0.0012232266042617734, -0.00029761597731814415, -0.0002883289141714096, 0.0013888458970425788, 0.0012365552597036755, 0.001079363116815002, 0.00028033172090635716]}}}, "s2_concept": {"s1_perspective": {"masked": 0.053383734608057246, "all_on_minus_masked": {"mean": -0.004890866810296829, "se": 0.0004628020995434032, "per_batch": [-0.004141266990366682, -0.005270580318341635, -0.006022149877069942, -0.003440582798462398, -0.006882116875962387, -0.005085054033723377, -0.0029630622194261325, -0.005322121369022083]}}, "s2_concept": {"masked": 0.23621946620440698, "all_on_minus_masked": {"mean": -0.1249750427048874, "se": 0.0012153532520739559, "per_batch": [-0.12898746109939635, -0.12571482473395934, -0.12004164377886353, -0.12944111908562542, -0.12654355538388878, -0.12526203740990027, -0.12146276866968994, -0.12234693147777567]}}, "s3_rules": {"masked": 0.3310471446862988, "all_on_minus_masked": {"mean": -0.10782208026734982, "se": 0.001322887591332039, "per_batch": [-0.10652568848129063, -0.1066150764640765, -0.10869454019349009, -0.10826626039717899, -0.10090856312146038, -0.10650287928221991, -0.11247916891054571, -0.11258446528853647]}}, "s4_arith": {"masked": 0.4342416340223047, "all_on_minus_masked": {"mean": -0.10636348465646667, "se": 0.003970104819277906, "per_batch": [-0.12350663199037493, -0.10730545782030254, -0.09177879960691415, -0.10175039669960245, -0.09154053914197507, -0.11418715011409564, -0.10594410764956663, -0.11489479422890198]}}, "s5_causal": {"masked": 0.044248592268924525, "all_on_minus_masked": {"mean": -0.0013644579509815898, "se": 0.000265537739003738, "per_batch": [-0.001256306392263623, -0.0013925972686145566, -0.0027939871502544733, -0.0018342863372351262, -0.001475675637896967, -0.0004494578474275174, -0.0012481587141789999, -0.00046519425998145514]}}, "s6_tryfail": {"masked": 0.05812095843306121, "all_on_minus_masked": {"mean": -0.0035350704699307614, "se": 0.0005574443481309332, "per_batch": [-0.003722800954070425, -0.006457588451223212, -0.0034597373662022465, -0.005021216223506236, -0.0026094927615413904, -0.001563236305017493, -0.0021997602192342716, -0.003246731478650816]}}, "s7_mixed": {"masked": 0.7196175523339106, "all_on_minus_masked": {"mean": -0.0008517655762294157, "se": 0.0003050649636828605, "per_batch": [-0.0006409793397657015, -0.0008830449208550029, -7.936999337321282e-05, -0.0012841944539926375, -0.00046942664553006885, -0.000577775715573603, -0.00012245164741431225, -0.002756881893330787]}}, "web": {"masked": 0.9522079911246301, "all_on_minus_masked": {"mean": -0.00017514283207238635, "se": 0.0003266070591697204, "per_batch": [-0.0011515098388519274, 0.0012522056809320947, 0.000547936725648901, 0.0007095144261345387, -0.0006568877349704128, -0.0012606328308243375, -0.0006914562477938757, -0.00015031283685407182]}}}, "s3_rules": {"s1_perspective": {"masked": 0.04846420413458449, "all_on_minus_masked": {"mean": 2.8663663175929922e-05, "se": 8.084781906304394e-05, "per_batch": [0.000334318149903777, -9.864817364333561e-05, 6.94433697437033e-05, 6.734195672846582e-05, -0.0004422802589829078, 2.2422238106237224e-05, 7.772272204539105e-05, 0.00019898930150610838]}}, "s2_concept": {"masked": 0.1105438696382179, "all_on_minus_masked": {"mean": 0.0007005538613016719, "se": 6.484400365846034e-05, "per_batch": [0.0006020467671766228, 0.0005459051805855875, 0.0008773802422922944, 0.00045893570382740667, 0.0008218728417263149, 0.0005370373251504801, 0.0008263551395644442, 0.000934897690090225]}}, "s3_rules": {"masked": 0.31442722232089365, "all_on_minus_masked": {"mean": -0.09120215790194464, "se": 0.001226366671635572, "per_batch": [-0.08595576801833843, -0.08849954190941886, -0.09337645393993552, -0.09358874502459433, -0.0881194817488434, -0.09464587938378527, -0.0952678331427993, -0.09016356004784201]}}, "s4_arith": {"masked": 0.32800237189711046, "all_on_minus_masked": {"mean": -0.00012422253127241517, "se": 8.961084918552508e-05, "per_batch": [-3.0526920528073376e-06, -0.0003668174964590043, -9.781513267820285e-05, -0.0005460019208267108, 0.00021022729284841946, -0.0002834704042846137, 1.4983988456451769e-05, 7.816611481714641e-05]}}, "s5_causal": {"masked": 0.042906928401350936, "all_on_minus_masked": {"mean": -2.279408340800579e-05, "se": 8.710054722417997e-05, "per_batch": [9.582658327761395e-06, 9.297005893594418e-05, 3.5901378367612924e-05, 0.00022301312804851037, 5.4730791408171e-05, -0.00012942231923210312, -0.0005805811825658064, 0.00011145281944586333]}}, "s6_tryfail": {"masked": 0.05475724717688387, "all_on_minus_masked": {"mean": -0.00017135921375342842, "se": 8.452365759652707e-05, "per_batch": [-0.00026178984136022987, 0.00030805854890017526, -0.00010966581221424215, -0.00047982084699155214, -4.342561934281286e-06, -0.00030018496816534845, -0.000295767163821313, -0.0002273610644406357]}}, "s7_mixed": {"masked": 0.7186833586977973, "all_on_minus_masked": {"mean": 8.242805988383645e-05, "se": 0.0001894956827222854, "per_batch": [0.0001938244475223616, 0.00030604312721038696, 0.0006414092963927809, -0.0001618356744619298, -0.0006678946246256867, 0.0008368675790983993, 9.493442327629387e-05, -0.0005839240953419145]}}, "web": {"masked": 0.9516742752144244, "all_on_minus_masked": {"mean": 0.0003585730781333063, "se": 0.0001587293052135205, "per_batch": [-0.00027543021535669787, 0.00045953764310546763, 0.00037148252586360897, 0.0011253684759208182, 0.00044405920452783576, 0.0004990076614784789, -0.0002464511386863766, 0.0004910104682133154]}}}, "s4_arith": {"s1_perspective": {"masked": 0.048525698345280005, "all_on_minus_masked": {"mean": -3.283054751958888e-05, "se": 4.441886936057066e-05, "per_batch": [-2.1336597622670372e-05, -9.071009941444969e-05, -0.00011308396740013216, 0.0001394403086448573, -0.0001577242142147134, 0.000165253829647799, -2.517933497788588e-05, -0.00015930430481951585]}}, "s2_concept": {"masked": 0.11186845179925958, "all_on_minus_masked": {"mean": -0.0006240282997400197, "se": 6.694477333874955e-05, "per_batch": [-0.0005954899286124693, -0.0005166573810232306, -0.0002888233642927285, -0.0006598651846135561, -0.0007188552338595056, -0.0006039170785047537, -0.0009633393209766555, -0.0006452789060372582]}}, "s3_rules": {"masked": 0.22324770700982668, "all_on_minus_masked": {"mean": -2.2642590877655633e-05, "se": 8.90060517258909e-05, "per_batch": [0.00018625721088441627, 0.00018283905569851933, -0.00016245911157128368, 0.00030273246118125985, -0.00019066426631275668, -0.0004651270792584705, 4.697276151691132e-05, -8.169175915984095e-05]}}, "s4_arith": {"masked": 0.4726139140071512, "all_on_minus_masked": {"mean": -0.14473576464131316, "se": 0.0023425873442205936, "per_batch": [-0.15399163174737562, -0.13531541225307386, -0.14443365293001237, -0.14222230000521835, -0.13678994850590892, -0.1520042432296677, -0.1451798856521061, -0.14794904280714222]}}, "s5_causal": {"masked": 0.0429240272389672, "all_on_minus_masked": {"mean": -3.989292102425917e-05, "se": 7.008635999887405e-05, "per_batch": [5.680801936309987e-06, 0.00031679473011821113, -9.823434038970513e-05, -9.448296881775509e-05, -0.0003071233932360959, -0.0001970948051177035, -0.00010299342280654933, 0.00015831003011921446]}}, "s6_tryfail": {"masked": 0.05863153864588247, "all_on_minus_masked": {"mean": -0.004045650682752025, "se": 0.000427159480028995, "per_batch": [-0.0050108058986712, -0.006080328383182121, -0.003742482218678546, -0.002907549444402316, -0.0027804918866195794, -0.0027699794470855896, -0.004380193888077838, -0.004693374295299008]}}, "s7_mixed": {"masked": 0.7187640964906906, "all_on_minus_masked": {"mean": 1.6902669905186074e-06, "se": 0.00011781173156115497, "per_batch": [0.00016527532747923068, 0.00022469533335223169, -0.00023570222300750565, 0.0003580678790963132, 0.0002331224832445855, -0.0005229992412738027, 0.00018363447545877154, -0.0003925718984256754]}}, "web": {"masked": 0.9517864294006183, "all_on_minus_masked": {"mean": 0.0002464188919393817, "se": 0.00015177058894160433, "per_batch": [-0.0004598816084071533, 0.0006576616568993998, 0.0004058790560361736, 0.0008266346113721124, 0.0002866950789883749, -3.147282510784777e-05, -0.00013956392117508987, 0.000425399086909084]}}}, "s5_causal": {"s1_perspective": {"masked": 0.04831749991026598, "all_on_minus_masked": {"mean": 0.0001753678874944391, "se": 0.00013999636986397027, "per_batch": [0.00020691662681931317, -0.0003441104120872826, 0.00034729746558608693, -0.00036309299717626753, 0.00041624101075071057, 0.00045059991971839464, 0.0007421481341357353, -5.305664779117758e-05]}}, "s2_concept": {"masked": 0.11197586705725354, "all_on_minus_masked": {"mean": -0.000731443557733984, "se": 0.0001664510238279948, "per_batch": [-0.001157819452355463, -0.0003865955013082345, -0.000452303622853506, -0.0002215244032269803, -0.0009973918858475067, -0.00026410085823119767, -0.0008613536090150142, -0.0015104591290339697]}}, "s3_rules": {"masked": 0.22299590828559, "all_on_minus_masked": {"mean": 0.000229156133359016, "se": 0.0001624295431140461, "per_batch": [0.0003984193085550536, 0.0009203436582609636, -0.0004669543949238908, -0.0001915241795670819, 0.0005110894427015622, -1.0598430859404262e-05, 0.0006097644886340903, 6.27091740708352e-05]}}, "s4_arith": {"masked": 0.3277956836847493, "all_on_minus_masked": {"mean": 8.246568108872393e-05, "se": 0.00010873900516492195, "per_batch": [0.00023942134783239144, -0.00018234460557725596, -0.00044367224356334223, 0.0001918681448688231, 0.000262445525216648, -0.00015048481950497017, 0.0002590918635248518, 0.00048340023591264547]}}, "s5_causal": {"masked": 0.4558421121504317, "all_on_minus_masked": {"mean": -0.4129579778324888, "se": 0.005436268842770763, "per_batch": [-0.3877692667770011, -0.4153893204168929, -0.41597012195223015, -0.4138392354102086, -0.43088242857721853, -0.4315495250948551, -0.3945923559825127, -0.4136715684489909]}}, "s6_tryfail": {"masked": 0.05546645928976492, "all_on_minus_masked": {"mean": -0.0008805713266344765, "se": 0.00019744654645023388, "per_batch": [-0.0006766334930726828, -0.00035400478896972787, -0.001935659361003074, -0.0009153077912653837, -0.0014516948070172897, -0.00045348062912034526, -0.0003895030829993845, -0.0008682866596279246]}}, "s7_mixed": {"masked": 0.7188670469178344, "all_on_minus_masked": {"mean": -0.00010126016015330613, "se": 0.00019064778884214122, "per_batch": [-0.0002490308784492967, 0.000549226595530472, -0.0008797772504887114, 0.00045033657128423776, 0.00013715616406306808, -0.000816917591598354, -0.00021360245237161957, 0.00021252756080375468]}}, "web": {"masked": 0.9517240211961865, "all_on_minus_masked": {"mean": 0.00030882709637124195, "se": 0.00014462833718290952, "per_batch": [-6.406353744625726e-05, -3.869609644380212e-06, 0.0004199816334069517, 0.0009594052178382162, 0.0005918782929443811, -8.33255943429867e-05, -3.8352131142338486e-05, 0.0006889624993563492]}}}, "s6_tryfail": {"s1_perspective": {"masked": 0.04848232344738301, "all_on_minus_masked": {"mean": 1.0544350377410597e-05, "se": 6.43439620703554e-05, "per_batch": [0.0001103806151068823, -8.921600013478415e-07, -0.00026047309918956346, 8.529802037011286e-05, -2.2723207745246887e-05, -0.00021439787213185096, 0.00030312748383246046, 8.403502277783831e-05]}}, "s2_concept": {"masked": 0.11088773282524456, "all_on_minus_masked": {"mean": 0.00035669067427500367, "se": 0.00012021640540750844, "per_batch": [0.00020496032416575383, 0.0010448375996542775, 0.0005773672567778027, 0.00018096874437038835, 0.0002644878191956357, 0.00011479841945091773, -2.731299474013904e-05, 0.0004934182253253927]}}, "s3_rules": {"masked": 0.22318794303865186, "all_on_minus_masked": {"mean": 3.7121380297166356e-05, "se": 5.294052418851752e-05, "per_batch": [0.0002621875512403282, 3.2139257880009486e-05, -0.00023574521867020248, 0.00015949241084389776, 0.00011357304306353067, -6.184926081653774e-05, 6.771816877720882e-06, 2.0401441958584066e-05]}}, "s4_arith": {"masked": 0.32865318918195124, "all_on_minus_masked": {"mean": -0.0007750398161132377, "se": 0.00010928215199558901, "per_batch": [-0.000938767299734633, -0.0006475791739925518, -0.0005816023295552863, -0.0012646314274569193, -0.0008696087762564697, -0.001007775338393302, -0.00026094067702159274, -0.0006294135064951467]}}, "s5_causal": {"masked": 0.042966527107947204, "all_on_minus_masked": {"mean": -8.239279000426831e-05, "se": 9.838852266470529e-05, "per_batch": [-0.0006975025128633169, 1.9213686776077543e-05, 0.00016501197904502524, -6.329767720426871e-05, -5.314801357444304e-05, -6.195406274440984e-05, 0.00020481252657516408, -0.0001722782460439748]}}, "s6_tryfail": {"masked": 0.4697191342745754, "all_on_minus_masked": {"mean": -0.415133246311445, "se": 0.0035097610509610463, "per_batch": [-0.42208600237402033, -0.3952651037425674, -0.4228392702614259, -0.41288486605936603, -0.4166165133710402, -0.4247268658459707, -0.4197081487453598, -0.40693920009180945]}}, "s7_mixed": {"masked": 0.718881555266772, "all_on_minus_masked": {"mean": -0.0001157685090909244, "se": 0.00017021455623028183, "per_batch": [-0.0005450130205841841, 0.0002602957420809737, 0.0004505085539351361, -0.0002393998500009875, -0.00014867900167092607, -0.001006829433813583, 0.00025857591557221227, 4.4393021753963424e-05]}}, "web": {"masked": 0.9521329851910225, "all_on_minus_masked": {"mean": -0.00010013689846484486, "se": 0.00015454286798471408, "per_batch": [-0.0008588813584087251, 0.0002759461633093929, 0.00045575402478637006, -0.0005074348113707217, 0.0001268372050113875, -0.00034792090269542, 3.190278173503813e-05, 2.2701709913919288e-05]}}}, "s7_mixed": {"s1_perspective": {"masked": 0.048797346957739504, "all_on_minus_masked": {"mean": -0.00030447915997908453, "se": 0.0001063028084656894, "per_batch": [-0.0006426346727802443, -0.00028325542597104625, 0.0001642488060318234, -0.0003280300342316106, -0.000561184764223184, -0.00038217769696419907, 0.00011842080403472177, -0.0005212202957289372]}}, "s2_concept": {"masked": 0.11188156547638788, "all_on_minus_masked": {"mean": -0.0006371419768683127, "se": 0.00014238033252011493, "per_batch": [-0.0007091704608327976, -0.0006969919393685631, -0.0007945598469861759, -0.0010622293452227788, 7.361932348495803e-05, -0.0008479819579104431, -0.0009472589431209905, -0.00011256264498971102]}}, "s3_rules": {"masked": 0.22408305361735653, "all_on_minus_masked": {"mean": -0.0008579891984075161, "se": 0.0002711441874389761, "per_batch": [0.000571627335805236, -0.0011028172508263934, -0.00122685973776121, -0.000384037259376635, -0.0016208934887192439, -0.0013455062690251873, -0.0015227558885706483, -0.00023267102878604673]}}, "s4_arith": {"masked": 0.3379026848682451, "all_on_minus_masked": {"mean": -0.010024535502407061, "se": 0.0005104595202619117, "per_batch": [-0.011460536892533124, -0.0094270355265626, -0.009596717909470143, -0.009549143208675226, -0.00755938992951799, -0.012203630931248599, -0.010862037267530544, -0.009537792353718266]}}, "s5_causal": {"masked": 0.04301686327236076, "all_on_minus_masked": {"mean": -0.00013272895441783332, "se": 6.830627059325744e-05, "per_batch": [-7.144535528889884e-05, -0.00011425559920914125, 0.0002395772071097299, -0.0001650334768763806, -0.0003993517769902005, -7.07869842035691e-05, -0.00033894824533248213, -0.00014158740455172403]}}, "s6_tryfail": {"masked": 0.05490541224602387, "all_on_minus_masked": {"mean": -0.000319524282893429, "se": 0.0001584763149306961, "per_batch": [-0.0003116863079418025, -0.0005455504663681859, -0.0011272441617067655, 0.00019735009186507696, 0.00014533608889475552, -3.055379281734216e-05, -0.0002115494094769299, -0.0006722963055962386]}}, "s7_mixed": {"masked": 0.7747712329571779, "all_on_minus_masked": {"mean": -0.056005446199496635, "se": 0.03226567339074146, "per_batch": [-0.0006179336645499856, -0.00011411048884735209, -0.0354403788719736, -0.0010676682965563566, -0.0008022990662749363, -0.18144655517014996, -0.006821863829124142, -0.22173276020849675]}}, "web": {"masked": 0.9527351179495246, "all_on_minus_masked": {"mean": -0.0007022696569669168, "se": 0.0002334784367719282, "per_batch": [-0.0007775335645507919, -0.00040837280447370006, 0.0002590918635247963, -0.0007193174372336353, -0.0010063564815236958, -0.0019098673378314945, -0.0009886422684848295, -6.715922516198347e-05]}}}}, "member_alone": {"s1_perspective": {"s1_perspective": {"alone": 0.05251879489204885, "alone_minus_off": {"mean": -0.5013145064265103, "se": 0.004723722359706749, "per_batch": [-0.5020161539899823, -0.5206146212027206, -0.511447510580648, -0.5087860038161363, -0.4873552833133265, -0.5025627202288822, -0.47857692335015567, -0.4991568349302314]}}, "s2_concept": {"alone": 0.23945795595196098, "alone_minus_off": {"mean": -0.48172597674480316, "se": 0.0029625557017275653, "per_batch": [-0.4753027767620657, -0.4804484116845556, -0.4714891474705084, -0.49853255940543073, -0.48017119714919604, -0.4800707377832608, -0.47906046869643004, -0.4887325150069781]}}, "s3_rules": {"alone": 0.508758760689348, "alone_minus_off": {"mean": -0.42085437549215554, "se": 0.004040788518956882, "per_batch": [-0.4068824458170247, -0.41901358606093464, -0.4391522815246798, -0.40577522151077006, -0.4185499638298712, -0.4321406348403284, -0.4199813324373397, -0.42533953791629586]}}, "s4_arith": {"alone": 0.7225499640202783, "alone_minus_off": {"mean": -0.5113249030739563, "se": 0.003602454892240459, "per_batch": [-0.5097063582233129, -0.5105177293744209, -0.5238398924712184, -0.49785761349211977, -0.5006369821170508, -0.5244675431555423, -0.5047830108770632, -0.5187900948809221]}}, "s5_causal": {"alone": 0.4626770842822263, "alone_minus_off": {"mean": -0.19776059832930892, "se": 0.003074781006523783, "per_batch": [-0.18829842997184976, -0.20876677318163606, -0.20486710956898452, -0.20499300086941613, -0.19802718756151977, -0.18360145478979822, -0.19417430322990303, -0.19935652746136384]}}, "s6_tryfail": {"alone": 0.49460838826503134, "alone_minus_off": {"mean": -0.1782128728751334, "se": 0.002204184674331319, "per_batch": [-0.1800673403015557, -0.18936055182791578, -0.1759819784216351, -0.1815247642806299, -0.17975192411987334, -0.17801537229844877, -0.17237782399882867, -0.16862322775217997]}}, "s7_mixed": {"alone": 0.7729224570816076, "alone_minus_off": {"mean": -0.010053471583414728, "se": 0.009004724163521997, "per_batch": [0.0017783866012468508, 0.0011675042253822543, 0.0005177537704225132, 0.001957162566818771, 0.0010375713326553626, -0.06640314643307121, 0.008738567477327486, -0.029221572208099844]}}, "web": {"alone": 0.9520279682848395, "alone_minus_off": {"mean": 0.0014312073737332287, "se": 0.0004766934609748917, "per_batch": [0.0014245322970964924, -0.0004134462926742577, 0.00041473616255571777, 1.4618525323362164e-05, 0.0028610173884282553, 0.0033452345419324736, 0.002158726233630004, 0.0016442401335737822]}}}, "s2_concept": {"s1_perspective": {"alone": 0.5265382088974727, "alone_minus_off": {"mean": -0.02729509242108652, "se": 0.0008478035966868174, "per_batch": [-0.023477093695305706, -0.029719504978010347, -0.026994955823041678, -0.02971271166330125, -0.0301825252697957, -0.02641580424626122, -0.025215322347576063, -0.02664282134540019]}}, "s2_concept": {"alone": 0.35026574371684355, "alone_minus_off": {"mean": -0.3709181889799206, "se": 0.0024118108231179045, "per_batch": [-0.37237346048863007, -0.3698774117831641, -0.3638556972371791, -0.385719013225945, -0.3722404964016817, -0.36824759669443125, -0.36423210276642376, -0.37079973324191]}}, "s3_rules": {"alone": 0.6188593181726778, "alone_minus_off": {"mean": -0.31075381800882573, "se": 0.002804570723181844, "per_batch": [-0.3059784617336435, -0.31155671901437876, -0.3166699782028437, -0.29686763780852365, -0.3055774841831569, -0.3180009949335336, -0.3101597899327456, -0.3212194782617802]}}, "s4_arith": {"alone": 0.8801855050550988, "alone_minus_off": {"mean": -0.3536893620391357, "se": 0.004028817126679558, "per_batch": [-0.3418542146862763, -0.3484509532254211, -0.36770071336251886, -0.35094263787112157, -0.33855532946875133, -0.36934916707103904, -0.36129874517547633, -0.3513631354524811]}}, "s5_causal": {"alone": 0.6589420516116804, "alone_minus_off": {"mean": -0.001495630999854905, "se": 0.000777079044912105, "per_batch": [0.0004168429500286397, -0.0005169798484936372, -0.005893587471092143, -0.0007514781929449654, -0.0015341712370217087, -0.001300575801487236, -0.003323435740935632, 0.0009383373431074427]}}, "s6_tryfail": {"alone": 0.6732533568015419, "alone_minus_off": {"mean": 0.0004320956613771132, "se": 0.001552406797931269, "per_batch": [-0.001823876012400083, -0.007744722733654008, 0.0013104218082491403, -0.002184394644271337, 0.0010933367071975786, 0.004876267095575981, 0.006227835753043975, 0.001701897317275658]}}, "s7_mixed": {"alone": 0.7829373884278554, "alone_minus_off": {"mean": -3.8540237166762026e-05, "se": 0.0010946548710171276, "per_batch": [0.000501845375217691, 0.0012001809290462795, 0.0021090232475305948, -8.031589795287619e-05, 0.000561265381090803, 0.0019007092616730725, 0.0009867504593253917, -0.007487780653265053]}}, "web": {"alone": 0.9516445652114878, "alone_minus_off": {"mean": 0.0010478043003816634, "se": 0.000252314831290859, "per_batch": [9.347257074376891e-05, 0.0012871181590572434, 0.0019100393204823929, 0.0013596088463959655, 0.0010667223919766933, -7.928400204770814e-05, 0.0010015409672995412, 0.0017432161491454101]}}}, "s3_rules": {"s1_perspective": {"alone": 0.55350150916384, "alone_minus_off": {"mean": -0.00033179215471922097, "se": 0.00018341479131836675, "per_batch": [0.0003735463176740561, -0.00032229548771689487, -0.0008996842423260043, -0.0003734173306858546, -0.0009771194308770825, -0.0004045461904920167, 0.00045029357562154093, -0.0005011144489515118]}}, "s2_concept": {"alone": 0.7242528502377963, "alone_minus_off": {"mean": 0.0030689175410321956, "se": 0.00019465440878027398, "per_batch": [0.002732804322210014, 0.003277301394841614, 0.0030090514551583958, 0.0019815841032413406, 0.0035649423784097634, 0.003502254702170249, 0.0028635111368657817, 0.0036198908353604065]}}, "s3_rules": {"alone": 0.8236434891082838, "alone_minus_off": {"mean": -0.10596964707321978, "se": 0.009346493007512462, "per_batch": [-0.11229495399364031, -0.10826746427573508, -0.09548760547277069, -0.1291266380596071, -0.08108225264517621, -0.11603213699688908, -0.14432027336743114, -0.0611458517745086]}}, "s4_arith": {"alone": 1.2339533341786908, "alone_minus_off": {"mean": 7.84670844561075e-05, "se": 0.00040067832550554967, "per_batch": [0.0005466468557673299, -0.00020706711163875902, -8.02299066273715e-05, 0.0007894003674602246, -7.782214951546074e-05, 0.0008427149892276109, -0.002416270253297048, 0.0012303638842723341]}}, "s5_causal": {"alone": 0.6675733556594478, "alone_minus_off": {"mean": 0.007135673047912558, "se": 0.0005647387818478271, "per_batch": [0.007655162768217094, 0.008555921902111074, 0.003638808926955339, 0.00856619786550028, 0.007419245566896104, 0.007850750037911025, 0.006295510926158543, 0.007103786389551003]}}, "s6_tryfail": {"alone": 0.6738381569335117, "alone_minus_off": {"mean": 0.0010168957933469058, "se": 0.00017347574800339347, "per_batch": [0.0017297155110527207, 0.0012240865175160431, 0.0009007161382312834, 0.0013980039732010008, 0.00020771204657954456, 0.0009364025382853081, 0.0012371142033188454, 0.0005014154185905006]}}, "s7_mixed": {"alone": 0.782468375615579, "alone_minus_off": {"mean": -0.0005075530494431726, "se": 0.0008395569992632177, "per_batch": [2.7775198114365907e-05, -2.235774461123441e-06, 0.0007041829639578001, 0.0003203176872319524, -0.0004014075071138157, 0.0012020727382056062, 0.0003502426684821036, -0.00626137236996227]}}, "web": {"alone": 0.9511259085321484, "alone_minus_off": {"mean": 0.0005291476210421558, "se": 0.0001432382824627441, "per_batch": [-0.00010852105269454349, 0.0007161357581927374, 0.0011528857000588921, 0.0004110385355621249, 0.0007747818421370845, 0.0008065986325466179, 0.0002391418760246955, 0.00024111967650963795]}}}, "s4_arith": {"s1_perspective": {"alone": 0.5529388195514218, "alone_minus_off": {"mean": -0.0008944817671374533, "se": 0.00022219662593559158, "per_batch": [-0.0008394043231986226, -0.0012983400270260015, 0.0003624964323560853, -0.0017084756536711598, -0.0014122355375599804, -0.0008876454567656733, -0.0007099873784243149, -0.0006622621928099592]}}, "s2_concept": {"alone": 0.7207903557747224, "alone_minus_off": {"mean": -0.0003935769220416857, "se": 0.00036656930724860673, "per_batch": [-0.000407598882544824, -0.00045631296840176194, 0.001248465058275916, -0.00226862314753129, 0.0005794525464195566, -0.0009882123118576391, -0.0004823253443447806, -0.00037346032634866244]}}, "s3_rules": {"alone": 0.9280330993212799, "alone_minus_off": {"mean": -0.001580036860223663, "se": 0.0002472548294138001, "per_batch": [-0.0006835450458542169, -0.0017735710870226962, -0.0018863057146633055, -0.0008067706151975162, -0.0008262906460703157, -0.0023852273848162797, -0.0022952804584150943, -0.00198330392974988]}}, "s4_arith": {"alone": 1.126693075548818, "alone_minus_off": {"mean": -0.10718179154541654, "se": 0.019733005800031626, "per_batch": [-0.12218541827078422, -0.000754745863311479, -0.15389351564505827, -0.15261577054047337, -0.042793669092263675, -0.11302235461546473, -0.12726140021967436, -0.14492745811630225]}}, "s5_causal": {"alone": 0.6571682010528499, "alone_minus_off": {"mean": -0.0032694815586852416, "se": 0.00036918803111562825, "per_batch": [-0.004515404498403042, -0.0033061944801866527, -0.003756875016772576, -0.0038970838728884516, -0.0014912615656313966, -0.0025041533878878974, -0.002423407533307831, -0.004261472114404086]}}, "s6_tryfail": {"alone": 0.6593261748623824, "alone_minus_off": {"mean": -0.013495086277782314, "se": 0.0007129585002580727, "per_batch": [-0.01444473685464187, -0.015171234567549097, -0.014263424144969727, -0.016344715190048764, -0.014004375277107628, -0.011502156694041865, -0.010623712309097422, -0.011606335184802141]}}, "s7_mixed": {"alone": 0.7828199027794849, "alone_minus_off": {"mean": -0.0001560258855374183, "se": 0.0002352169625457898, "per_batch": [0.00038696096444135186, 0.0003690747687516982, -0.001014955614066837, 0.0002759461633093929, 4.695126368559066e-05, -0.0010977222647946538, 0.00047183440264214616, -0.0006862967682680354]}}, "web": {"alone": 0.9507103339541697, "alone_minus_off": {"mean": 0.00011357304306353067, "se": 0.00010289644243726181, "per_batch": [-0.0003639152892258579, 4.480148054974986e-05, -4.5575402478625904e-05, 0.0002953802028569097, 5.933401454827347e-06, 0.0003312385855618327, 3.989997500020159e-05, 0.0006008213907892079]}}}, "s5_causal": {"s1_perspective": {"alone": 0.5355407643694168, "alone_minus_off": {"mean": -0.0182925369491425, "se": 0.0010792101601506997, "per_batch": [-0.017861086222706102, -0.017265596294093633, -0.023324072131700513, -0.01829319263299889, -0.012617335197899093, -0.02079786196883937, -0.017867363589462615, -0.018313787555439776]}}, "s2_concept": {"alone": 0.7102577029047991, "alone_minus_off": {"mean": -0.010926229791965086, "se": 0.0004629922489753803, "per_batch": [-0.012461389929229183, -0.010230344990911466, -0.010227593268497648, -0.013156930764981145, -0.009059831069139102, -0.010504141371085018, -0.010754161149776897, -0.011015445792100231]}}, "s3_rules": {"alone": 0.9173749367374315, "alone_minus_off": {"mean": -0.01223819944407202, "se": 0.0003759474969709794, "per_batch": [-0.01179603204870372, -0.013579406146825623, -0.013483955775596801, -0.010735930988785447, -0.01172250946545983, -0.011976441849458785, -0.01136538749094329, -0.013245931786802667]}}, "s4_arith": {"alone": 1.2247629037840526, "alone_minus_off": {"mean": -0.009111963310181992, "se": 0.0003133156648215143, "per_batch": [-0.00816814402007826, -0.009321803642065918, -0.009664909030537183, -0.009344677334630846, -0.007534817908275926, -0.010003456878760808, -0.010028738328437425, -0.008829159338669568]}}, "s5_causal": {"alone": 0.09109720145841224, "alone_minus_off": {"mean": -0.5693404811531231, "se": 0.007627785122552896, "per_batch": [-0.5395056362790345, -0.5729805168000802, -0.5723726226227741, -0.5818498663344731, -0.6069291611541324, -0.5535619986868231, -0.5485124375726715, -0.579011609774995]}}, "s6_tryfail": {"alone": 0.6662963307322131, "alone_minus_off": {"mean": -0.00652493040795174, "se": 0.00038463576812808813, "per_batch": [-0.006364690947468166, -0.005736481319529085, -0.006751049972631429, -0.00707196959914147, -0.008866049617279548, -0.006194643101427477, -0.005688197190299338, -0.005526361515837408]}}, "s7_mixed": {"alone": 0.7819243246229481, "alone_minus_off": {"mean": -0.0010516040420741543, "se": 0.0008258915662142286, "per_batch": [0.0004442311871786231, -9.149477025904851e-05, 0.00015272059396620463, 8.925899579781404e-05, 0.00026373539509816357, -0.00625333218103441, -0.00026807795703237547, -0.002749873600308206]}}, "web": {"alone": 0.9508772001211694, "alone_minus_off": {"mean": 0.0002804392100631825, "se": 0.00013042933220109836, "per_batch": [0.0004127583620707753, -5.262669116401497e-05, 0.00043288033222166344, 0.0006003914341621286, 0.0007997193265121272, -0.0001322546585135198, -0.00021454835695133845, 0.00039719393216763876]}}}, "s6_tryfail": {"s1_perspective": {"alone": 0.5518238130302154, "alone_minus_off": {"mean": -0.0020094882883437554, "se": 0.0003256805712077641, "per_batch": [-0.0010533937365346802, -0.002643029378459616, -0.0009457755927573253, -0.001739604513477322, -0.0036941873405331727, -0.0021753225594381975, -0.002429684900064233, -0.0013949082854854966]}}, "s2_concept": {"alone": 0.7223676892826529, "alone_minus_off": {"mean": 0.0011837565858887206, "se": 0.00030040574308138647, "per_batch": [0.0005063599198028568, 0.00019365246487146326, 0.0026010656116491893, 0.0005574817627718165, 0.0017863837945120142, 0.0005542570880681108, 0.0017871577164408903, 0.0014836943289934235]}}, "s3_rules": {"alone": 0.9301771102893455, "alone_minus_off": {"mean": 0.0005639741078418553, "se": 0.0001864979651283735, "per_batch": [0.0006568877349705238, -0.0003355381518332923, 0.0004389857163272781, 0.0014670980031851188, 0.0005839670910047223, 0.0010127198396056025, 0.00033837586557250443, 0.00034929676390238473]}}, "s4_arith": {"alone": 1.2374947041850164, "alone_minus_off": {"mean": 0.00361983709078198, "se": 0.000311374566884852, "per_batch": [0.004592366734664122, 0.0031974154535159283, 0.002824987023072545, 0.004043054147808256, 0.0025683029166598814, 0.005088364699752379, 0.0034874641941959883, 0.0031567415565867396]}}, "s5_causal": {"alone": 0.6506552986815429, "alone_minus_off": {"mean": -0.009782383929992228, "se": 0.00038481175604609925, "per_batch": [-0.008596939764341949, -0.009804085990748002, -0.010320377908638045, -0.011653931383428295, -0.00987790954363088, -0.010220327001498664, -0.008083184590552017, -0.00970231525709997]}}, "s6_tryfail": {"alone": 0.16377145340459293, "alone_minus_off": {"mean": -0.5090498077355718, "se": 0.004677470835012531, "per_batch": [-0.5175160259054989, -0.49329147411171215, -0.5123662633993884, -0.5249514593394076, -0.5133465000158124, -0.5087508763596975, -0.5168361354909756, -0.485339727262082]}}, "s7_mixed": {"alone": 0.7829311486823038, "alone_minus_off": {"mean": -4.477998271842226e-05, "se": 0.00020200262136100378, "per_batch": [-0.0001941684128241583, -9.149477025904851e-05, 0.0008069425978483036, 0.00044371523922603906, 4.1963766810537706e-05, -0.0005567938321683341, 0.00021566624418195568, -0.0010240706945626732]}}, "web": {"alone": 0.9507240925662387, "alone_minus_off": {"mean": 0.00012733165513255373, "se": 0.00013788984803610982, "per_batch": [-0.0002956381768333127, -0.00018548328895551247, 0.0008409951627191825, 0.0002792998250011891, 0.0002971860206910648, 0.00027371038884815846, -0.00032573514073419574, 0.00013431845032385592]}}}, "s7_mixed": {"s1_perspective": {"alone": 0.4970373744855765, "alone_minus_off": {"mean": -0.056795926832982666, "se": 0.0018110331876671725, "per_batch": [-0.05812665334294931, -0.05100399185746607, -0.050193609606600575, -0.06252992515298983, -0.057888758341143376, -0.05970347627738498, -0.06299913682020625, -0.05192186326512094]}}, "s2_concept": {"alone": 0.7153484807361215, "alone_minus_off": {"mean": -0.005835451960642682, "se": 0.0007318243036266138, "per_batch": [-0.008731774162618389, -0.005678437174862827, -0.0043266965347438235, -0.007888328247124377, -0.0026645702054801657, -0.006611959003746115, -0.003970520464806948, -0.0068113298917588105]}}, "s3_rules": {"alone": 0.867875922062003, "alone_minus_off": {"mean": -0.061737214119500436, "se": 0.001966162214666222, "per_batch": [-0.060009046452305426, -0.0667788855295437, -0.06605509654338759, -0.05384596216731141, -0.06646923076666511, -0.058477798920348434, -0.06718940811715324, -0.05507228445928858]}}, "s4_arith": {"alone": 1.195411129960334, "alone_minus_off": {"mean": -0.03846373713390064, "se": 0.0014722165440657325, "per_batch": [-0.032631902183383854, -0.03863573053367908, -0.0428992664398935, -0.03808968561718973, -0.033519590635812335, -0.041296302142526775, -0.0440718441534762, -0.03656557536524363]}}, "s5_causal": {"alone": 0.5576567389956736, "alone_minus_off": {"mean": -0.10278094361586164, "se": 0.002443483866586268, "per_batch": [-0.09823747907307878, -0.10830788019868787, -0.11166214382979056, -0.10734787304157156, -0.10408054664047994, -0.08949444004682883, -0.10261293268934224, -0.10050425340711333]}}, "s6_tryfail": {"alone": 0.5926274278736829, "alone_minus_off": {"mean": -0.08019383326648177, "se": 0.003441452803867667, "per_batch": [-0.08156569587675166, -0.10006397782090459, -0.07540334251802361, -0.08326673328077305, -0.08425219387021687, -0.07515306476535555, -0.0677320133806254, -0.07411364461920344]}}, "s7_mixed": {"alone": 0.7241658592632068, "alone_minus_off": {"mean": -0.058810069401815435, "se": 0.034794470993815295, "per_batch": [-0.0003199737219301557, -0.0002995077864775819, -0.03512848833463389, 6.879306034512922e-05, -0.0005225692846466679, -0.20927880852891734, -0.00042617300883801557, -0.22457382760942496]}}, "web": {"alone": 0.9504116968298622, "alone_minus_off": {"mean": -0.00018506408124407958, "se": 0.0001933827163928398, "per_batch": [9.8890024244902e-06, -0.0006554258824381654, 0.0008509701564691774, -0.0003986557846999972, -9.381653604556561e-05, -0.0010095381605647047, -0.00016467338820114197, -1.926205689672944e-05]}}}}, "battery": {"s1_perspective": {"acc_off": {"in": 0.55, "transfer": 0.4625}, "acc_all_on": {"in": 0.9, "transfer": 0.4125}, "acc_member_masked": {"in": 0.55, "transfer": 0.4375}, "acc_member_alone": {"in": 0.8375, "transfer": 0.4625}, "margin_all_on_minus_off": {"in": {"mean": 2.1786898315593133, "se": 0.18238070780221505, "per_batch": null}, "transfer": {"mean": -0.24194687381386756, "se": 0.14483030691607612, "per_batch": null}}}, "s2_concept": {"acc_off": {"in": 0.2976190476190476, "transfer": 0.6779661016949152}, "acc_all_on": {"in": 0.9880952380952381, "transfer": 0.6864406779661016}, "acc_member_masked": {"in": 0.7261904761904762, "transfer": 0.6779661016949152}, "acc_member_alone": {"in": 0.9642857142857143, "transfer": 0.7203389830508474}, "margin_all_on_minus_off": {"in": {"mean": 4.215096542168231, "se": 0.3270012040208027, "per_batch": null}, "transfer": {"mean": 0.20048295794907262, "se": 0.22334445470985073, "per_batch": null}}}, "s3_rules": {"acc_off": {"in": 0.155, "transfer": 0.07, "held-out": 0.05}, "acc_all_on": {"in": 0.87, "transfer": 0.32, "held-out": 0.44166666666666665}, "acc_member_masked": {"in": 0.81, "transfer": 0.25, "held-out": 0.4083333333333333}, "acc_member_alone": {"in": 0.32, "transfer": 0.08, "held-out": 0.10833333333333334}, "margin_all_on_minus_off": {"in": {"mean": 3.6436909707443554, "se": 0.17905952477194872, "per_batch": null}, "transfer": {"mean": 0.9199708187580109, "se": 0.1802350690514543, "per_batch": null}, "held-out": {"mean": 1.769368333951085, "se": 0.3311439870578026, "per_batch": null}}}, "s4_arith": {"acc_off": {"in": 0.34375, "transfer": 0.24166666666666667, "held-out": 0.35}, "acc_all_on": {"in": 0.64375, "transfer": 0.24166666666666667, "held-out": 0.325}, "acc_member_masked": {"in": 0.5125, "transfer": 0.23333333333333334, "held-out": 0.3}, "acc_member_alone": {"in": 0.38125, "transfer": 0.24583333333333332, "held-out": 0.3875}, "margin_all_on_minus_off": {"in": {"mean": 1.818516574592286, "se": 0.2384041733542834, "per_batch": null}, "transfer": {"mean": -0.3892519031961759, "se": 0.07495719462712842, "per_batch": null}, "held-out": {"mean": -0.1310452938079834, "se": 0.117156809174464, "per_batch": null}}}}, "probes_all_on": {"p0_grounding": {"acc": 0.43333333333333335, "n": 30, "families": {"count": 0.125, "permanence": 0.5714285714285714, "sequence": 0.625, "attribute": 0.42857142857142855}}, "p1_self_other": {"acc": 0.4666666666666667, "n": 30, "families": {"belief": 0.125, "reality": 0.5, "perspective": 0.8, "reference": 0.8571428571428571, "second_order": 0.0}}, "p2_difference": {"acc": 0.5666666666666667, "n": 30, "families": {"category": 1.0, "antonym": 0.6666666666666666, "odd_one_out": 0.3333333333333333, "property": 0.5, "analogy": 0.3333333333333333}}, "p3_logic": {"acc": 0.2, "n": 30, "families": {"d1": 0.5, "d2": 0.5, "d3": 0.25, "d4": 0.0, "d5": 0.0, "d6": 0.0, "negation": 0.25, "contrapositive": 0.0}}, "p4_arithmetic": {"acc": 0.4, "n": 30, "families": {"add2": 0.4, "add3": 0.0, "sub2": 0.75, "sub3": 0.4, "mul1": 0.5, "mul2": 0.3333333333333333, "word": 0.4}}, "p5_causal": {"acc": 0.5, "n": 30, "families": {"effect": 0.5714285714285714, "cause": 0.42857142857142855, "counterfactual": 0.5, "prevention": 0.4, "chain": 0.6}}, "p6_learning": {"acc": 0.36666666666666664, "n": 30, "families": {"error_detect": 0.25, "correction": 0.25, "strategy": 0.7142857142857143, "check": 0.2857142857142857}}, "p7_composite": {"acc": 0.36666666666666664, "n": 30, "families": {"logic_arith": 0.375, "causal_category": 0.375, "belief_causal": 0.2857142857142857, "three_step": 0.42857142857142855}}, "p8_articulation": {"acc": 0.4666666666666667, "n": 30, "families": {"definition": 0.375, "register": 0.42857142857142855, "grammar": 0.875, "paraphrase": 0.14285714285714285}}}, "probes_mean_off": 0.41111111111111115, "probes_mean_all_on": 0.4185185185185185, "gates": {"s1_perspective": {"mean": 0.2099039478571495, "max": 0.5712027580899203}, "s2_concept": {"mean": 0.27674460361666475, "max": 0.4977012170865557}, "s3_rules": {"mean": 0.12033936123137423, "max": 0.1738736117222015}, "s4_arith": {"mean": 0.10767731599767934, "max": 0.15932411779931022}, "s5_causal": {"mean": 0.18518384792916404, "max": 0.3681551185947585}, "s6_tryfail": {"mean": 0.23037076111374397, "max": 0.5373235543789295}, "s7_mixed": {"mean": 0.2511517146927655, "max": 0.463826973625139}}, "chain_proof": "compiled chain: 23 cache entries (limit 256), 0 limit warnings", "curve": [{"step": 1, "train_bpb": {"s7_mixed": 0.3642366388088402}, "quiet_kl": {"s7_mixed": 0.00016790055087767541}}, {"step": 50, "train_bpb": {"s7_mixed": 0.46902312306627336}, "quiet_kl": {"s7_mixed": 0.0001510203882935457}}, {"step": 100, "train_bpb": {"s7_mixed": 0.5390128901834064}, "quiet_kl": {"s7_mixed": 0.00014959990381612442}}, {"step": 150, "train_bpb": {"s7_mixed": 0.4871148901560276}, "quiet_kl": {"s7_mixed": 0.00017191540289786643}}, {"step": 200, "train_bpb": {"s7_mixed": 0.3962228283674392}, "quiet_kl": {"s7_mixed": 0.00021447284321766347}}, {"step": 250, "train_bpb": {"s7_mixed": 0.43212297082472584}, "quiet_kl": {"s7_mixed": 0.00027403544227126985}}, {"step": 300, "train_bpb": {"s7_mixed": 0.5527450762582127}, "quiet_kl": {"s7_mixed": 0.00028941785712959244}}, {"step": 350, "train_bpb": {"s7_mixed": 0.44438169408716144}, "quiet_kl": {"s7_mixed": 0.000400753513968084}}, {"step": 400, "train_bpb": {"s7_mixed": 0.48180019274059743}, "quiet_kl": {"s7_mixed": 0.0004926037209224887}}, {"step": 450, "train_bpb": {"s7_mixed": 0.4725850659597629}, "quiet_kl": {"s7_mixed": 0.0005281461944105103}}, {"step": 500, "train_bpb": {"s7_mixed": 0.46816784641048215}, "quiet_kl": {"s7_mixed": 0.00042564627452520656}}, {"step": 550, "train_bpb": {"s7_mixed": 0.43819326896473126}, "quiet_kl": {"s7_mixed": 0.0004602784806047566}}, {"step": 600, "train_bpb": {"s7_mixed": 0.3918486518835922}, "quiet_kl": {"s7_mixed": 0.00034343611157964914}}, {"step": 650, "train_bpb": {"s7_mixed": 0.4018861649281106}, "quiet_kl": {"s7_mixed": 0.0005043115210719407}}, {"step": 700, "train_bpb": {"s7_mixed": 0.4464929599884592}, "quiet_kl": {"s7_mixed": 0.0004680038316291757}}, {"step": 750, "train_bpb": {"s7_mixed": 0.40829855628494227}, "quiet_kl": {"s7_mixed": 0.0006034824170637876}}, {"step": 800, "train_bpb": {"s7_mixed": 0.42667008766034425}, "quiet_kl": {"s7_mixed": 0.00043090361898066474}}, {"step": 850, "train_bpb": {"s7_mixed": 0.3861280573539475}, "quiet_kl": {"s7_mixed": 0.000651449701981619}}, {"step": 900, "train_bpb": {"s7_mixed": 0.4337167188943003}, "quiet_kl": {"s7_mixed": 0.0005231048390851356}}, {"step": 950, "train_bpb": {"s7_mixed": 0.511076978312675}, "quiet_kl": {"s7_mixed": 0.0006589995080139488}}, {"step": 1000, "train_bpb": {"s7_mixed": 0.4723076572989862}, "quiet_kl": {"s7_mixed": 0.0006425097622559405}}, {"step": 1050, "train_bpb": {"s7_mixed": 0.5202561042339272}, "quiet_kl": {"s7_mixed": 0.0007421331288060174}}, {"step": 1100, "train_bpb": {"s7_mixed": 0.3957248623296346}, "quiet_kl": {"s7_mixed": 0.0006784632848575712}}, {"step": 1150, "train_bpb": {"s7_mixed": 0.42217836511922036}, "quiet_kl": {"s7_mixed": 0.0008336551420507021}}, {"step": 1200, "train_bpb": {"s7_mixed": 0.40625395677102255}, "quiet_kl": {"s7_mixed": 0.000526208998344373}}, {"step": 1250, "train_bpb": {"s7_mixed": 0.44920271240108256}, "quiet_kl": {"s7_mixed": 0.0005948563566198573}}, {"step": 1300, "train_bpb": {"s7_mixed": 0.5418893463469133}, "quiet_kl": {"s7_mixed": 0.0007853034313302488}}, {"step": 1350, "train_bpb": {"s7_mixed": 0.3959808230340221}, "quiet_kl": {"s7_mixed": 0.0009045846288790926}}, {"step": 1400, "train_bpb": {"s7_mixed": 0.3689174519750214}, "quiet_kl": {"s7_mixed": 0.0006089582294225693}}, {"step": 1450, "train_bpb": {"s7_mixed": 0.3994638702300428}, "quiet_kl": {"s7_mixed": 0.0005905353254638613}}, {"step": 1500, "train_bpb": {"s7_mixed": 0.4600651865582908}, "quiet_kl": {"s7_mixed": 0.0005561842781025917}}, {"step": 1550, "train_bpb": {"s7_mixed": 0.4147487242222266}, "quiet_kl": {"s7_mixed": 0.000571558731899131}}, {"step": 1600, "train_bpb": {"s7_mixed": 0.3934197020591179}, "quiet_kl": {"s7_mixed": 0.0005696060249465517}}, {"step": 1650, "train_bpb": {"s7_mixed": 0.36743036614815294}, "quiet_kl": {"s7_mixed": 0.0006183415334089659}}, {"step": 1700, "train_bpb": {"s7_mixed": 0.49994185995156465}, "quiet_kl": {"s7_mixed": 0.000572950791392941}}, {"step": 1750, "train_bpb": {"s7_mixed": 0.4460065222164649}, "quiet_kl": {"s7_mixed": 0.0005554246320389211}}, {"step": 1800, "train_bpb": {"s7_mixed": 0.5316604451503559}, "quiet_kl": {"s7_mixed": 0.0006525515223620459}}, {"step": 1850, "train_bpb": {"s7_mixed": 0.48753551124242833}, "quiet_kl": {"s7_mixed": 0.000650390861555934}}, {"step": 1900, "train_bpb": {"s7_mixed": 0.4752556098751317}, "quiet_kl": {"s7_mixed": 0.0006546224729390814}}, {"step": 1950, "train_bpb": {"s7_mixed": 0.42374496701722614}, "quiet_kl": {"s7_mixed": 0.0006760741607286036}}, {"step": 2000, "train_bpb": {"s7_mixed": 0.3462074052558746}, "quiet_kl": {"s7_mixed": 0.0006167851929785684}}, {"step": 2050, "train_bpb": {"s7_mixed": 0.49217845409761246}, "quiet_kl": {"s7_mixed": 0.0005724033911246807}}, {"step": 2100, "train_bpb": {"s7_mixed": 0.3943440749484211}, "quiet_kl": {"s7_mixed": 0.000594364617136307}}, {"step": 2150, "train_bpb": {"s7_mixed": 0.4072405806081212}, "quiet_kl": {"s7_mixed": 0.0005634973046835512}}, {"step": 2200, "train_bpb": {"s7_mixed": 0.3955424059353343}, "quiet_kl": {"s7_mixed": 0.0006106325666769408}}, {"step": 2250, "train_bpb": {"s7_mixed": 0.2781348105471577}, "quiet_kl": {"s7_mixed": 0.0005198777513578534}}, {"step": 2300, "train_bpb": {"s7_mixed": 0.39866033084638774}, "quiet_kl": {"s7_mixed": 0.0005882594105787576}}, {"step": 2350, "train_bpb": {"s7_mixed": 0.384167268317957}, "quiet_kl": {"s7_mixed": 0.0006667332642246038}}, {"step": 2400, "train_bpb": {"s7_mixed": 0.4195725818084799}, "quiet_kl": {"s7_mixed": 0.0006168914231238886}}, {"step": 2450, "train_bpb": {"s7_mixed": 0.4183964302954123}, "quiet_kl": {"s7_mixed": 0.0007282642397331073}}, {"step": 2500, "train_bpb": {"s7_mixed": 0.48656071400729467}, "quiet_kl": {"s7_mixed": 0.0006262975744903088}}, {"step": 2550, "train_bpb": {"s7_mixed": 0.38767664127455725}, "quiet_kl": {"s7_mixed": 0.0006999711645767092}}, {"step": 2600, "train_bpb": {"s7_mixed": 0.4070559242332488}, "quiet_kl": {"s7_mixed": 0.0006666993693215772}}, {"step": 2650, "train_bpb": {"s7_mixed": 0.37047537820820436}, "quiet_kl": {"s7_mixed": 0.0006070201058173552}}, {"step": 2700, "train_bpb": {"s7_mixed": 0.4671347314170335}, "quiet_kl": {"s7_mixed": 0.0006557619560044258}}, {"step": 2750, "train_bpb": {"s7_mixed": 0.376819534857116}, "quiet_kl": {"s7_mixed": 0.0006533763039624318}}, {"step": 2800, "train_bpb": {"s7_mixed": 0.45487450642358823}, "quiet_kl": {"s7_mixed": 0.0006340613274369389}}, {"step": 2850, "train_bpb": {"s7_mixed": 0.44002908804359764}, "quiet_kl": {"s7_mixed": 0.0005979629437206313}}, {"step": 2900, "train_bpb": {"s7_mixed": 0.41176210317435036}, "quiet_kl": {"s7_mixed": 0.000628843087470159}}, {"step": 2950, "train_bpb": {"s7_mixed": 0.38827030850953803}, "quiet_kl": {"s7_mixed": 0.0006400322035187855}}, {"step": 3000, "train_bpb": {"s7_mixed": 0.4966694058339929}, "quiet_kl": {"s7_mixed": 0.000613626838894561}}, {"step": 3050, "train_bpb": {"s7_mixed": 0.44479213745526236}, "quiet_kl": {"s7_mixed": 0.0006889841874362901}}, {"step": 3100, "train_bpb": {"s7_mixed": 0.3772843826258007}, "quiet_kl": {"s7_mixed": 0.0007423702668165788}}, {"step": 3150, "train_bpb": {"s7_mixed": 0.38945356279859716}, "quiet_kl": {"s7_mixed": 0.0006747859693132341}}, {"step": 3200, "train_bpb": {"s7_mixed": 0.3850043320110219}, "quiet_kl": {"s7_mixed": 0.0007386741292430088}}, {"step": 3250, "train_bpb": {"s7_mixed": 0.40347374838705397}, "quiet_kl": {"s7_mixed": 0.0006668977852677927}}, {"step": 3300, "train_bpb": {"s7_mixed": 0.43751872704941597}, "quiet_kl": {"s7_mixed": 0.0006218865688424558}}, {"step": 3350, "train_bpb": {"s7_mixed": 0.4424722296724341}, "quiet_kl": {"s7_mixed": 0.0006448291632113978}}, {"step": 3400, "train_bpb": {"s7_mixed": 0.3765403509054259}, "quiet_kl": {"s7_mixed": 0.0006232062156777829}}, {"step": 3450, "train_bpb": {"s7_mixed": 0.3871429037590311}, "quiet_kl": {"s7_mixed": 0.0006700096605345607}}, {"step": 3500, "train_bpb": {"s7_mixed": 0.40825074779523135}, "quiet_kl": {"s7_mixed": 0.0007485594123136252}}, {"step": 3550, "train_bpb": {"s7_mixed": 0.3569012913235362}, "quiet_kl": {"s7_mixed": 0.0006780203530797735}}, {"step": 3600, "train_bpb": {"s7_mixed": 0.36137753478114737}, "quiet_kl": {"s7_mixed": 0.0007626503059873358}}, {"step": 3650, "train_bpb": {"s7_mixed": 0.4564877505002093}, "quiet_kl": {"s7_mixed": 0.0006500640523154289}}, {"step": 3700, "train_bpb": {"s7_mixed": 0.37513895849779083}, "quiet_kl": {"s7_mixed": 0.0006405016331700608}}, {"step": 3750, "train_bpb": {"s7_mixed": 0.46916800049893614}, "quiet_kl": {"s7_mixed": 0.0006066528399242088}}, {"step": 3800, "train_bpb": {"s7_mixed": 0.4174741085141989}, "quiet_kl": {"s7_mixed": 0.0011492666270351037}}, {"step": 3850, "train_bpb": {"s7_mixed": 0.43142920184023437}, "quiet_kl": {"s7_mixed": 0.0008312835078686476}}, {"step": 3900, "train_bpb": {"s7_mixed": 0.3549475959269588}, "quiet_kl": {"s7_mixed": 0.0007130836939904839}}, {"step": 3950, "train_bpb": {"s7_mixed": 0.4112838657501275}, "quiet_kl": {"s7_mixed": 0.0007259232515934855}}, {"step": 4000, "train_bpb": {"s7_mixed": 0.4131729150717732}, "quiet_kl": {"s7_mixed": 0.0006511309003690257}}], "steps_done": {"s1_perspective": 3200, "s2_concept": 2400, "s3_rules": 800, "s4_arith": 800, "s5_causal": 4000, "s6_tryfail": 4000, "s7_mixed": 4000, "s8_register": 0}, "stage_seconds": 3113.776780605316, "inert_at_birth": true}, {"close": 8, "attached": ["s1_perspective", "s2_concept", "s3_rules", "s4_arith", "s5_causal", "s6_tryfail", "s7_mixed", "s8_register"], "rows": {"s1_perspective": {"off": 0.5538333013185592, "all_on": 0.049211667943427066, "all_on_minus_off": {"mean": -0.5046216333751321, "se": 0.004827993887953663, "per_batch": [-0.5055083477070771, -0.5245492349147021, -0.5152920860004448, -0.511084374587809, -0.49221514218911855, -0.506423607128222, -0.48044241378960667, -0.5014578606840768]}}, "s2_concept": {"off": 0.7211839326967642, "all_on": 0.11045867104531781, "all_on_minus_off": {"mean": -0.6107252616514464, "se": 0.003846341077262944, "per_batch": [-0.6113034323896711, -0.6101417970722497, -0.5961741400911013, -0.6332875124460897, -0.6089768940844618, -0.6080903665149267, -0.6021887603527382, -0.6156391902603323]}}, "s3_rules": {"off": 0.9296131361815035, "all_on": 0.2237152122866791, "all_on_minus_off": {"mean": -0.7058979238948244, "se": 0.005308375345307719, "per_batch": [-0.6856023238216852, -0.7057338820054194, -0.7209254965170958, -0.6893075825453677, -0.6943165987495783, -0.7272337126615868, -0.7115760251660723, -0.7124877696917901]}}, "s4_arith": {"off": 1.2338748670942346, "all_on": 0.32794756855052143, "all_on_minus_off": {"mean": -0.9059272985437131, "se": 0.006203701657997884, "per_batch": [-0.9215295646299871, -0.8978054608523797, -0.9203441742089151, -0.8795382805953238, -0.8832252016688579, -0.9259297192544808, -0.9091697304560923, -0.9098762566836682]}}, "s5_causal": {"off": 0.6604376826115352, "all_on": 0.04305856436254439, "all_on_minus_off": {"mean": -0.6173791182489908, "se": 0.006926127398144821, "per_batch": [-0.583298126862931, -0.6267407079947315, -0.627996482315669, -0.6287680367958733, -0.6432474711503268, -0.6157478456744727, -0.594548022629362, -0.6186862525685604]}}, "s6_tryfail": {"off": 0.6728212611401647, "all_on": 0.05483553152977349, "all_on_minus_off": {"mean": -0.6179857296103912, "se": 0.0037034955412284013, "per_batch": [-0.6278270095366173, -0.6235137383975564, -0.622208379328592, -0.6269420728066026, -0.6221451703299421, -0.6174263748490624, -0.6039438003091304, -0.5998792913256269]}}, "s7_mixed": {"off": 0.7829759286650222, "all_on": 0.719981644980245, "all_on_minus_off": {"mean": -0.06299428368477725, "se": 0.039024436449205856, "per_batch": [-5.477647429974475e-05, 0.0014537693477434388, -0.032563539079665915, 0.003692295531373624, 0.0006430431315760377, -0.2387040081956387, 0.00438267688759969, -0.24280373062690636]}}, "s8_register": {"off": 0.4906504977761471, "all_on": 0.02118320337400121, "all_on_minus_off": {"mean": -0.4694672944021459, "se": 0.008515764702868935, "per_batch": [-0.42507686263300665, -0.48827611067279303, -0.4582893339064262, -0.4883499516926639, -0.4621684738082032, -0.49672457566486194, -0.48354009005453785, -0.4533129567846746]}}, "web": {"off": 0.9505967609111062, "all_on": 0.9524873016985469, "all_on_minus_off": {"mean": 0.0018905407874407892, "se": 0.0003216060834452311, "per_batch": [0.0005862888567913505, 0.0010316379312005353, 0.001940652232335971, 0.0012538395161153515, 0.0033688821564261673, 0.0019352347788337276, 0.0025564361137502267, 0.002451354714072984]}}}, "member_in_group": {"s1_perspective": {"s1_perspective": {"masked": 0.43244818281573383, "all_on_minus_masked": {"mean": -0.3832365148723068, "se": 0.005157017851679235, "per_batch": [-0.38494885341755003, -0.40404155039874745, -0.3974797135888121, -0.3861240162453534, -0.3716344940335353, -0.38239234206139006, -0.3566542032041066, -0.382616946028959]}}, "s2_concept": {"masked": 0.3278164210303229, "all_on_minus_masked": {"mean": -0.21735774998500507, "se": 0.001286414032913593, "per_batch": [-0.21324885604143443, -0.21597471656411582, -0.21226120267319198, -0.221139086846634, -0.21746277495287492, -0.21727645324849643, -0.21844965439918845, -0.2230492551541045]}}, "s3_rules": {"masked": 0.4068656183895294, "all_on_minus_masked": {"mean": -0.18315040610285027, "se": 0.0025870547558581964, "per_batch": [-0.18032163814868774, -0.18376903038523273, -0.18332329434985903, -0.17473170754552522, -0.18242855460874524, -0.1992293032953881, -0.18419617079648185, -0.1772035496928823]}}, "s4_arith": {"masked": 0.4559598235260316, "all_on_minus_masked": {"mean": -0.12801225497551014, "se": 0.004912314076564061, "per_batch": [-0.14006245588435778, -0.14609921891229322, -0.10131635548448747, -0.12883564341434717, -0.13720793084099958, -0.1257281533894012, -0.12501259807265508, -0.11983568380553966]}}, "s5_causal": {"masked": 0.05582799436597497, "all_on_minus_masked": {"mean": -0.012769430003430584, "se": 0.001044915635451976, "per_batch": [-0.013504056247916375, -0.015144593380038866, -0.015570099956104817, -0.013021255264052838, -0.01384929260892103, -0.014704575767806372, -0.008052265334601572, -0.008309301468002801]}}, "s6_tryfail": {"masked": 0.08285324264882162, "all_on_minus_masked": {"mean": -0.028017711119048133, "se": 0.002311243546890652, "per_batch": [-0.04141044487812126, -0.026706143207664693, -0.030906260511373504, -0.024598753795317264, -0.02722222552169911, -0.025109671255367956, -0.018702162002292848, -0.029486027780548422]}}, "s7_mixed": {"masked": 0.7228516215898917, "all_on_minus_masked": {"mean": -0.0028699766096466142, "se": 0.0028568772401661125, "per_batch": [-0.00031438428577712507, -0.00029116662791062176, -0.0011474682465566488, 0.0022101490462379525, 0.00033252845544318177, -0.0218930474895338, 0.003280569065208072, -0.005136992794283923]}}, "s8_register": {"masked": 0.04835342043513964, "all_on_minus_masked": {"mean": -0.027170217061138438, "se": 0.003183107468926375, "per_batch": [-0.04552271263760066, -0.02019739260503871, -0.030140437890454924, -0.023471699083249354, -0.02188482053819606, -0.022853446982072294, -0.019150757249237132, -0.034140469503258365]}}, "web": {"masked": 0.9522344549550316, "all_on_minus_masked": {"mean": 0.0002528467435153753, "se": 0.0004887731927533195, "per_batch": [-0.000766784648871699, -0.0004991796441292662, -0.0017040900960740846, -0.0009386813084092394, 0.0013282220126134003, 0.001486704025383534, 0.0020255256705367097, 0.0010910579370736473]}}}, "s2_concept": {"s1_perspective": {"masked": 0.05493950982197912, "all_on_minus_masked": {"mean": -0.005727841878552048, "se": 0.00047044566677306287, "per_batch": [-0.00525768699398476, -0.006389940150397906, -0.006061888618334933, -0.003926240306667619, -0.007905886600885974, -0.0060482912400011, -0.003936822614153512, -0.006295978503990579]}}, "s2_concept": {"masked": 0.23069932024906098, "all_on_minus_masked": {"mean": -0.12024064920374317, "se": 0.0011098290697982628, "per_batch": [-0.1230749555367093, -0.12150984892036365, -0.11594605968013229, -0.12513587738657672, -0.12090367456956631, -0.12028593438549846, -0.11657618261506227, -0.11849266053603628]}}, "s3_rules": {"masked": 0.3323613071172043, "all_on_minus_masked": {"mean": -0.1086460948305252, "se": 0.0014622121196552456, "per_batch": [-0.10675289906091173, -0.10703733686760747, -0.10996418061565341, -0.10974638608616707, -0.10165105522089785, -0.10620087774730483, -0.11350420700751943, -0.11431181603813981]}}, "s4_arith": {"masked": 0.43398541211926617, "all_on_minus_masked": {"mean": -0.10603784356874471, "se": 0.003818014664932071, "per_batch": [-0.12301407367830386, -0.1058705850663228, -0.09127738418832365, -0.10363098399112458, -0.0923003799913183, -0.11450353369818911, -0.10509752305069447, -0.11260828488568098]}}, "s5_causal": {"masked": 0.0443593947789718, "all_on_minus_masked": {"mean": -0.0013008304164274188, "se": 0.00026591176947018226, "per_batch": [-0.0015958915108500796, -0.001163828096219982, -0.0026223592136090726, -0.0015130791772307098, -0.00173091670237982, -0.0006407616742231878, -0.0010018849326012894, -0.00013792202430520928]}}, "s6_tryfail": {"masked": 0.058309898170214744, "all_on_minus_masked": {"mean": -0.003474366640441262, "se": 0.0008470863315530741, "per_batch": [-0.0035635826405763446, -0.008703203544743784, -0.0027583544947788405, -0.004962946351610778, -0.0025744889176329763, -0.0013106421610205865, -0.0018324966427745865, -0.0020892183703922]}}, "s7_mixed": {"masked": 0.7208044529776354, "all_on_minus_masked": {"mean": -0.0008228079973903665, "se": 0.00043252837514485366, "per_batch": [-0.0013986489081416753, -0.0011429107063087862, 0.0012669961889063552, -0.0002189339145484137, -0.000549226595530361, -0.0015171019589235168, -0.00013191069321177862, -0.0028907273913647558]}}, "s8_register": {"masked": 0.021885734867835994, "all_on_minus_masked": {"mean": -0.0007025314938347883, "se": 0.00038100275628170574, "per_batch": [-0.003125393687623726, -0.0003935527369814222, -0.0008676793458920681, -0.0009558217980737173, -0.00019942597620556737, 0.0003369623831607979, 0.00014484701323136048, -0.0005601878022939641]}}, "web": {"masked": 0.952914076646663, "all_on_minus_masked": {"mean": -0.0004267749481160349, "se": 0.000354372403966084, "per_batch": [-0.001634351131149181, 0.001280152861697359, -0.0014497277554480492, 0.0006656588501645633, -0.0006089905667052697, -0.000998445279584037, -0.00018651518486068053, -0.0004819813790429839]}}}, "s3_rules": {"s1_perspective": {"masked": 0.04913599389752936, "all_on_minus_masked": {"mean": 7.567404589770878e-05, "se": 0.00011566059726850928, "per_batch": [0.0005357797020160426, -1.9068576414413285e-05, 8.817872977206348e-05, -0.00014438480985717178, -0.00024407562832136354, -6.619182275082597e-05, -0.00016698709230102832, 0.000622141865038367]}}, "s2_concept": {"masked": 0.10991907279100682, "all_on_minus_masked": {"mean": 0.0005395982543109808, "se": 8.511747905276193e-05, "per_batch": [0.0005699935006220624, 0.0007234342719387304, 0.00017896944605411136, 0.00031716825493804945, 0.0008100597833951978, 0.00037262191092563013, 0.0008384584186189037, 0.0005060804479951608]}}, "s3_rules": {"masked": 0.3137152168335505, "all_on_minus_masked": {"mean": -0.09000000454687146, "se": 0.0012519694752443542, "per_batch": [-0.08526628957102944, -0.08695049417309758, -0.09220606900515141, -0.09280689039594067, -0.08670494594332817, -0.09350556991507078, -0.09418681469296977, -0.08837296267838388]}}, "s4_arith": {"masked": 0.3280995501565347, "all_on_minus_masked": {"mean": -0.0001519816060132559, "se": 9.031846457985421e-05, "per_batch": [0.00022637216419812978, -0.0002849537546483205, -0.0005296635689947538, -0.00032646606700031944, -0.00019257757330365388, -0.0002298763107094759, -5.0584397185027274e-05, 0.00017189665953737387]}}, "s5_causal": {"masked": 0.04317212497717416, "all_on_minus_masked": {"mean": -0.00011356061462977282, "se": 9.729376613696273e-05, "per_batch": [-7.334791336406632e-05, 1.9307739788268496e-05, 1.940179280045945e-05, 0.00016872572941209474, -0.000202162918860313, -0.0002684676052257598, -0.0006873232897153733, 0.0001153815481265072]}}, "s6_tryfail": {"masked": 0.05511960252972167, "all_on_minus_masked": {"mean": -0.00028407099994818327, "se": 7.65438160477377e-05, "per_batch": [-0.00016222263542634702, -0.00027538721969407043, -0.0005262991583872348, -0.0005974999758444274, -0.0002949986163502741, -0.0003712191774295284, 5.395955670820657e-05, -9.890077316179052e-05]}}, "s7_mixed": {"masked": 0.7198716755111628, "all_on_minus_masked": {"mean": 0.00010996946908219118, "se": 0.00014303862465725415, "per_batch": [0.00022701709913897083, -5.8904057920416975e-05, 0.0007617111606713634, 0.00035918576632698596, -0.0006159558640651541, -0.00013053483200486937, 0.0002803747165691095, 5.68617639415403e-05]}}, "s8_register": {"masked": 0.021275203343297133, "all_on_minus_masked": {"mean": -9.199996929592363e-05, "se": 5.965723713550016e-05, "per_batch": [-0.0001442383558810413, -0.00039522150614057505, -0.00012014331777226994, -5.360484249080158e-05, -0.00022376017768819395, 9.633715677237129e-05, -1.045063326883533e-05, 0.00011508192210195681]}}, "web": {"masked": 0.9523970538025067, "all_on_minus_masked": {"mean": 9.024789604025751e-05, "se": 0.0001601420187726477, "per_batch": [0.00027955779897759214, 0.000465986992512768, -0.00038696096444124084, -7.188874806063339e-05, -0.00012881500549632996, 0.0009368754905750842, -0.000395302123008201, 2.2529727263020938e-05]}}}, "s4_arith": {"s1_perspective": {"masked": 0.04919292116267579, "all_on_minus_masked": {"mean": 1.874678075127357e-05, "se": 6.364014766710639e-05, "per_batch": [-0.00010080870569480205, 3.614322897038669e-05, -0.0001409881525026302, 0.00024501615844326613, 0.00030083527756392253, 7.768241361159195e-05, -8.004180060312144e-05, -0.00018786417377842507]}}, "s2_concept": {"masked": 0.1110565808235725, "all_on_minus_masked": {"mean": -0.0005979097782546922, "se": 7.583500521583865e-05, "per_batch": [-0.0007611522170560825, -0.00044788581850946363, -0.00031257846794315036, -0.0008945570095472283, -0.00046536355540340024, -0.0004032348227792082, -0.0007160282690359399, -0.0007824780657630648]}}, "s3_rules": {"masked": 0.22378718433883626, "all_on_minus_masked": {"mean": -7.197205215716401e-05, "se": 7.329844587638798e-05, "per_batch": [0.00015964289566341994, -0.00019866145957789239, -7.629580348900156e-05, -0.00010503840401446252, 0.00011337956258131165, -0.0004898495853200013, 7.165227191571755e-05, -5.0605895016403446e-05]}}, "s4_arith": {"masked": 0.4710815915836263, "all_on_minus_masked": {"mean": -0.14313402303310482, "se": 0.0023185327143662397, "per_batch": [-0.15170907800512445, -0.13356761706584924, -0.14244123391976676, -0.14094851050160295, -0.13528701361785017, -0.15089828729546306, -0.14438253108704335, -0.1458379127721386]}}, "s5_causal": {"masked": 0.043089405017048915, "all_on_minus_masked": {"mean": -3.0840654504526825e-05, "se": 0.00010553725731538536, "per_batch": [-0.00023385340951070227, 0.0003053847561250364, -0.0005311092981535523, -8.289832494479121e-05, -0.00012238984114917167, -0.0001054791095572924, 0.00013680413707460592, 0.0003868158540796529]}}, "s6_tryfail": {"masked": 0.05847575058010442, "all_on_minus_masked": {"mean": -0.003640219050330934, "se": 0.0003843096076324444, "per_batch": [-0.004188293496461978, -0.005932127708258947, -0.0033515119086889866, -0.0030860781849135904, -0.0025688672347329022, -0.002584936863672886, -0.0035355548429435446, -0.0038743821629746364]}}, "s7_mixed": {"masked": 0.7198329794147187, "all_on_minus_masked": {"mean": 0.00014866556552631943, "se": 0.00013001274668035547, "per_batch": [-0.0002925424891175865, -2.235774461123441e-06, -0.00016114774385844743, 0.0008342018480099744, -4.7037255010984325e-05, 0.0001007388377429197, 0.0004703295544470909, 0.00028701754645871214]}}, "s8_register": {"masked": 0.021212522049226322, "all_on_minus_masked": {"mean": -2.9318675225114936e-05, "se": 3.6379121072651076e-05, "per_batch": [-5.2736867549724176e-06, -0.00020882993381016526, -5.7651809243917246e-05, -8.521471627360305e-05, -6.925257649039937e-05, 6.988138805761462e-05, -8.63944097693678e-06, 0.00013043137369146002]}}, "web": {"masked": 0.9522881995334262, "all_on_minus_masked": {"mean": 0.00019910216512075696, "se": 0.00013099393473965182, "per_batch": [-0.00029271447176837384, 0.0006511263161665948, -4.660729838379396e-05, 0.0005745940365327051, -7.42965051727662e-05, 0.0006495784723088427, 6.801913841625318e-05, 6.311763286659389e-05]}}}, "s5_causal": {"s1_perspective": {"masked": 0.04852981517998504, "all_on_minus_masked": {"mean": 0.0006818527634420302, "se": 0.00017029495425816582, "per_batch": [0.0007162002516868105, -6.310688395097519e-05, 0.0010756439919900745, 0.00017163868556108186, 0.0014471372667694202, 0.0004986636961767099, 0.0008447572832067582, 0.0007638878160963619]}}, "s2_concept": {"masked": 0.11095687522534925, "all_on_minus_masked": {"mean": -0.0004982041800314398, "se": 0.00022944062832020107, "per_batch": [-0.0010421933663972704, -0.0002186329449093416, 0.00031233124288254643, -4.9369769713314415e-05, -0.0008525072514112553, 0.00016430792506805236, -0.0007612382083815039, -0.0015383310673894313]}}, "s3_rules": {"masked": 0.2232551882551392, "all_on_minus_masked": {"mean": 0.0004600240315398921, "se": 0.00020334208028465882, "per_batch": [0.0006968092078020183, 0.0012970931528072938, -0.000335215684362955, -5.512043960154145e-05, 0.000831643606078486, -4.699425934825974e-05, 0.0009595342048263067, 0.0003324424641177881]}}, "s4_arith": {"masked": 0.32771942012800737, "all_on_minus_masked": {"mean": 0.0002281484225141045, "se": 0.00015070948866851, "per_batch": [0.00042905371823998006, 0.00019661916559887693, -6.331111334889616e-05, 0.00019767255933544892, 0.000383263337447759, -7.749968204506796e-05, -0.0003207691416904912, 0.0010801585365752264]}}, "s5_causal": {"masked": 0.45805035863859445, "all_on_minus_masked": {"mean": -0.4149917942760501, "se": 0.005613757765991339, "per_batch": [-0.38852807573043924, -0.42190184657615115, -0.4169701580713347, -0.416439287936555, -0.43334118391973653, -0.43333311417129056, -0.3967670282325524, -0.4126536595703413]}}, "s6_tryfail": {"masked": 0.055308242640850665, "all_on_minus_masked": {"mean": -0.00047271111107717763, "se": 8.464371324594121e-05, "per_batch": [-0.0005654682071212364, -0.00013286197224936236, -0.0009017157873893872, -0.0005303192528511441, -0.00040570169892752833, -0.0005647372808550641, -0.0001958721159592211, -0.0004850125732644775]}}, "s7_mixed": {"masked": 0.7195940498297786, "all_on_minus_masked": {"mean": 0.00038759515046637993, "se": 0.00029919286030501683, "per_batch": [-4.7037255010984325e-05, -0.00045377622430153863, 0.0009722179253274232, 0.00206026616601096, 0.00011350854956948542, -0.00046581500986186963, 0.0001757662691818096, 0.0007456307828157538]}}, "s8_register": {"masked": 0.021440051905233473, "all_on_minus_masked": {"mean": -0.0002568485312322683, "se": 0.00010474271489733196, "per_batch": [-0.00034611911570478146, 6.704636154729673e-05, -0.00033975710123732017, -0.0007269773832693921, -0.00025979726111625273, -0.0005465393666106468, 1.0052923388716162e-05, 8.730269314423389e-05]}}, "web": {"masked": 0.952135446692713, "all_on_minus_masked": {"mean": 0.00035185500583397034, "se": 0.00012019484202317439, "per_batch": [6.363358082017712e-06, -3.8524113793236836e-05, 0.0005568798234938388, 0.0009626728882046187, 0.0006151819421362781, 0.0002047453458521309, 0.000330894620260036, 0.0001766261824360793]}}}, "s6_tryfail": {"s1_perspective": {"masked": 0.04929523471656565, "all_on_minus_masked": {"mean": -8.356677313857853e-05, "se": 7.36575248390683e-05, "per_batch": [-6.287040780603159e-05, -0.00023980293433899236, -0.00027579567848987074, 0.00030826277829806153, -0.00019131995016919562, -0.0002558295676162725, 0.0001337272599615108, -8.490568494783773e-05]}}, "s2_concept": {"masked": 0.11029371415446565, "all_on_minus_masked": {"mean": 0.00016495689085217063, "se": 0.00014514503134958106, "per_batch": [-0.00013199668453721392, 0.0009291577691176861, -0.0003691715089926967, 0.00024035650349645, 0.0005467328470928762, -4.753170513220606e-05, 5.7614188039012415e-06, 0.00014634648696856822]}}, "s3_rules": {"masked": 0.22387401407969063, "all_on_minus_masked": {"mean": -0.00015880179301152458, "se": 5.8830351672599784e-05, "per_batch": [7.152328492757154e-05, -0.0005012649337710062, -0.00013917696021070736, -0.00017935640701857714, -0.00019982234247120645, -0.0001236985216330755, -1.967051569243261e-05, -0.00017894794822276294]}}, "s4_arith": {"masked": 0.32884865821357256, "all_on_minus_masked": {"mean": -0.000901089663051094, "se": 0.00010956229330762269, "per_batch": [-0.000500684492324377, -0.0009692297267687167, -0.0007054298381765634, -0.0014258651626408159, -0.0012682860587878153, -0.0007891423934839881, -0.0008599992456394534, -0.000690080386587022]}}, "s5_causal": {"masked": 0.04313994070820977, "all_on_minus_masked": {"mean": -8.137634566538106e-05, "se": 0.00015397238476287946, "per_batch": [-0.0009440396428751566, 0.0002886782539310642, 0.00029215552815319706, 0.00021960840900720618, -8.596445314221213e-05, -0.000504290753534628, 0.00010199108641948534, -1.914919328200454e-05]}}, "s6_tryfail": {"masked": 0.4686174349109796, "all_on_minus_masked": {"mean": -0.41378190338120613, "se": 0.0032844948664612363, "per_batch": [-0.4114666433757729, -0.40629823150095945, -0.4329068605455849, -0.4057090511858505, -0.41605608103091457, -0.41664329429445424, -0.41683471635932234, -0.4043403487567902]}}, "s7_mixed": {"masked": 0.719856538350658, "all_on_minus_masked": {"mean": 0.00012510662958704205, "se": 0.00019522428815324432, "per_batch": [-0.00023535825770570895, -0.0001802378181041675, 0.0006317782679445827, 0.0006230931440759369, -0.0007464906960700235, -0.0001795068918380438, 0.0008626649767278782, 0.00022491031166588238]}}, "s8_register": {"masked": 0.021753321326912618, "all_on_minus_masked": {"mean": -0.0005701179529114088, "se": 0.0001966525694457617, "per_batch": [-0.0004037857047077577, -0.0016825331456800548, -0.0009075712592054794, -0.0005096759602897724, -0.0006866152048950261, -0.00010381034039814302, -0.0004255858493190795, 0.00015863384120404223]}}, "web": {"masked": 0.9527345375080779, "all_on_minus_masked": {"mean": -0.00024723580953098234, "se": 0.00012628248836158055, "per_batch": [0.00014016586045317858, -0.000472522333245573, -0.0006014233300671856, 0.00028643710501197184, -0.00039702194951685144, 7.833809746793374e-05, -0.0004067389692905543, -0.0006051209570607785]}}}, "s7_mixed": {"s1_perspective": {"masked": 0.04960258719606715, "all_on_minus_masked": {"mean": -0.0003909192526400859, "se": 0.00012767979675169512, "per_batch": [-0.0007706488840584086, -0.00021657452755683565, 0.0002722001661952661, -0.00031623847373183495, -0.000712550994813814, -0.0005123578147516669, -0.00014332604166279045, -0.000727857450740603]}}, "s2_concept": {"masked": 0.11098699637431052, "all_on_minus_masked": {"mean": -0.0005283253289927077, "se": 0.00014557330501007527, "per_batch": [-0.0007609587365738635, -0.00041652048255835794, -0.0008846572582069251, -0.0008734461391538134, 0.00014149872599737712, -0.0007954627559031979, -0.0006832010805525035, 4.6145095009622605e-05]}}, "s3_rules": {"masked": 0.22471775559036575, "all_on_minus_masked": {"mean": -0.001002543303686633, "se": 0.0003333943953259469, "per_batch": [0.0006149669638226829, -0.0014189643587749767, -0.000511196931858332, -0.001105074523119004, -0.0019678684868349727, -0.0017272217626151942, -0.001880071343569517, -2.4915986543749824e-05]}}, "s4_arith": {"masked": 0.33821385522823433, "all_on_minus_masked": {"mean": -0.010266286677712871, "se": 0.00045066235569934043, "per_batch": [-0.011211291035770232, -0.008932477916175297, -0.01037210169088515, -0.010032951903383658, -0.008284490283386958, -0.011981278861514344, -0.01151234666610551, -0.009803355064481822]}}, "s5_causal": {"masked": 0.04321698087410955, "all_on_minus_masked": {"mean": -0.00015841651156515892, "se": 0.00011149557059484788, "per_batch": [0.00021817880322191202, 5.702299767670199e-05, 9.805698328099743e-05, -0.00020156366681121224, -0.000709226892640108, -0.0001154245437892179, -0.0005202206465707987, -9.415512688954603e-05]}}, "s6_tryfail": {"masked": 0.055185743295123026, "all_on_minus_masked": {"mean": -0.00035021176534953333, "se": 0.00010259174161096671, "per_batch": [-0.00043126799486985223, -0.00037095045453755526, -0.0006802075075359321, 0.00012393499777801215, -0.0002224649333488829, -8.096083289366868e-05, -0.0003919000911957876, -0.0007478773061926]}}, "s7_mixed": {"masked": 0.7767741052879211, "all_on_minus_masked": {"mean": -0.05679246030767622, "se": 0.032771926955067206, "per_batch": [-0.0009523539291527161, -0.0009249226963401469, -0.03497654166259656, -4.153381018334734e-05, -0.0007876805409516852, -0.18575506453956508, -0.006918389091920885, -0.22398319619069934]}}, "s8_register": {"masked": 0.021236894039865634, "all_on_minus_masked": {"mean": -5.3690665864426e-05, "se": 6.61117614984459e-05, "per_batch": [-0.00017525435207258258, -0.00017117379495797114, 5.1312636222270586e-05, -0.0003069231946815805, 0.0002922643609244435, -3.081982848040024e-05, -0.00014608313853443564, 5.7151984664848e-05]}}, "web": {"masked": 0.953443396250356, "all_on_minus_masked": {"mean": -0.0009560945518090752, "se": 0.00018380444804242692, "per_batch": [-0.001162602719832595, -0.0008753057015662424, -0.00108168488260163, -0.00033657004773846033, -0.001350579757225745, -0.0015372669247372128, -3.379459089458692e-05, -0.0012709517898761291]}}}, "s8_register": {"s1_perspective": {"masked": 0.04849286779776042, "all_on_minus_masked": {"mean": 0.0007188001456666475, "se": 0.00017096681102505884, "per_batch": [0.000591405340654487, 0.0004065347398926403, 0.0008034384513369644, 0.00042899997366159515, 0.0010610630878716137, 0.0007815536590146735, 3.803503812987996e-05, 0.001639370874771326]}}, "s2_concept": {"masked": 0.11124442349951956, "all_on_minus_masked": {"mean": -0.000785752454201755, "se": 0.00012752336205234502, "per_batch": [-0.0010723978194550482, -0.0010509214859285437, -0.00033933251906799466, -0.0007183500348225957, -0.00106423939245473, -0.00017303604459933952, -0.0007642909004343251, -0.001103451436851463]}}, "s3_rules": {"masked": 0.22322506441894902, "all_on_minus_masked": {"mean": 0.0004901478677300786, "se": 0.00020819227082427883, "per_batch": [-0.00036987018851181164, 0.0007035810246797392, 0.0004941491515915719, 0.000354327256440079, 5.282017164623398e-05, 0.0002567271020754436, 0.0008098448050815887, 0.0016196036188377838]}}, "s4_arith": {"masked": 0.32787814936583803, "all_on_minus_masked": {"mean": 6.941918468340857e-05, "se": 9.810336671362854e-05, "per_batch": [0.00029426231562623695, 0.00022420088323099607, 0.00028020273391821116, 0.000321736544101614, -7.552188156012551e-05, 1.0899400498420864e-05, -0.0005027912797974099, 2.364761449324959e-06]}}, "s5_causal": {"masked": 0.04288413431794293, "all_on_minus_masked": {"mean": 0.00017443004460145507, "se": 4.302449341222884e-05, "per_batch": [0.00021727858153380158, -1.8074301714111896e-05, 0.00030101263467263023, 0.00027231571703881935, 7.803175337115636e-05, 6.096784973087105e-05, 0.0003047156361240255, 0.00017919248605444832]}}, "s6_tryfail": {"masked": 0.05458588796313044, "all_on_minus_masked": {"mean": 0.00024964356664304316, "se": 0.00017293105271589974, "per_batch": [0.00038492941937799674, 0.00013228690526052855, 0.00023116080613313356, -0.0003353984159295137, 3.8808960058756004e-05, -0.00026412235606255996, 0.0011628929405559166, 0.0006465902737500875]}}, "s7_mixed": {"masked": 0.7187657867576811, "all_on_minus_masked": {"mean": 0.0012158582225638426, "se": 0.00034442994337215144, "per_batch": [9.897601557162794e-05, 0.00048662491061635116, 0.001804012016225487, 0.0023164343244711505, 0.0006815672453692745, 0.0028102395107610367, 0.0005747230235207956, 0.0009542887339750172]}}, "s8_register": {"masked": 0.37574392004843205, "all_on_minus_masked": {"mean": -0.3545607166744309, "se": 0.006614908165433324, "per_batch": [-0.3205802684622582, -0.3703931393829811, -0.34871639232365426, -0.3691243978388914, -0.3497612581392122, -0.37509195262961365, -0.3643391297198388, -0.3384791948989975]}}, "web": {"masked": 0.9520328482925577, "all_on_minus_masked": {"mean": 0.0004544534059892774, "se": 8.126702235291003e-05, "per_batch": [9.854605894443758e-05, 0.0006427851575997456, 0.0007175116193997022, 0.00031146058071251925, 0.0005633291729010281, 0.000366237055012264, 0.0002464511386863766, 0.0006893064646581459]}}}}, "member_alone": {"s1_perspective": {"s1_perspective": {"alone": 0.05251879489204885, "alone_minus_off": {"mean": -0.5013145064265103, "se": 0.004723722359706749, "per_batch": [-0.5020161539899823, -0.5206146212027206, -0.511447510580648, -0.5087860038161363, -0.4873552833133265, -0.5025627202288822, -0.47857692335015567, -0.4991568349302314]}}, "s2_concept": {"alone": 0.23945795595196098, "alone_minus_off": {"mean": -0.48172597674480316, "se": 0.0029625557017275653, "per_batch": [-0.4753027767620657, -0.4804484116845556, -0.4714891474705084, -0.49853255940543073, -0.48017119714919604, -0.4800707377832608, -0.47906046869643004, -0.4887325150069781]}}, "s3_rules": {"alone": 0.508758760689348, "alone_minus_off": {"mean": -0.42085437549215554, "se": 0.004040788518956882, "per_batch": [-0.4068824458170247, -0.41901358606093464, -0.4391522815246798, -0.40577522151077006, -0.4185499638298712, -0.4321406348403284, -0.4199813324373397, -0.42533953791629586]}}, "s4_arith": {"alone": 0.7225499640202783, "alone_minus_off": {"mean": -0.5113249030739563, "se": 0.003602454892240459, "per_batch": [-0.5097063582233129, -0.5105177293744209, -0.5238398924712184, -0.49785761349211977, -0.5006369821170508, -0.5244675431555423, -0.5047830108770632, -0.5187900948809221]}}, "s5_causal": {"alone": 0.4626770842822263, "alone_minus_off": {"mean": -0.19776059832930892, "se": 0.003074781006523783, "per_batch": [-0.18829842997184976, -0.20876677318163606, -0.20486710956898452, -0.20499300086941613, -0.19802718756151977, -0.18360145478979822, -0.19417430322990303, -0.19935652746136384]}}, "s6_tryfail": {"alone": 0.49460838826503134, "alone_minus_off": {"mean": -0.1782128728751334, "se": 0.002204184674331319, "per_batch": [-0.1800673403015557, -0.18936055182791578, -0.1759819784216351, -0.1815247642806299, -0.17975192411987334, -0.17801537229844877, -0.17237782399882867, -0.16862322775217997]}}, "s7_mixed": {"alone": 0.7729224570816076, "alone_minus_off": {"mean": -0.010053471583414728, "se": 0.009004724163521997, "per_batch": [0.0017783866012468508, 0.0011675042253822543, 0.0005177537704225132, 0.001957162566818771, 0.0010375713326553626, -0.06640314643307121, 0.008738567477327486, -0.029221572208099844]}}, "s8_register": {"alone": 0.3787972650978744, "alone_minus_off": {"mean": -0.11185323267827268, "se": 0.002333843553660825, "per_batch": [-0.09853311724991193, -0.1145322762987146, -0.1086452241683552, -0.11632412054239144, -0.10945930404621429, -0.11954634449329432, -0.11680008252865426, -0.11098539209864544]}}, "web": {"alone": 0.9520279682848395, "alone_minus_off": {"mean": 0.0014312073737332287, "se": 0.0004766934609748917, "per_batch": [0.0014245322970964924, -0.0004134462926742577, 0.00041473616255571777, 1.4618525323362164e-05, 0.0028610173884282553, 0.0033452345419324736, 0.002158726233630004, 0.0016442401335737822]}}}, "s2_concept": {"s1_perspective": {"alone": 0.5265382088974727, "alone_minus_off": {"mean": -0.02729509242108652, "se": 0.0008478035966868174, "per_batch": [-0.023477093695305706, -0.029719504978010347, -0.026994955823041678, -0.02971271166330125, -0.0301825252697957, -0.02641580424626122, -0.025215322347576063, -0.02664282134540019]}}, "s2_concept": {"alone": 0.35026574371684355, "alone_minus_off": {"mean": -0.3709181889799206, "se": 0.0024118108231179045, "per_batch": [-0.37237346048863007, -0.3698774117831641, -0.3638556972371791, -0.385719013225945, -0.3722404964016817, -0.36824759669443125, -0.36423210276642376, -0.37079973324191]}}, "s3_rules": {"alone": 0.6188593181726778, "alone_minus_off": {"mean": -0.31075381800882573, "se": 0.002804570723181844, "per_batch": [-0.3059784617336435, -0.31155671901437876, -0.3166699782028437, -0.29686763780852365, -0.3055774841831569, -0.3180009949335336, -0.3101597899327456, -0.3212194782617802]}}, "s4_arith": {"alone": 0.8801855050550988, "alone_minus_off": {"mean": -0.3536893620391357, "se": 0.004028817126679558, "per_batch": [-0.3418542146862763, -0.3484509532254211, -0.36770071336251886, -0.35094263787112157, -0.33855532946875133, -0.36934916707103904, -0.36129874517547633, -0.3513631354524811]}}, "s5_causal": {"alone": 0.6589420516116804, "alone_minus_off": {"mean": -0.001495630999854905, "se": 0.000777079044912105, "per_batch": [0.0004168429500286397, -0.0005169798484936372, -0.005893587471092143, -0.0007514781929449654, -0.0015341712370217087, -0.001300575801487236, -0.003323435740935632, 0.0009383373431074427]}}, "s6_tryfail": {"alone": 0.6732533568015419, "alone_minus_off": {"mean": 0.0004320956613771132, "se": 0.001552406797931269, "per_batch": [-0.001823876012400083, -0.007744722733654008, 0.0013104218082491403, -0.002184394644271337, 0.0010933367071975786, 0.004876267095575981, 0.006227835753043975, 0.001701897317275658]}}, "s7_mixed": {"alone": 0.7829373884278554, "alone_minus_off": {"mean": -3.8540237166762026e-05, "se": 0.0010946548710171276, "per_batch": [0.000501845375217691, 0.0012001809290462795, 0.0021090232475305948, -8.031589795287619e-05, 0.000561265381090803, 0.0019007092616730725, 0.0009867504593253917, -0.007487780653265053]}}, "s8_register": {"alone": 0.469142223846714, "alone_minus_off": {"mean": -0.021508273929433143, "se": 0.0008167612527105201, "per_batch": [-0.019257628343374855, -0.022873004634150096, -0.0193475322741134, -0.022005610134523512, -0.024771349134373732, -0.024399694625859225, -0.019647040060590926, -0.019764332228479398]}}, "web": {"alone": 0.9516445652114878, "alone_minus_off": {"mean": 0.0010478043003816634, "se": 0.000252314831290859, "per_batch": [9.347257074376891e-05, 0.0012871181590572434, 0.0019100393204823929, 0.0013596088463959655, 0.0010667223919766933, -7.928400204770814e-05, 0.0010015409672995412, 0.0017432161491454101]}}}, "s3_rules": {"s1_perspective": {"alone": 0.55350150916384, "alone_minus_off": {"mean": -0.00033179215471922097, "se": 0.00018341479131836675, "per_batch": [0.0003735463176740561, -0.00032229548771689487, -0.0008996842423260043, -0.0003734173306858546, -0.0009771194308770825, -0.0004045461904920167, 0.00045029357562154093, -0.0005011144489515118]}}, "s2_concept": {"alone": 0.7242528502377963, "alone_minus_off": {"mean": 0.0030689175410321956, "se": 0.00019465440878027398, "per_batch": [0.002732804322210014, 0.003277301394841614, 0.0030090514551583958, 0.0019815841032413406, 0.0035649423784097634, 0.003502254702170249, 0.0028635111368657817, 0.0036198908353604065]}}, "s3_rules": {"alone": 0.8236434891082838, "alone_minus_off": {"mean": -0.10596964707321978, "se": 0.009346493007512462, "per_batch": [-0.11229495399364031, -0.10826746427573508, -0.09548760547277069, -0.1291266380596071, -0.08108225264517621, -0.11603213699688908, -0.14432027336743114, -0.0611458517745086]}}, "s4_arith": {"alone": 1.2339533341786908, "alone_minus_off": {"mean": 7.84670844561075e-05, "se": 0.00040067832550554967, "per_batch": [0.0005466468557673299, -0.00020706711163875902, -8.02299066273715e-05, 0.0007894003674602246, -7.782214951546074e-05, 0.0008427149892276109, -0.002416270253297048, 0.0012303638842723341]}}, "s5_causal": {"alone": 0.6675733556594478, "alone_minus_off": {"mean": 0.007135673047912558, "se": 0.0005647387818478271, "per_batch": [0.007655162768217094, 0.008555921902111074, 0.003638808926955339, 0.00856619786550028, 0.007419245566896104, 0.007850750037911025, 0.006295510926158543, 0.007103786389551003]}}, "s6_tryfail": {"alone": 0.6738381569335117, "alone_minus_off": {"mean": 0.0010168957933469058, "se": 0.00017347574800339347, "per_batch": [0.0017297155110527207, 0.0012240865175160431, 0.0009007161382312834, 0.0013980039732010008, 0.00020771204657954456, 0.0009364025382853081, 0.0012371142033188454, 0.0005014154185905006]}}, "s7_mixed": {"alone": 0.782468375615579, "alone_minus_off": {"mean": -0.0005075530494431726, "se": 0.0008395569992632177, "per_batch": [2.7775198114365907e-05, -2.235774461123441e-06, 0.0007041829639578001, 0.0003203176872319524, -0.0004014075071138157, 0.0012020727382056062, 0.0003502426684821036, -0.00626137236996227]}}, "s8_register": {"alone": 0.49080977520867747, "alone_minus_off": {"mean": 0.0001592774325303442, "se": 0.00028115811822754163, "per_batch": [0.0004771658648188293, 0.0013107227778881292, 0.0001049524126890411, -0.0004726083245709667, -0.000431418479689305, 0.0010830607438085949, 0.0002226315415420066, -0.0010202870762435756]}}, "web": {"alone": 0.9511259085321484, "alone_minus_off": {"mean": 0.0005291476210421558, "se": 0.0001432382824627441, "per_batch": [-0.00010852105269454349, 0.0007161357581927374, 0.0011528857000588921, 0.0004110385355621249, 0.0007747818421370845, 0.0008065986325466179, 0.0002391418760246955, 0.00024111967650963795]}}}, "s4_arith": {"s1_perspective": {"alone": 0.5529388195514218, "alone_minus_off": {"mean": -0.0008944817671374533, "se": 0.00022219662593559158, "per_batch": [-0.0008394043231986226, -0.0012983400270260015, 0.0003624964323560853, -0.0017084756536711598, -0.0014122355375599804, -0.0008876454567656733, -0.0007099873784243149, -0.0006622621928099592]}}, "s2_concept": {"alone": 0.7207903557747224, "alone_minus_off": {"mean": -0.0003935769220416857, "se": 0.00036656930724860673, "per_batch": [-0.000407598882544824, -0.00045631296840176194, 0.001248465058275916, -0.00226862314753129, 0.0005794525464195566, -0.0009882123118576391, -0.0004823253443447806, -0.00037346032634866244]}}, "s3_rules": {"alone": 0.9280330993212799, "alone_minus_off": {"mean": -0.001580036860223663, "se": 0.0002472548294138001, "per_batch": [-0.0006835450458542169, -0.0017735710870226962, -0.0018863057146633055, -0.0008067706151975162, -0.0008262906460703157, -0.0023852273848162797, -0.0022952804584150943, -0.00198330392974988]}}, "s4_arith": {"alone": 1.126693075548818, "alone_minus_off": {"mean": -0.10718179154541654, "se": 0.019733005800031626, "per_batch": [-0.12218541827078422, -0.000754745863311479, -0.15389351564505827, -0.15261577054047337, -0.042793669092263675, -0.11302235461546473, -0.12726140021967436, -0.14492745811630225]}}, "s5_causal": {"alone": 0.6571682010528499, "alone_minus_off": {"mean": -0.0032694815586852416, "se": 0.00036918803111562825, "per_batch": [-0.004515404498403042, -0.0033061944801866527, -0.003756875016772576, -0.0038970838728884516, -0.0014912615656313966, -0.0025041533878878974, -0.002423407533307831, -0.004261472114404086]}}, "s6_tryfail": {"alone": 0.6593261748623824, "alone_minus_off": {"mean": -0.013495086277782314, "se": 0.0007129585002580727, "per_batch": [-0.01444473685464187, -0.015171234567549097, -0.014263424144969727, -0.016344715190048764, -0.014004375277107628, -0.011502156694041865, -0.010623712309097422, -0.011606335184802141]}}, "s7_mixed": {"alone": 0.7828199027794849, "alone_minus_off": {"mean": -0.0001560258855374183, "se": 0.0002352169625457898, "per_batch": [0.00038696096444135186, 0.0003690747687516982, -0.001014955614066837, 0.0002759461633093929, 4.695126368559066e-05, -0.0010977222647946538, 0.00047183440264214616, -0.0006862967682680354]}}, "s8_register": {"alone": 0.4891133974596029, "alone_minus_off": {"mean": -0.0015371003165441724, "se": 0.0002106886696758639, "per_batch": [-0.001695619950519145, -0.0003266810453138591, -0.001291288738340668, -0.0024274921212658063, -0.0015957410260305505, -0.001409999763098746, -0.0018442129608645663, -0.0017057669269200382]}}, "web": {"alone": 0.9507103339541697, "alone_minus_off": {"mean": 0.00011357304306353067, "se": 0.00010289644243726181, "per_batch": [-0.0003639152892258579, 4.480148054974986e-05, -4.5575402478625904e-05, 0.0002953802028569097, 5.933401454827347e-06, 0.0003312385855618327, 3.989997500020159e-05, 0.0006008213907892079]}}}, "s5_causal": {"s1_perspective": {"alone": 0.5355407643694168, "alone_minus_off": {"mean": -0.0182925369491425, "se": 0.0010792101601506997, "per_batch": [-0.017861086222706102, -0.017265596294093633, -0.023324072131700513, -0.01829319263299889, -0.012617335197899093, -0.02079786196883937, -0.017867363589462615, -0.018313787555439776]}}, "s2_concept": {"alone": 0.7102577029047991, "alone_minus_off": {"mean": -0.010926229791965086, "se": 0.0004629922489753803, "per_batch": [-0.012461389929229183, -0.010230344990911466, -0.010227593268497648, -0.013156930764981145, -0.009059831069139102, -0.010504141371085018, -0.010754161149776897, -0.011015445792100231]}}, "s3_rules": {"alone": 0.9173749367374315, "alone_minus_off": {"mean": -0.01223819944407202, "se": 0.0003759474969709794, "per_batch": [-0.01179603204870372, -0.013579406146825623, -0.013483955775596801, -0.010735930988785447, -0.01172250946545983, -0.011976441849458785, -0.01136538749094329, -0.013245931786802667]}}, "s4_arith": {"alone": 1.2247629037840526, "alone_minus_off": {"mean": -0.009111963310181992, "se": 0.0003133156648215143, "per_batch": [-0.00816814402007826, -0.009321803642065918, -0.009664909030537183, -0.009344677334630846, -0.007534817908275926, -0.010003456878760808, -0.010028738328437425, -0.008829159338669568]}}, "s5_causal": {"alone": 0.09109720145841224, "alone_minus_off": {"mean": -0.5693404811531231, "se": 0.007627785122552896, "per_batch": [-0.5395056362790345, -0.5729805168000802, -0.5723726226227741, -0.5818498663344731, -0.6069291611541324, -0.5535619986868231, -0.5485124375726715, -0.579011609774995]}}, "s6_tryfail": {"alone": 0.6662963307322131, "alone_minus_off": {"mean": -0.00652493040795174, "se": 0.00038463576812808813, "per_batch": [-0.006364690947468166, -0.005736481319529085, -0.006751049972631429, -0.00707196959914147, -0.008866049617279548, -0.006194643101427477, -0.005688197190299338, -0.005526361515837408]}}, "s7_mixed": {"alone": 0.7819243246229481, "alone_minus_off": {"mean": -0.0010516040420741543, "se": 0.0008258915662142286, "per_batch": [0.0004442311871786231, -9.149477025904851e-05, 0.00015272059396620463, 8.925899579781404e-05, 0.00026373539509816357, -0.00625333218103441, -0.00026807795703237547, -0.002749873600308206]}}, "s8_register": {"alone": 0.4842518187573023, "alone_minus_off": {"mean": -0.006398679018844861, "se": 0.0003353391925517653, "per_batch": [-0.004630245913516684, -0.006516422641191788, -0.00590304651688961, -0.00766930834125068, -0.006491356169828544, -0.007350065545586593, -0.0067434827359934, -0.005885504286501586]}}, "web": {"alone": 0.9508772001211694, "alone_minus_off": {"mean": 0.0002804392100631825, "se": 0.00013042933220109836, "per_batch": [0.0004127583620707753, -5.262669116401497e-05, 0.00043288033222166344, 0.0006003914341621286, 0.0007997193265121272, -0.0001322546585135198, -0.00021454835695133845, 0.00039719393216763876]}}}, "s6_tryfail": {"s1_perspective": {"alone": 0.5518238130302154, "alone_minus_off": {"mean": -0.0020094882883437554, "se": 0.0003256805712077641, "per_batch": [-0.0010533937365346802, -0.002643029378459616, -0.0009457755927573253, -0.001739604513477322, -0.0036941873405331727, -0.0021753225594381975, -0.002429684900064233, -0.0013949082854854966]}}, "s2_concept": {"alone": 0.7223676892826529, "alone_minus_off": {"mean": 0.0011837565858887206, "se": 0.00030040574308138647, "per_batch": [0.0005063599198028568, 0.00019365246487146326, 0.0026010656116491893, 0.0005574817627718165, 0.0017863837945120142, 0.0005542570880681108, 0.0017871577164408903, 0.0014836943289934235]}}, "s3_rules": {"alone": 0.9301771102893455, "alone_minus_off": {"mean": 0.0005639741078418553, "se": 0.0001864979651283735, "per_batch": [0.0006568877349705238, -0.0003355381518332923, 0.0004389857163272781, 0.0014670980031851188, 0.0005839670910047223, 0.0010127198396056025, 0.00033837586557250443, 0.00034929676390238473]}}, "s4_arith": {"alone": 1.2374947041850164, "alone_minus_off": {"mean": 0.00361983709078198, "se": 0.000311374566884852, "per_batch": [0.004592366734664122, 0.0031974154535159283, 0.002824987023072545, 0.004043054147808256, 0.0025683029166598814, 0.005088364699752379, 0.0034874641941959883, 0.0031567415565867396]}}, "s5_causal": {"alone": 0.6506552986815429, "alone_minus_off": {"mean": -0.009782383929992228, "se": 0.00038481175604609925, "per_batch": [-0.008596939764341949, -0.009804085990748002, -0.010320377908638045, -0.011653931383428295, -0.00987790954363088, -0.010220327001498664, -0.008083184590552017, -0.00970231525709997]}}, "s6_tryfail": {"alone": 0.16377145340459293, "alone_minus_off": {"mean": -0.5090498077355718, "se": 0.004677470835012531, "per_batch": [-0.5175160259054989, -0.49329147411171215, -0.5123662633993884, -0.5249514593394076, -0.5133465000158124, -0.5087508763596975, -0.5168361354909756, -0.485339727262082]}}, "s7_mixed": {"alone": 0.7829311486823038, "alone_minus_off": {"mean": -4.477998271842226e-05, "se": 0.00020200262136100378, "per_batch": [-0.0001941684128241583, -9.149477025904851e-05, 0.0008069425978483036, 0.00044371523922603906, 4.1963766810537706e-05, -0.0005567938321683341, 0.00021566624418195568, -0.0010240706945626732]}}, "s8_register": {"alone": 0.4903608897410098, "alone_minus_off": {"mean": -0.0002896080351372926, "se": 0.00018477535886866244, "per_batch": [-0.0008486483906825493, -0.0005564928625293453, 0.0005621682900078251, 0.0001129066102913967, -0.00011226167535066667, -0.00040364328157505014, -4.794016392795086e-05, -0.0010229528073320004]}}, "web": {"alone": 0.9507240925662387, "alone_minus_off": {"mean": 0.00012733165513255373, "se": 0.00013788984803610982, "per_batch": [-0.0002956381768333127, -0.00018548328895551247, 0.0008409951627191825, 0.0002792998250011891, 0.0002971860206910648, 0.00027371038884815846, -0.00032573514073419574, 0.00013431845032385592]}}}, "s7_mixed": {"s1_perspective": {"alone": 0.4970373744855765, "alone_minus_off": {"mean": -0.056795926832982666, "se": 0.0018110331876671725, "per_batch": [-0.05812665334294931, -0.05100399185746607, -0.050193609606600575, -0.06252992515298983, -0.057888758341143376, -0.05970347627738498, -0.06299913682020625, -0.05192186326512094]}}, "s2_concept": {"alone": 0.7153484807361215, "alone_minus_off": {"mean": -0.005835451960642682, "se": 0.0007318243036266138, "per_batch": [-0.008731774162618389, -0.005678437174862827, -0.0043266965347438235, -0.007888328247124377, -0.0026645702054801657, -0.006611959003746115, -0.003970520464806948, -0.0068113298917588105]}}, "s3_rules": {"alone": 0.867875922062003, "alone_minus_off": {"mean": -0.061737214119500436, "se": 0.001966162214666222, "per_batch": [-0.060009046452305426, -0.0667788855295437, -0.06605509654338759, -0.05384596216731141, -0.06646923076666511, -0.058477798920348434, -0.06718940811715324, -0.05507228445928858]}}, "s4_arith": {"alone": 1.195411129960334, "alone_minus_off": {"mean": -0.03846373713390064, "se": 0.0014722165440657325, "per_batch": [-0.032631902183383854, -0.03863573053367908, -0.0428992664398935, -0.03808968561718973, -0.033519590635812335, -0.041296302142526775, -0.0440718441534762, -0.03656557536524363]}}, "s5_causal": {"alone": 0.5576567389956736, "alone_minus_off": {"mean": -0.10278094361586164, "se": 0.002443483866586268, "per_batch": [-0.09823747907307878, -0.10830788019868787, -0.11166214382979056, -0.10734787304157156, -0.10408054664047994, -0.08949444004682883, -0.10261293268934224, -0.10050425340711333]}}, "s6_tryfail": {"alone": 0.5926274278736829, "alone_minus_off": {"mean": -0.08019383326648177, "se": 0.003441452803867667, "per_batch": [-0.08156569587675166, -0.10006397782090459, -0.07540334251802361, -0.08326673328077305, -0.08425219387021687, -0.07515306476535555, -0.0677320133806254, -0.07411364461920344]}}, "s7_mixed": {"alone": 0.7241658592632068, "alone_minus_off": {"mean": -0.058810069401815435, "se": 0.034794470993815295, "per_batch": [-0.0003199737219301557, -0.0002995077864775819, -0.03512848833463389, 6.879306034512922e-05, -0.0005225692846466679, -0.20927880852891734, -0.00042617300883801557, -0.22457382760942496]}}, "s8_register": {"alone": 0.48266726197803556, "alone_minus_off": {"mean": -0.007983235798111513, "se": 0.00023482390733178961, "per_batch": [-0.00844305828748243, -0.007983692627027916, -0.007460822372742204, -0.00862118931811362, -0.006764765589037713, -0.008812047064908568, -0.007952219801919902, -0.007828091323659747]}}, "web": {"alone": 0.9504116968298622, "alone_minus_off": {"mean": -0.00018506408124407958, "se": 0.0001933827163928398, "per_batch": [9.8890024244902e-06, -0.0006554258824381654, 0.0008509701564691774, -0.0003986557846999972, -9.381653604556561e-05, -0.0010095381605647047, -0.00016467338820114197, -1.926205689672944e-05]}}}, "s8_register": {"s1_perspective": {"alone": 0.5365892404770283, "alone_minus_off": {"mean": -0.017244060841530906, "se": 0.0006719440752896968, "per_batch": [-0.016769340355028972, -0.017993426872545015, -0.015500323369975089, -0.020143038025679294, -0.017358079964595152, -0.016784216854328626, -0.01917597957987771, -0.014228081710217388]}}, "s2_concept": {"alone": 0.713278524422626, "alone_minus_off": {"mean": -0.007905408274138243, "se": 0.00106024374276821, "per_batch": [-0.008583009169622069, -0.008665345863722584, -0.009178112137270045, -0.01224963629035436, -0.007087061076754608, -0.008324046293085363, -0.0015420824389613674, -0.0076139729233355435]}}, "s3_rules": {"alone": 0.9248188295679793, "alone_minus_off": {"mean": -0.004794306613524119, "se": 0.0002440026230490317, "per_batch": [-0.00465428048897476, -0.005635613494797909, -0.0055971753723301765, -0.0043165065626802335, -0.003674753300985545, -0.005269806396412724, -0.004904171280678438, -0.004302146011333163]}}, "s4_arith": {"alone": 1.2273313034409532, "alone_minus_off": {"mean": -0.006543563653281098, "se": 0.00034973144457505715, "per_batch": [-0.00574215674700751, -0.00602214450261207, -0.00784361275790002, -0.004937621906271339, -0.0062069398609641, -0.006951366765223899, -0.007616939624062846, -0.007027727062207001]}}, "s5_causal": {"alone": 0.6415767644991226, "alone_minus_off": {"mean": -0.01886091811241275, "se": 0.000764795110381384, "per_batch": [-0.018075548588332047, -0.020532191768919184, -0.01738856388946053, -0.01865719391355003, -0.017148648091506957, -0.01681229302208198, -0.018939761408917732, -0.02333314421653354]}}, "s6_tryfail": {"alone": 0.6508539493922052, "alone_minus_off": {"mean": -0.02196731174795953, "se": 0.0006708837435624013, "per_batch": [-0.022805845408988223, -0.02028685851746337, -0.02500472959159461, -0.01964123564612441, -0.02164307070650473, -0.021333587926277042, -0.02423523621597168, -0.020787929970752184]}}, "s7_mixed": {"alone": 0.7828018875968069, "alone_minus_off": {"mean": -0.000174041068215329, "se": 0.00036676702214151854, "per_batch": [0.0001026736425651098, 0.00018350548847068104, 0.0006153539247870654, 0.0007295504049600332, 0.000742707077751037, -0.001697683742329481, -0.0002445163338641865, -0.0018239190080628909]}}, "s8_register": {"alone": 0.06783037737981043, "alone_minus_off": {"mean": -0.4228201203963367, "se": 0.011232764742466362, "per_batch": [-0.35262292986788246, -0.4407578201873516, -0.4154033155051069, -0.4361712471182384, -0.4287622506424421, -0.4492752555968743, -0.4491613224651355, -0.4104068217876622]}}, "web": {"alone": 0.9510764527711095, "alone_minus_off": {"mean": 0.00047969186000340613, "se": 0.00014252589789469095, "per_batch": [0.00020637918103538766, 0.000561265381090803, 0.0011725777135825899, 0.0009258686009199213, 0.0004278928353467215, -4.806915091615238e-05, 0.00017705613906326967, 0.0004145641799047084]}}}}, "battery": {"s1_perspective": {"acc_off": {"in": 0.55, "transfer": 0.4625}, "acc_all_on": {"in": 0.8875, "transfer": 0.4375}, "acc_member_masked": {"in": 0.5625, "transfer": 0.45}, "acc_member_alone": {"in": 0.8375, "transfer": 0.4625}, "margin_all_on_minus_off": {"in": {"mean": 2.14360880900058, "se": 0.18661705138647003, "per_batch": null}, "transfer": {"mean": -0.22658184878528118, "se": 0.185168445674904, "per_batch": null}}}, "s2_concept": {"acc_off": {"in": 0.2976190476190476, "transfer": 0.6779661016949152}, "acc_all_on": {"in": 0.9761904761904762, "transfer": 0.7203389830508474}, "acc_member_masked": {"in": 0.6785714285714286, "transfer": 0.6949152542372882}, "acc_member_alone": {"in": 0.9642857142857143, "transfer": 0.7203389830508474}, "margin_all_on_minus_off": {"in": {"mean": 4.041511316916773, "se": 0.31925993285782656, "per_batch": null}, "transfer": {"mean": 0.28084098636093785, "se": 0.22139000195955616, "per_batch": null}}}, "s3_rules": {"acc_off": {"in": 0.155, "transfer": 0.07, "held-out": 0.05}, "acc_all_on": {"in": 0.855, "transfer": 0.31, "held-out": 0.43333333333333335}, "acc_member_masked": {"in": 0.81, "transfer": 0.25, "held-out": 0.425}, "acc_member_alone": {"in": 0.32, "transfer": 0.08, "held-out": 0.10833333333333334}, "margin_all_on_minus_off": {"in": {"mean": 3.6816665181398274, "se": 0.18325546279257193, "per_batch": null}, "transfer": {"mean": 0.8239017653465271, "se": 0.19632172126941658, "per_batch": null}, "held-out": {"mean": 1.7243452899424787, "se": 0.34806204515730227, "per_batch": null}}}, "s4_arith": {"acc_off": {"in": 0.34375, "transfer": 0.24166666666666667, "held-out": 0.35}, "acc_all_on": {"in": 0.64375, "transfer": 0.24166666666666667, "held-out": 0.425}, "acc_member_masked": {"in": 0.51875, "transfer": 0.23333333333333334, "held-out": 0.3}, "acc_member_alone": {"in": 0.38125, "transfer": 0.24583333333333332, "held-out": 0.3875}, "margin_all_on_minus_off": {"in": {"mean": 1.8523645000241231, "se": 0.24326049955897172, "per_batch": null}, "transfer": {"mean": -0.43729203070203465, "se": 0.08219488957325416, "per_batch": null}, "held-out": {"mean": -0.15771321058273316, "se": 0.1294458657229396, "per_batch": null}}}}, "probes_all_on": {"p0_grounding": {"acc": 0.4666666666666667, "n": 30, "families": {"count": 0.125, "permanence": 0.5714285714285714, "sequence": 0.75, "attribute": 0.42857142857142855}}, "p1_self_other": {"acc": 0.5, "n": 30, "families": {"belief": 0.25, "reality": 0.5, "perspective": 0.8, "reference": 0.8571428571428571, "second_order": 0.0}}, "p2_difference": {"acc": 0.5666666666666667, "n": 30, "families": {"category": 1.0, "antonym": 0.6666666666666666, "odd_one_out": 0.3333333333333333, "property": 0.5, "analogy": 0.3333333333333333}}, "p3_logic": {"acc": 0.23333333333333334, "n": 30, "families": {"d1": 0.5, "d2": 0.5, "d3": 0.25, "d4": 0.25, "d5": 0.0, "d6": 0.0, "negation": 0.25, "contrapositive": 0.0}}, "p4_arithmetic": {"acc": 0.3333333333333333, "n": 30, "families": {"add2": 0.4, "add3": 0.0, "sub2": 0.5, "sub3": 0.4, "mul1": 0.25, "mul2": 0.3333333333333333, "word": 0.4}}, "p5_causal": {"acc": 0.5, "n": 30, "families": {"effect": 0.42857142857142855, "cause": 0.42857142857142855, "counterfactual": 0.6666666666666666, "prevention": 0.4, "chain": 0.6}}, "p6_learning": {"acc": 0.3333333333333333, "n": 30, "families": {"error_detect": 0.125, "correction": 0.25, "strategy": 0.7142857142857143, "check": 0.2857142857142857}}, "p7_composite": {"acc": 0.36666666666666664, "n": 30, "families": {"logic_arith": 0.375, "causal_category": 0.375, "belief_causal": 0.2857142857142857, "three_step": 0.42857142857142855}}, "p8_articulation": {"acc": 0.5, "n": 30, "families": {"definition": 0.5, "register": 0.5714285714285714, "grammar": 0.875, "paraphrase": 0.0}}}, "probes_mean_off": 0.41111111111111115, "probes_mean_all_on": 0.4222222222222222, "gates": {"s1_perspective": {"mean": 0.2099039478571495, "max": 0.5712027580899203}, "s2_concept": {"mean": 0.27674460361666475, "max": 0.4977012170865557}, "s3_rules": {"mean": 0.12033936123137423, "max": 0.1738736117222015}, "s4_arith": {"mean": 0.10767731599767934, "max": 0.15932411779931022}, "s5_causal": {"mean": 0.18518384792916404, "max": 0.3681551185947585}, "s6_tryfail": {"mean": 0.23037076111374397, "max": 0.5373235543789295}, "s7_mixed": {"mean": 0.2511517146927655, "max": 0.463826973625139}, "s8_register": {"mean": 0.20274797688933244, "max": 0.4520048633318294}}, "chain_proof": "compiled chain: 28 cache entries (limit 256), 0 limit warnings", "curve": [{"step": 1, "train_bpb": {"s8_register": 0.3284731475315582}, "quiet_kl": {"s8_register": 0.00012053394311806187}}, {"step": 50, "train_bpb": {"s8_register": 0.37337323304432213}, "quiet_kl": {"s8_register": 0.00015407804210553877}}, {"step": 100, "train_bpb": {"s8_register": 0.35240036410970454}, "quiet_kl": {"s8_register": 0.00029328941920539367}}, {"step": 150, "train_bpb": {"s8_register": 0.33517199647008644}, "quiet_kl": {"s8_register": 0.00045233469165395946}}, {"step": 200, "train_bpb": {"s8_register": 0.30011607879472396}, "quiet_kl": {"s8_register": 0.000573032955871895}}, {"step": 250, "train_bpb": {"s8_register": 0.2689926102147035}, "quiet_kl": {"s8_register": 0.0007519861130276695}}, {"step": 300, "train_bpb": {"s8_register": 0.23407220755895825}, "quiet_kl": {"s8_register": 0.0009210047998931259}}, {"step": 350, "train_bpb": {"s8_register": 0.19826027177862454}, "quiet_kl": {"s8_register": 0.0008790462068282068}}, {"step": 400, "train_bpb": {"s8_register": 0.1661887133643011}, "quiet_kl": {"s8_register": 0.0008035184373147786}}, {"step": 450, "train_bpb": {"s8_register": 0.14119757384177767}, "quiet_kl": {"s8_register": 0.0007913070730865002}}, {"step": 500, "train_bpb": {"s8_register": 0.11734982055759266}, "quiet_kl": {"s8_register": 0.0007852105260826647}}, {"step": 550, "train_bpb": {"s8_register": 0.09817307930954992}, "quiet_kl": {"s8_register": 0.0007539926707977429}}, {"step": 600, "train_bpb": {"s8_register": 0.08123581756646907}, "quiet_kl": {"s8_register": 0.0007763803843408823}}, {"step": 650, "train_bpb": {"s8_register": 0.06934630391793221}, "quiet_kl": {"s8_register": 0.0007571315142558887}}, {"step": 700, "train_bpb": {"s8_register": 0.05929549785062752}, "quiet_kl": {"s8_register": 0.000765610738308169}}, {"step": 750, "train_bpb": {"s8_register": 0.04965903749151648}, "quiet_kl": {"s8_register": 0.0007589922321494669}}, {"step": 800, "train_bpb": {"s8_register": 0.04369733709771878}, "quiet_kl": {"s8_register": 0.0007505005429266021}}, {"step": 850, "train_bpb": {"s8_register": 0.04038011944326968}, "quiet_kl": {"s8_register": 0.0006916806881781668}}, {"step": 900, "train_bpb": {"s8_register": 0.03612310581785546}, "quiet_kl": {"s8_register": 0.0007369291270151734}}, {"step": 950, "train_bpb": {"s8_register": 0.03344802682747258}, "quiet_kl": {"s8_register": 0.0005980553483823315}}, {"step": 1000, "train_bpb": {"s8_register": 0.032557015294017486}, "quiet_kl": {"s8_register": 0.0006548272771760822}}, {"step": 1050, "train_bpb": {"s8_register": 0.030075962185836826}, "quiet_kl": {"s8_register": 0.0005793917458504439}}, {"step": 1100, "train_bpb": {"s8_register": 0.029520088084638546}, "quiet_kl": {"s8_register": 0.0005496665462851524}}, {"step": 1150, "train_bpb": {"s8_register": 0.028969627889800273}, "quiet_kl": {"s8_register": 0.0005668047216022388}}, {"step": 1200, "train_bpb": {"s8_register": 0.028590518363327468}, "quiet_kl": {"s8_register": 0.0005481418850831688}}, {"step": 1250, "train_bpb": {"s8_register": 0.02929263764296404}, "quiet_kl": {"s8_register": 0.0006800263409968466}}, {"step": 1300, "train_bpb": {"s8_register": 0.027195222525687134}, "quiet_kl": {"s8_register": 0.0005403092427877709}}, {"step": 1350, "train_bpb": {"s8_register": 0.026387788299307977}, "quiet_kl": {"s8_register": 0.000497638585511595}}, {"step": 1400, "train_bpb": {"s8_register": 0.026319035547396646}, "quiet_kl": {"s8_register": 0.0004816311597824097}}, {"step": 1450, "train_bpb": {"s8_register": 0.025496958986845918}, "quiet_kl": {"s8_register": 0.00046699840284418313}}, {"step": 1500, "train_bpb": {"s8_register": 0.025136793833066856}, "quiet_kl": {"s8_register": 0.0004553849680814892}}, {"step": 1550, "train_bpb": {"s8_register": 0.025790678912187025}, "quiet_kl": {"s8_register": 0.00044847407552879305}}, {"step": 1600, "train_bpb": {"s8_register": 0.025377687056795457}, "quiet_kl": {"s8_register": 0.0004364796355366707}}, {"step": 1650, "train_bpb": {"s8_register": 0.02528682456462896}, "quiet_kl": {"s8_register": 0.0004172971667139791}}, {"step": 1700, "train_bpb": {"s8_register": 0.025200664158753928}, "quiet_kl": {"s8_register": 0.00038426660641562195}}, {"step": 1750, "train_bpb": {"s8_register": 0.02489323697428845}, "quiet_kl": {"s8_register": 0.00040256197680719195}}, {"step": 1800, "train_bpb": {"s8_register": 0.024346336067898943}, "quiet_kl": {"s8_register": 0.00041426937328651545}}, {"step": 1850, "train_bpb": {"s8_register": 0.02441436786316339}, "quiet_kl": {"s8_register": 0.00037298401701264084}}, {"step": 1900, "train_bpb": {"s8_register": 0.024554263334642678}, "quiet_kl": {"s8_register": 0.0003680531258578412}}, {"step": 1950, "train_bpb": {"s8_register": 0.02413660440755247}, "quiet_kl": {"s8_register": 0.0004078659042716026}}, {"step": 2000, "train_bpb": {"s8_register": 0.024424851334369627}, "quiet_kl": {"s8_register": 0.00039103347837226465}}, {"step": 2050, "train_bpb": {"s8_register": 0.02406696051678678}, "quiet_kl": {"s8_register": 0.00044414905074518173}}, {"step": 2100, "train_bpb": {"s8_register": 0.02395335192168465}, "quiet_kl": {"s8_register": 0.0003821491810958833}}, {"step": 2150, "train_bpb": {"s8_register": 0.02370573331639407}, "quiet_kl": {"s8_register": 0.00038368793379049747}}, {"step": 2200, "train_bpb": {"s8_register": 0.0243890238816696}, "quiet_kl": {"s8_register": 0.0003926025703549385}}, {"step": 2250, "train_bpb": {"s8_register": 0.02358709613668673}, "quiet_kl": {"s8_register": 0.00034730643819784744}}, {"step": 2300, "train_bpb": {"s8_register": 0.023897751462381187}, "quiet_kl": {"s8_register": 0.00040502122486941517}}, {"step": 2350, "train_bpb": {"s8_register": 0.023572393581715066}, "quiet_kl": {"s8_register": 0.0003809566324343905}}, {"step": 2400, "train_bpb": {"s8_register": 0.023607916974462823}, "quiet_kl": {"s8_register": 0.00035115331440465524}}, {"step": 2450, "train_bpb": {"s8_register": 0.023209809411334752}, "quiet_kl": {"s8_register": 0.0004396457053371705}}, {"step": 2500, "train_bpb": {"s8_register": 0.023777857492580375}, "quiet_kl": {"s8_register": 0.0003461380413500592}}, {"step": 2550, "train_bpb": {"s8_register": 0.023503733592946056}, "quiet_kl": {"s8_register": 0.0003935579551034607}}, {"step": 2600, "train_bpb": {"s8_register": 0.02360184980275244}, "quiet_kl": {"s8_register": 0.00032782236259663476}}, {"step": 2650, "train_bpb": {"s8_register": 0.0230264229133607}, "quiet_kl": {"s8_register": 0.0003733018494676799}}, {"step": 2700, "train_bpb": {"s8_register": 0.022939626064188316}, "quiet_kl": {"s8_register": 0.00032826498412759977}}, {"step": 2750, "train_bpb": {"s8_register": 0.023352673249623354}, "quiet_kl": {"s8_register": 0.0003341206710319966}}, {"step": 2800, "train_bpb": {"s8_register": 0.02305001608459643}, "quiet_kl": {"s8_register": 0.0003360060328850523}}, {"step": 2850, "train_bpb": {"s8_register": 0.023429324949710492}, "quiet_kl": {"s8_register": 0.0003471303329570219}}, {"step": 2900, "train_bpb": {"s8_register": 0.022994468294979788}, "quiet_kl": {"s8_register": 0.0003480083469185047}}, {"step": 2950, "train_bpb": {"s8_register": 0.023111641633605396}, "quiet_kl": {"s8_register": 0.000336370020522736}}, {"step": 3000, "train_bpb": {"s8_register": 0.023008007360446076}, "quiet_kl": {"s8_register": 0.00031759764038724825}}, {"step": 3050, "train_bpb": {"s8_register": 0.022953830810002594}, "quiet_kl": {"s8_register": 0.00034626718465005977}}, {"step": 3100, "train_bpb": {"s8_register": 0.02277677300485741}, "quiet_kl": {"s8_register": 0.0003643619839567691}}, {"step": 3150, "train_bpb": {"s8_register": 0.023491181788512585}, "quiet_kl": {"s8_register": 0.00034642097278265283}}, {"step": 3200, "train_bpb": {"s8_register": 0.02317441159279441}, "quiet_kl": {"s8_register": 0.00036094941169722005}}, {"step": 3250, "train_bpb": {"s8_register": 0.02257360127338834}, "quiet_kl": {"s8_register": 0.00035801238467684013}}, {"step": 3300, "train_bpb": {"s8_register": 0.02268269653315834}, "quiet_kl": {"s8_register": 0.0003430224471958354}}, {"step": 3350, "train_bpb": {"s8_register": 0.022990105283233423}, "quiet_kl": {"s8_register": 0.00034256197628565135}}, {"step": 3400, "train_bpb": {"s8_register": 0.023069682542573185}, "quiet_kl": {"s8_register": 0.0003270630422048271}}, {"step": 3450, "train_bpb": {"s8_register": 0.022545602309721847}, "quiet_kl": {"s8_register": 0.00037098663859069347}}, {"step": 3500, "train_bpb": {"s8_register": 0.022929804833677063}, "quiet_kl": {"s8_register": 0.000317269783699885}}, {"step": 3550, "train_bpb": {"s8_register": 0.022755475183948052}, "quiet_kl": {"s8_register": 0.000319466183136683}}, {"step": 3600, "train_bpb": {"s8_register": 0.022302778365759002}, "quiet_kl": {"s8_register": 0.0003178187366575003}}, {"step": 3650, "train_bpb": {"s8_register": 0.021981639568858315}, "quiet_kl": {"s8_register": 0.00031899034918751566}}, {"step": 3700, "train_bpb": {"s8_register": 0.022191135693589974}, "quiet_kl": {"s8_register": 0.00032774368533864615}}, {"step": 3750, "train_bpb": {"s8_register": 0.022632202937438516}, "quiet_kl": {"s8_register": 0.0003055658849189058}}, {"step": 3800, "train_bpb": {"s8_register": 0.022115721650526384}, "quiet_kl": {"s8_register": 0.0003529132701805793}}, {"step": 3850, "train_bpb": {"s8_register": 0.02211923259572688}, "quiet_kl": {"s8_register": 0.00031297659006668255}}, {"step": 3900, "train_bpb": {"s8_register": 0.022480119781044382}, "quiet_kl": {"s8_register": 0.00030225280934246255}}, {"step": 3950, "train_bpb": {"s8_register": 0.02282011695931339}, "quiet_kl": {"s8_register": 0.00029369181400397794}}, {"step": 4000, "train_bpb": {"s8_register": 0.022593154410921558}, "quiet_kl": {"s8_register": 0.0003139365094830282}}], "steps_done": {"s1_perspective": 3200, "s2_concept": 2400, "s3_rules": 800, "s4_arith": 800, "s5_causal": 4000, "s6_tryfail": 4000, "s7_mixed": 4000, "s8_register": 4000}, "stage_seconds": 3300.242875099182, "inert_at_birth": true}], "seconds": 12223.74996805191, "log": ["[ 0.2s] THE JOB: gXA: the group ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail', 's7_mixed', 's8_register'], staged, EARLIER ARMS FROZEN (on, not training); 4000 steps per stage; each step = one task chunk + one abstention chunk of 2 x 4096 bytes per newest member; lambda 2.0; pure Adam lr 0.001; the compiled chain with recompute; trunk step 245,674; card 31.4 GB with 16.4 GB free; host RSS 0.6 GB; the first 4 members loaded from gRA:4 (the later members attach fresh); a member's partner rows come from every other member of the roster, attached or not", "[ 15.7s] mini-beatrix-3 @ 245,674: trunk loaded from /root/v3/runs/mission/mini-beatrix-3/checkpoints/step_00245674.safetensors and frozen (376.0M); amoe-lora 0.2.11, torch 2.8.0+cu128", "[ 82.1s] ARMS OFF (the bare trunk): s1_perspective rows 0.5538; s2_concept rows 0.7212; s3_rules rows 0.9296; s4_arith rows 1.2339; s5_causal rows 0.6604; s6_tryfail rows 0.6728; s7_mixed rows 0.7830; s8_register rows 0.4907; web rows 0.9506; s1_perspective battery in 0.550, transfer 0.463; s2_concept battery in 0.298, transfer 0.678; s3_rules battery in 0.155, transfer 0.070, held-out 0.050; s4_arith battery in 0.344, transfer 0.242, held-out 0.350; probes mean 0.411", "[ 86.8s] RESUMED from gRA at its close 4: ['s1_perspective', 's2_concept', 's3_rules', 's4_arith'] loaded (on; frozen while the later arms train); the next stage attaches over them", "[ 223.4s] CLOSE 4/8: the group ['s1_perspective', 's2_concept', 's3_rules', 's4_arith'] all on, against arms off:", "[ 223.4s] s1_perspective rows: off 0.5538 -> all on 0.0482 (-0.5057 +- 0.0048); with s1_perspective masked 0.5249 (its share -0.4767); s1_perspective alone 0.0525 (-0.5013); gate mean 0.210 / widest 0.571", "[ 223.4s] s2_concept rows: off 0.7212 -> all on 0.1123 (-0.6089 +- 0.0038); with s2_concept masked 0.2405 (its share -0.1283); s2_concept alone 0.3503 (-0.3709); gate mean 0.277 / widest 0.498", "[ 223.4s] s3_rules rows: off 0.9296 -> all on 0.2236 (-0.7060 +- 0.0052); with s3_rules masked 0.3189 (its share -0.0952); s3_rules alone 0.8236 (-0.1060); gate mean 0.120 / widest 0.174", "[ 223.4s] s4_arith rows: off 1.2339 -> all on 0.3384 (-0.8955 +- 0.0059); with s4_arith masked 0.4861 (its share -0.1477); s4_arith alone 1.1267 (-0.1072); gate mean 0.108 / widest 0.159", "[ 223.4s] web holdout: off 0.9506 -> all on 0.9523 (+0.00171 +- 0.00035; the quiet bar +.012); each member's share: s1_perspective +0.00035, s2_concept -0.00010, s3_rules +0.00040, s4_arith +0.00010; alone: s1_perspective +0.00143, s2_concept +0.00105, s3_rules +0.00053, s4_arith +0.00011", "[ 223.4s] a member's share on a PARTNER's rows (all on minus that member masked; near 0 = it stays out): s1_perspective on s2_concept -0.2405; s1_perspective on s3_rules -0.1848; s1_perspective on s4_arith -0.1575; s2_concept on s1_perspective -0.0043; s2_concept on s3_rules -0.1064; s2_concept on s4_arith -0.1204; s3_rules on s1_perspective +0.0001; s3_rules on s2_concept +0.0007; s3_rules on s4_arith +0.0000; s4_arith on s1_perspective -0.0001; s4_arith on s2_concept -0.0006; s4_arith on s3_rules +0.0000", "[ 223.4s] a member ALONE on a partner's rows (alone minus off): s1_perspective on s2_concept -0.4817; s1_perspective on s3_rules -0.4209; s1_perspective on s4_arith -0.5113; s2_concept on s1_perspective -0.0273; s2_concept on s3_rules -0.3108; s2_concept on s4_arith -0.3537; s3_rules on s1_perspective -0.0003; s3_rules on s2_concept +0.0031; s3_rules on s4_arith +0.0001; s4_arith on s1_perspective -0.0009; s4_arith on s2_concept -0.0004; s4_arith on s3_rules -0.0016", "[ 223.4s] the group on the rows of stages not yet attached: s5_causal -0.1818; s6_tryfail -0.1849; s7_mixed -0.0082; s8_register -0.1085", "[ 223.4s] s1_perspective battery accuracy, off -> all on: in 0.550 -> 0.963, transfer 0.463 -> 0.475; alone: in 0.838, transfer 0.463", "[ 223.4s] s2_concept battery accuracy, off -> all on: in 0.298 -> 0.976, transfer 0.678 -> 0.780; alone: in 0.964, transfer 0.720", "[ 223.4s] s3_rules battery accuracy, off -> all on: in 0.155 -> 0.905, transfer 0.070 -> 0.300, held-out 0.050 -> 0.458; alone: in 0.320, transfer 0.080, held-out 0.108", "[ 223.4s] s4_arith battery accuracy, off -> all on: in 0.344 -> 0.613, transfer 0.242 -> 0.237, held-out 0.350 -> 0.350; alone: in 0.381, transfer 0.246, held-out 0.388", "[ 223.4s] probes mean: off 0.411 -> all on 0.419", "[ 223.4s] compiled chain: 8 cache entries (limit 256), 0 limit warnings; card peak 3.0 GB; host RSS 5.3 GB", "[ 226.9s] STAGE 5/8: s5_causal attached over ['s1_perspective', 's2_concept', 's3_rules', 's4_arith'] (on, frozen); at birth the group reads exactly as before on its rows and on web text; 4000 steps of 1 task + 1 abstention chunks", "[ 237.0s] stage 5/8 s5_causal step 1/4000: train bpb s5 0.4826; quiet KL s5 0.00013; 10.03 s/step; the job: 3.9 min spent, about 2674.3 min of steps left; card peak 13.9 GB", "[ 269.8s] stage 5/8 s5_causal step 50/4000: train bpb s5 0.4627; quiet KL s5 0.00023; 0.86 s/step; the job: 4.5 min spent, about 228.1 min of steps left; card peak 14.0 GB", "[ 301.3s] stage 5/8 s5_causal step 100/4000: train bpb s5 0.4353; quiet KL s5 0.00058; 0.74 s/step; the job: 5.0 min spent, about 197.2 min of steps left; card peak 14.0 GB", "[ 332.7s] stage 5/8 s5_causal step 150/4000: train bpb s5 0.4018; quiet KL s5 0.00065; 0.71 s/step; the job: 5.5 min spent, about 186.4 min of steps left; card peak 14.0 GB", "[ 364.3s] stage 5/8 s5_causal step 200/4000: train bpb s5 0.3648; quiet KL s5 0.00068; 0.69 s/step; the job: 6.1 min spent, about 180.9 min of steps left; card peak 14.0 GB", "[ 395.7s] stage 5/8 s5_causal step 250/4000: train bpb s5 0.3277; quiet KL s5 0.00077; 0.68 s/step; the job: 6.6 min spent, about 177.2 min of steps left; card peak 14.0 GB", "[ 427.1s] stage 5/8 s5_causal step 300/4000: train bpb s5 0.2895; quiet KL s5 0.00074; 0.67 s/step; the job: 7.1 min spent, about 174.6 min of steps left; card peak 14.0 GB", "[ 458.6s] stage 5/8 s5_causal step 350/4000: train bpb s5 0.2472; quiet KL s5 0.00088; 0.66 s/step; the job: 7.6 min spent, about 172.7 min of steps left; card peak 14.0 GB", "[ 490.0s] stage 5/8 s5_causal step 400/4000: train bpb s5 0.2082; quiet KL s5 0.00084; 0.66 s/step; the job: 8.2 min spent, about 171.0 min of steps left; card peak 14.0 GB", "[ 521.3s] stage 5/8 s5_causal step 450/4000: train bpb s5 0.1748; quiet KL s5 0.00085; 0.65 s/step; the job: 8.7 min spent, about 169.5 min of steps left; card peak 14.0 GB", "[ 552.8s] stage 5/8 s5_causal step 500/4000: train bpb s5 0.1429; quiet KL s5 0.00086; 0.65 s/step; the job: 9.2 min spent, about 168.4 min of steps left; card peak 14.0 GB", "[ 584.2s] stage 5/8 s5_causal step 550/4000: train bpb s5 0.1156; quiet KL s5 0.00079; 0.65 s/step; the job: 9.7 min spent, about 167.3 min of steps left; card peak 14.0 GB", "[ 615.6s] stage 5/8 s5_causal step 600/4000: train bpb s5 0.0952; quiet KL s5 0.00074; 0.65 s/step; the job: 10.3 min spent, about 166.2 min of steps left; card peak 14.0 GB", "[ 647.1s] stage 5/8 s5_causal step 650/4000: train bpb s5 0.0807; quiet KL s5 0.00072; 0.65 s/step; the job: 10.8 min spent, about 165.4 min of steps left; card peak 14.0 GB", "[ 678.5s] stage 5/8 s5_causal step 700/4000: train bpb s5 0.0719; quiet KL s5 0.00073; 0.65 s/step; the job: 11.3 min spent, about 164.5 min of steps left; card peak 14.0 GB", "[ 709.9s] stage 5/8 s5_causal step 750/4000: train bpb s5 0.0631; quiet KL s5 0.00070; 0.64 s/step; the job: 11.8 min spent, about 163.7 min of steps left; card peak 14.0 GB", "[ 741.5s] stage 5/8 s5_causal step 800/4000: train bpb s5 0.0586; quiet KL s5 0.00065; 0.64 s/step; the job: 12.4 min spent, about 163.0 min of steps left; card peak 14.0 GB", "[ 772.9s] stage 5/8 s5_causal step 850/4000: train bpb s5 0.0560; quiet KL s5 0.00061; 0.64 s/step; the job: 12.9 min spent, about 162.2 min of steps left; card peak 14.0 GB", "[ 804.3s] stage 5/8 s5_causal step 900/4000: train bpb s5 0.0534; quiet KL s5 0.00058; 0.64 s/step; the job: 13.4 min spent, about 161.4 min of steps left; card peak 14.0 GB", "[ 835.7s] stage 5/8 s5_causal step 950/4000: train bpb s5 0.0519; quiet KL s5 0.00055; 0.64 s/step; the job: 13.9 min spent, about 160.7 min of steps left; card peak 14.0 GB", "[ 867.3s] stage 5/8 s5_causal step 1000/4000: train bpb s5 0.0505; quiet KL s5 0.00052; 0.64 s/step; the job: 14.5 min spent, about 160.1 min of steps left; card peak 14.0 GB", "[ 898.6s] stage 5/8 s5_causal step 1050/4000: train bpb s5 0.0496; quiet KL s5 0.00049; 0.64 s/step; the job: 15.0 min spent, about 159.4 min of steps left; card peak 14.0 GB", "[ 930.1s] stage 5/8 s5_causal step 1100/4000: train bpb s5 0.0497; quiet KL s5 0.00047; 0.64 s/step; the job: 15.5 min spent, about 158.7 min of steps left; card peak 14.0 GB", "[ 961.6s] stage 5/8 s5_causal step 1150/4000: train bpb s5 0.0495; quiet KL s5 0.00049; 0.64 s/step; the job: 16.0 min spent, about 158.1 min of steps left; card peak 14.0 GB", "[ 993.0s] stage 5/8 s5_causal step 1200/4000: train bpb s5 0.0479; quiet KL s5 0.00048; 0.64 s/step; the job: 16.6 min spent, about 157.5 min of steps left; card peak 14.0 GB", "[ 1024.6s] stage 5/8 s5_causal step 1250/4000: train bpb s5 0.0476; quiet KL s5 0.00045; 0.64 s/step; the job: 17.1 min spent, about 156.9 min of steps left; card peak 14.0 GB", "[ 1056.0s] stage 5/8 s5_causal step 1300/4000: train bpb s5 0.0471; quiet KL s5 0.00041; 0.64 s/step; the job: 17.6 min spent, about 156.2 min of steps left; card peak 14.0 GB", "[ 1087.4s] stage 5/8 s5_causal step 1350/4000: train bpb s5 0.0469; quiet KL s5 0.00044; 0.64 s/step; the job: 18.1 min spent, about 155.6 min of steps left; card peak 14.0 GB", "[ 1118.8s] stage 5/8 s5_causal step 1400/4000: train bpb s5 0.0470; quiet KL s5 0.00042; 0.64 s/step; the job: 18.6 min spent, about 155.0 min of steps left; card peak 14.0 GB", "[ 1150.4s] stage 5/8 s5_causal step 1450/4000: train bpb s5 0.0460; quiet KL s5 0.00039; 0.64 s/step; the job: 19.2 min spent, about 154.4 min of steps left; card peak 14.0 GB", "[ 1182.2s] stage 5/8 s5_causal step 1500/4000: train bpb s5 0.0460; quiet KL s5 0.00042; 0.64 s/step; the job: 19.7 min spent, about 153.9 min of steps left; card peak 14.0 GB", "[ 1214.0s] stage 5/8 s5_causal step 1550/4000: train bpb s5 0.0461; quiet KL s5 0.00039; 0.64 s/step; the job: 20.2 min spent, about 153.4 min of steps left; card peak 14.0 GB", "[ 1245.8s] stage 5/8 s5_causal step 1600/4000: train bpb s5 0.0457; quiet KL s5 0.00038; 0.64 s/step; the job: 20.8 min spent, about 152.8 min of steps left; card peak 14.0 GB", "[ 1277.4s] stage 5/8 s5_causal step 1650/4000: train bpb s5 0.0452; quiet KL s5 0.00039; 0.64 s/step; the job: 21.3 min spent, about 152.3 min of steps left; card peak 14.0 GB", "[ 1308.7s] stage 5/8 s5_causal step 1700/4000: train bpb s5 0.0453; quiet KL s5 0.00037; 0.64 s/step; the job: 21.8 min spent, about 151.7 min of steps left; card peak 14.0 GB", "[ 1340.3s] stage 5/8 s5_causal step 1750/4000: train bpb s5 0.0451; quiet KL s5 0.00038; 0.64 s/step; the job: 22.3 min spent, about 151.1 min of steps left; card peak 14.0 GB", "[ 1371.7s] stage 5/8 s5_causal step 1800/4000: train bpb s5 0.0449; quiet KL s5 0.00035; 0.64 s/step; the job: 22.9 min spent, about 150.5 min of steps left; card peak 14.0 GB", "[ 1403.2s] stage 5/8 s5_causal step 1850/4000: train bpb s5 0.0450; quiet KL s5 0.00041; 0.64 s/step; the job: 23.4 min spent, about 149.9 min of steps left; card peak 14.0 GB", "[ 1434.8s] stage 5/8 s5_causal step 1900/4000: train bpb s5 0.0450; quiet KL s5 0.00038; 0.64 s/step; the job: 23.9 min spent, about 149.4 min of steps left; card peak 14.0 GB", "[ 1466.3s] stage 5/8 s5_causal step 1950/4000: train bpb s5 0.0447; quiet KL s5 0.00038; 0.64 s/step; the job: 24.4 min spent, about 148.8 min of steps left; card peak 14.0 GB", "[ 1497.7s] stage 5/8 s5_causal step 2000/4000: train bpb s5 0.0446; quiet KL s5 0.00036; 0.64 s/step; the job: 25.0 min spent, about 148.3 min of steps left; card peak 14.0 GB", "[ 1529.2s] stage 5/8 s5_causal step 2050/4000: train bpb s5 0.0447; quiet KL s5 0.00034; 0.64 s/step; the job: 25.5 min spent, about 147.7 min of steps left; card peak 14.0 GB", "[ 1560.7s] stage 5/8 s5_causal step 2100/4000: train bpb s5 0.0441; quiet KL s5 0.00033; 0.64 s/step; the job: 26.0 min spent, about 147.1 min of steps left; card peak 14.0 GB", "[ 1592.3s] stage 5/8 s5_causal step 2150/4000: train bpb s5 0.0446; quiet KL s5 0.00035; 0.64 s/step; the job: 26.5 min spent, about 146.6 min of steps left; card peak 14.0 GB", "[ 1624.2s] stage 5/8 s5_causal step 2200/4000: train bpb s5 0.0440; quiet KL s5 0.00033; 0.64 s/step; the job: 27.1 min spent, about 146.1 min of steps left; card peak 14.0 GB", "[ 1655.7s] stage 5/8 s5_causal step 2250/4000: train bpb s5 0.0440; quiet KL s5 0.00032; 0.64 s/step; the job: 27.6 min spent, about 145.5 min of steps left; card peak 14.0 GB", "[ 1687.3s] stage 5/8 s5_causal step 2300/4000: train bpb s5 0.0439; quiet KL s5 0.00031; 0.63 s/step; the job: 28.1 min spent, about 145.0 min of steps left; card peak 14.0 GB", "[ 1718.9s] stage 5/8 s5_causal step 2350/4000: train bpb s5 0.0438; quiet KL s5 0.00034; 0.63 s/step; the job: 28.6 min spent, about 144.4 min of steps left; card peak 14.0 GB", "[ 1750.3s] stage 5/8 s5_causal step 2400/4000: train bpb s5 0.0440; quiet KL s5 0.00035; 0.63 s/step; the job: 29.2 min spent, about 143.9 min of steps left; card peak 14.0 GB", "[ 1781.8s] stage 5/8 s5_causal step 2450/4000: train bpb s5 0.0439; quiet KL s5 0.00035; 0.63 s/step; the job: 29.7 min spent, about 143.3 min of steps left; card peak 14.0 GB", "[ 1813.4s] stage 5/8 s5_causal step 2500/4000: train bpb s5 0.0440; quiet KL s5 0.00035; 0.63 s/step; the job: 30.2 min spent, about 142.8 min of steps left; card peak 14.0 GB", "[ 1845.1s] stage 5/8 s5_causal step 2550/4000: train bpb s5 0.0438; quiet KL s5 0.00034; 0.63 s/step; the job: 30.8 min spent, about 142.2 min of steps left; card peak 14.0 GB", "[ 1876.6s] stage 5/8 s5_causal step 2600/4000: train bpb s5 0.0443; quiet KL s5 0.00037; 0.63 s/step; the job: 31.3 min spent, about 141.7 min of steps left; card peak 14.0 GB", "[ 1908.3s] stage 5/8 s5_causal step 2650/4000: train bpb s5 0.0440; quiet KL s5 0.00032; 0.63 s/step; the job: 31.8 min spent, about 141.2 min of steps left; card peak 14.0 GB", "[ 1939.8s] stage 5/8 s5_causal step 2700/4000: train bpb s5 0.0437; quiet KL s5 0.00033; 0.63 s/step; the job: 32.3 min spent, about 140.6 min of steps left; card peak 14.0 GB", "[ 1971.3s] stage 5/8 s5_causal step 2750/4000: train bpb s5 0.0437; quiet KL s5 0.00032; 0.63 s/step; the job: 32.9 min spent, about 140.1 min of steps left; card peak 14.0 GB", "[ 2002.9s] stage 5/8 s5_causal step 2800/4000: train bpb s5 0.0440; quiet KL s5 0.00035; 0.63 s/step; the job: 33.4 min spent, about 139.5 min of steps left; card peak 14.0 GB", "[ 2034.4s] stage 5/8 s5_causal step 2850/4000: train bpb s5 0.0433; quiet KL s5 0.00031; 0.63 s/step; the job: 33.9 min spent, about 139.0 min of steps left; card peak 14.0 GB", "[ 2065.8s] stage 5/8 s5_causal step 2900/4000: train bpb s5 0.0433; quiet KL s5 0.00030; 0.63 s/step; the job: 34.4 min spent, about 138.4 min of steps left; card peak 14.0 GB", "[ 2097.5s] stage 5/8 s5_causal step 2950/4000: train bpb s5 0.0432; quiet KL s5 0.00030; 0.63 s/step; the job: 35.0 min spent, about 137.9 min of steps left; card peak 14.0 GB", "[ 2129.0s] stage 5/8 s5_causal step 3000/4000: train bpb s5 0.0432; quiet KL s5 0.00031; 0.63 s/step; the job: 35.5 min spent, about 137.4 min of steps left; card peak 14.0 GB", "[ 2160.5s] stage 5/8 s5_causal step 3050/4000: train bpb s5 0.0433; quiet KL s5 0.00031; 0.63 s/step; the job: 36.0 min spent, about 136.8 min of steps left; card peak 14.0 GB", "[ 2192.2s] stage 5/8 s5_causal step 3100/4000: train bpb s5 0.0435; quiet KL s5 0.00031; 0.63 s/step; the job: 36.5 min spent, about 136.3 min of steps left; card peak 14.0 GB", "[ 2223.7s] stage 5/8 s5_causal step 3150/4000: train bpb s5 0.0432; quiet KL s5 0.00031; 0.63 s/step; the job: 37.1 min spent, about 135.8 min of steps left; card peak 14.0 GB", "[ 2255.2s] stage 5/8 s5_causal step 3200/4000: train bpb s5 0.0435; quiet KL s5 0.00029; 0.63 s/step; the job: 37.6 min spent, about 135.2 min of steps left; card peak 14.0 GB", "[ 2287.0s] stage 5/8 s5_causal step 3250/4000: train bpb s5 0.0433; quiet KL s5 0.00032; 0.63 s/step; the job: 38.1 min spent, about 134.7 min of steps left; card peak 14.0 GB", "[ 2318.5s] stage 5/8 s5_causal step 3300/4000: train bpb s5 0.0434; quiet KL s5 0.00034; 0.63 s/step; the job: 38.6 min spent, about 134.2 min of steps left; card peak 14.0 GB", "[ 2350.0s] stage 5/8 s5_causal step 3350/4000: train bpb s5 0.0431; quiet KL s5 0.00031; 0.63 s/step; the job: 39.2 min spent, about 133.6 min of steps left; card peak 14.0 GB", "[ 2381.6s] stage 5/8 s5_causal step 3400/4000: train bpb s5 0.0431; quiet KL s5 0.00032; 0.63 s/step; the job: 39.7 min spent, about 133.1 min of steps left; card peak 14.0 GB", "[ 2413.1s] stage 5/8 s5_causal step 3450/4000: train bpb s5 0.0430; quiet KL s5 0.00037; 0.63 s/step; the job: 40.2 min spent, about 132.5 min of steps left; card peak 14.0 GB", "[ 2444.6s] stage 5/8 s5_causal step 3500/4000: train bpb s5 0.0429; quiet KL s5 0.00030; 0.63 s/step; the job: 40.7 min spent, about 132.0 min of steps left; card peak 14.0 GB", "[ 2476.4s] stage 5/8 s5_causal step 3550/4000: train bpb s5 0.0431; quiet KL s5 0.00033; 0.63 s/step; the job: 41.3 min spent, about 131.5 min of steps left; card peak 14.0 GB", "[ 2507.9s] stage 5/8 s5_causal step 3600/4000: train bpb s5 0.0430; quiet KL s5 0.00034; 0.63 s/step; the job: 41.8 min spent, about 130.9 min of steps left; card peak 14.0 GB", "[ 2539.4s] stage 5/8 s5_causal step 3650/4000: train bpb s5 0.0429; quiet KL s5 0.00029; 0.63 s/step; the job: 42.3 min spent, about 130.4 min of steps left; card peak 14.0 GB", "[ 2571.2s] stage 5/8 s5_causal step 3700/4000: train bpb s5 0.0427; quiet KL s5 0.00031; 0.63 s/step; the job: 42.9 min spent, about 129.9 min of steps left; card peak 14.0 GB", "[ 2602.9s] stage 5/8 s5_causal step 3750/4000: train bpb s5 0.0426; quiet KL s5 0.00031; 0.63 s/step; the job: 43.4 min spent, about 129.4 min of steps left; card peak 14.0 GB", "[ 2634.6s] stage 5/8 s5_causal step 3800/4000: train bpb s5 0.0427; quiet KL s5 0.00030; 0.63 s/step; the job: 43.9 min spent, about 128.8 min of steps left; card peak 14.0 GB", "[ 2666.5s] stage 5/8 s5_causal step 3850/4000: train bpb s5 0.0427; quiet KL s5 0.00031; 0.63 s/step; the job: 44.4 min spent, about 128.3 min of steps left; card peak 14.0 GB", "[ 2698.3s] stage 5/8 s5_causal step 3900/4000: train bpb s5 0.0426; quiet KL s5 0.00030; 0.63 s/step; the job: 45.0 min spent, about 127.8 min of steps left; card peak 14.0 GB", "[ 2729.9s] stage 5/8 s5_causal step 3950/4000: train bpb s5 0.0430; quiet KL s5 0.00041; 0.63 s/step; the job: 45.5 min spent, about 127.3 min of steps left; card peak 14.0 GB", "[ 2761.7s] stage 5/8 s5_causal step 4000/4000: train bpb s5 0.0431; quiet KL s5 0.00035; 0.63 s/step; the job: 46.0 min spent, about 126.7 min of steps left; card peak 14.0 GB", "[ 2762.5s] anchors saved: /root/refit/group/anchors/gXA_close5_.safetensors for ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal']", "[ 2917.3s] CLOSE 5/8: the group ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal'] all on, against arms off:", "[ 2917.3s] s1_perspective rows: off 0.5538 -> all on 0.0489 (-0.5050 +- 0.0048); with s1_perspective masked 0.5074 (its share -0.4586); s1_perspective alone 0.0525 (-0.5013); gate mean 0.210 / widest 0.571", "[ 2917.3s] s2_concept rows: off 0.7212 -> all on 0.1116 (-0.6095 +- 0.0038); with s2_concept masked 0.2356 (its share -0.1239); s2_concept alone 0.3503 (-0.3709); gate mean 0.277 / widest 0.498", "[ 2917.3s] s3_rules rows: off 0.9296 -> all on 0.2241 (-0.7055 +- 0.0053); with s3_rules masked 0.3162 (its share -0.0920); s3_rules alone 0.8236 (-0.1060); gate mean 0.120 / widest 0.174", "[ 2917.3s] s4_arith rows: off 1.2339 -> all on 0.3385 (-0.8954 +- 0.0059); with s4_arith masked 0.4841 (its share -0.1456); s4_arith alone 1.1267 (-0.1072); gate mean 0.108 / widest 0.159", "[ 2917.3s] s5_causal rows: off 0.6604 -> all on 0.0432 (-0.6173 +- 0.0070); with s5_causal masked 0.4786 (its share -0.4354); s5_causal alone 0.0911 (-0.5693); gate mean 0.185 / widest 0.368", "[ 2917.3s] web holdout: off 0.9506 -> all on 0.9527 (+0.00211 +- 0.00042; the quiet bar +.012); each member's share: s1_perspective +0.00053, s2_concept -0.00026, s3_rules +0.00049, s4_arith +0.00022, s5_causal +0.00040; alone: s1_perspective +0.00143, s2_concept +0.00105, s3_rules +0.00053, s4_arith +0.00011, s5_causal +0.00028", "[ 2917.3s] a member's share on a PARTNER's rows (all on minus that member masked; near 0 = it stays out): s1_perspective on s2_concept -0.2330; s1_perspective on s3_rules -0.1819; s1_perspective on s4_arith -0.1539; s1_perspective on s5_causal -0.0159; s2_concept on s1_perspective -0.0057; s2_concept on s3_rules -0.1071; s2_concept on s4_arith -0.1193; s2_concept on s5_causal -0.0020; s3_rules on s1_perspective -0.0001; s3_rules on s2_concept +0.0008; s3_rules on s4_arith -0.0003; s3_rules on s5_causal +0.0001; s4_arith on s1_perspective +0.0001; s4_arith on s2_concept -0.0005; s4_arith on s3_rules +0.0002; s4_arith on s5_causal +0.0001; s5_causal on s1_perspective +0.0007; s5_causal on s2_concept -0.0006; s5_causal on s3_rules +0.0005; s5_causal on s4_arith +0.0001", "[ 2917.3s] a member ALONE on a partner's rows (alone minus off): s1_perspective on s2_concept -0.4817; s1_perspective on s3_rules -0.4209; s1_perspective on s4_arith -0.5113; s1_perspective on s5_causal -0.1978; s2_concept on s1_perspective -0.0273; s2_concept on s3_rules -0.3108; s2_concept on s4_arith -0.3537; s2_concept on s5_causal -0.0015; s3_rules on s1_perspective -0.0003; s3_rules on s2_concept +0.0031; s3_rules on s4_arith +0.0001; s3_rules on s5_causal +0.0071; s4_arith on s1_perspective -0.0009; s4_arith on s2_concept -0.0004; s4_arith on s3_rules -0.0016; s4_arith on s5_causal -0.0033; s5_causal on s1_perspective -0.0183; s5_causal on s2_concept -0.0109; s5_causal on s3_rules -0.0122; s5_causal on s4_arith -0.0091", "[ 2917.3s] the group on the rows of stages not yet attached: s6_tryfail -0.1896; s7_mixed -0.0081; s8_register -0.1126", "[ 2917.3s] s1_perspective battery accuracy, off -> all on: in 0.550 -> 0.912, transfer 0.463 -> 0.438; alone: in 0.838, transfer 0.463", "[ 2917.3s] s2_concept battery accuracy, off -> all on: in 0.298 -> 0.976, transfer 0.678 -> 0.746; alone: in 0.964, transfer 0.720", "[ 2917.3s] s3_rules battery accuracy, off -> all on: in 0.155 -> 0.880, transfer 0.070 -> 0.300, held-out 0.050 -> 0.508; alone: in 0.320, transfer 0.080, held-out 0.108", "[ 2917.3s] s4_arith battery accuracy, off -> all on: in 0.344 -> 0.606, transfer 0.242 -> 0.229, held-out 0.350 -> 0.338; alone: in 0.381, transfer 0.246, held-out 0.388", "[ 2917.3s] probes mean: off 0.411 -> all on 0.430", "[ 2917.3s] compiled chain: 13 cache entries (limit 256), 0 limit warnings; card peak 14.0 GB; host RSS 5.3 GB", "[ 2920.9s] STAGE 6/8: s6_tryfail attached over ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal'] (on, frozen); at birth the group reads exactly as before on its rows and on web text; 4000 steps of 1 task + 1 abstention chunks", "[ 2922.4s] stage 6/8 s6_tryfail step 1/4000: train bpb s6 0.5003; quiet KL s6 0.00012; 1.53 s/step; the job: 48.7 min spent, about 126.8 min of steps left; card peak 14.0 GB", "[ 2956.0s] stage 6/8 s6_tryfail step 50/4000: train bpb s6 0.4784; quiet KL s6 0.00018; 0.70 s/step; the job: 49.3 min spent, about 126.4 min of steps left; card peak 14.1 GB", "[ 2989.8s] stage 6/8 s6_tryfail step 100/4000: train bpb s6 0.4475; quiet KL s6 0.00040; 0.69 s/step; the job: 49.8 min spent, about 125.9 min of steps left; card peak 14.1 GB", "[ 3023.7s] stage 6/8 s6_tryfail step 150/4000: train bpb s6 0.4025; quiet KL s6 0.00051; 0.69 s/step; the job: 50.4 min spent, about 125.5 min of steps left; card peak 14.1 GB", "[ 3057.7s] stage 6/8 s6_tryfail step 200/4000: train bpb s6 0.3530; quiet KL s6 0.00061; 0.68 s/step; the job: 51.0 min spent, about 125.1 min of steps left; card peak 14.1 GB", "[ 3091.6s] stage 6/8 s6_tryfail step 250/4000: train bpb s6 0.3117; quiet KL s6 0.00074; 0.68 s/step; the job: 51.5 min spent, about 124.7 min of steps left; card peak 14.1 GB", "[ 3125.3s] stage 6/8 s6_tryfail step 300/4000: train bpb s6 0.2728; quiet KL s6 0.00072; 0.68 s/step; the job: 52.1 min spent, about 124.2 min of steps left; card peak 14.1 GB", "[ 3159.2s] stage 6/8 s6_tryfail step 350/4000: train bpb s6 0.2478; quiet KL s6 0.00068; 0.68 s/step; the job: 52.7 min spent, about 123.8 min of steps left; card peak 14.1 GB", "[ 3193.1s] stage 6/8 s6_tryfail step 400/4000: train bpb s6 0.2154; quiet KL s6 0.00073; 0.68 s/step; the job: 53.2 min spent, about 123.3 min of steps left; card peak 14.1 GB", "[ 3226.7s] stage 6/8 s6_tryfail step 450/4000: train bpb s6 0.1939; quiet KL s6 0.00070; 0.68 s/step; the job: 53.8 min spent, about 122.9 min of steps left; card peak 14.1 GB", "[ 3260.7s] stage 6/8 s6_tryfail step 500/4000: train bpb s6 0.1731; quiet KL s6 0.00068; 0.68 s/step; the job: 54.3 min spent, about 122.4 min of steps left; card peak 14.1 GB", "[ 3294.7s] stage 6/8 s6_tryfail step 550/4000: train bpb s6 0.1561; quiet KL s6 0.00075; 0.68 s/step; the job: 54.9 min spent, about 122.0 min of steps left; card peak 14.1 GB", "[ 3328.4s] stage 6/8 s6_tryfail step 600/4000: train bpb s6 0.1364; quiet KL s6 0.00066; 0.68 s/step; the job: 55.5 min spent, about 121.5 min of steps left; card peak 14.1 GB", "[ 3362.4s] stage 6/8 s6_tryfail step 650/4000: train bpb s6 0.1226; quiet KL s6 0.00068; 0.68 s/step; the job: 56.0 min spent, about 121.1 min of steps left; card peak 14.1 GB", "[ 3396.2s] stage 6/8 s6_tryfail step 700/4000: train bpb s6 0.1139; quiet KL s6 0.00069; 0.68 s/step; the job: 56.6 min spent, about 120.6 min of steps left; card peak 14.1 GB", "[ 3429.7s] stage 6/8 s6_tryfail step 750/4000: train bpb s6 0.1028; quiet KL s6 0.00070; 0.68 s/step; the job: 57.2 min spent, about 120.1 min of steps left; card peak 14.1 GB", "[ 3463.4s] stage 6/8 s6_tryfail step 800/4000: train bpb s6 0.0953; quiet KL s6 0.00071; 0.68 s/step; the job: 57.7 min spent, about 119.7 min of steps left; card peak 14.1 GB", "[ 3497.1s] stage 6/8 s6_tryfail step 850/4000: train bpb s6 0.0907; quiet KL s6 0.00060; 0.68 s/step; the job: 58.3 min spent, about 119.2 min of steps left; card peak 14.1 GB", "[ 3530.9s] stage 6/8 s6_tryfail step 900/4000: train bpb s6 0.0918; quiet KL s6 0.00059; 0.68 s/step; the job: 58.8 min spent, about 118.7 min of steps left; card peak 14.1 GB", "[ 3564.5s] stage 6/8 s6_tryfail step 950/4000: train bpb s6 0.0908; quiet KL s6 0.00059; 0.68 s/step; the job: 59.4 min spent, about 118.2 min of steps left; card peak 14.1 GB", "[ 3598.4s] stage 6/8 s6_tryfail step 1000/4000: train bpb s6 0.0857; quiet KL s6 0.00064; 0.68 s/step; the job: 60.0 min spent, about 117.8 min of steps left; card peak 14.1 GB", "[ 3632.1s] stage 6/8 s6_tryfail step 1050/4000: train bpb s6 0.0856; quiet KL s6 0.00057; 0.68 s/step; the job: 60.5 min spent, about 117.3 min of steps left; card peak 14.1 GB", "[ 3665.7s] stage 6/8 s6_tryfail step 1100/4000: train bpb s6 0.0815; quiet KL s6 0.00056; 0.68 s/step; the job: 61.1 min spent, about 116.8 min of steps left; card peak 14.1 GB", "[ 3699.4s] stage 6/8 s6_tryfail step 1150/4000: train bpb s6 0.0813; quiet KL s6 0.00060; 0.68 s/step; the job: 61.7 min spent, about 116.3 min of steps left; card peak 14.1 GB", "[ 3733.1s] stage 6/8 s6_tryfail step 1200/4000: train bpb s6 0.0821; quiet KL s6 0.00064; 0.68 s/step; the job: 62.2 min spent, about 115.9 min of steps left; card peak 14.1 GB", "[ 3766.7s] stage 6/8 s6_tryfail step 1250/4000: train bpb s6 0.0793; quiet KL s6 0.00057; 0.68 s/step; the job: 62.8 min spent, about 115.4 min of steps left; card peak 14.1 GB", "[ 3800.4s] stage 6/8 s6_tryfail step 1300/4000: train bpb s6 0.0783; quiet KL s6 0.00056; 0.68 s/step; the job: 63.3 min spent, about 114.9 min of steps left; card peak 14.1 GB", "[ 3834.2s] stage 6/8 s6_tryfail step 1350/4000: train bpb s6 0.0768; quiet KL s6 0.00059; 0.68 s/step; the job: 63.9 min spent, about 114.4 min of steps left; card peak 14.1 GB", "[ 3867.7s] stage 6/8 s6_tryfail step 1400/4000: train bpb s6 0.0758; quiet KL s6 0.00056; 0.68 s/step; the job: 64.5 min spent, about 113.9 min of steps left; card peak 14.1 GB", "[ 3901.5s] stage 6/8 s6_tryfail step 1450/4000: train bpb s6 0.0746; quiet KL s6 0.00055; 0.68 s/step; the job: 65.0 min spent, about 113.4 min of steps left; card peak 14.1 GB", "[ 3935.3s] stage 6/8 s6_tryfail step 1500/4000: train bpb s6 0.0767; quiet KL s6 0.00057; 0.68 s/step; the job: 65.6 min spent, about 112.9 min of steps left; card peak 14.1 GB", "[ 3969.0s] stage 6/8 s6_tryfail step 1550/4000: train bpb s6 0.0743; quiet KL s6 0.00053; 0.68 s/step; the job: 66.2 min spent, about 112.4 min of steps left; card peak 14.1 GB", "[ 4002.6s] stage 6/8 s6_tryfail step 1600/4000: train bpb s6 0.0737; quiet KL s6 0.00049; 0.68 s/step; the job: 66.7 min spent, about 111.9 min of steps left; card peak 14.1 GB", "[ 4036.4s] stage 6/8 s6_tryfail step 1650/4000: train bpb s6 0.0676; quiet KL s6 0.00053; 0.68 s/step; the job: 67.3 min spent, about 111.4 min of steps left; card peak 14.1 GB", "[ 4070.1s] stage 6/8 s6_tryfail step 1700/4000: train bpb s6 0.0693; quiet KL s6 0.00053; 0.68 s/step; the job: 67.8 min spent, about 110.9 min of steps left; card peak 14.1 GB", "[ 4103.7s] stage 6/8 s6_tryfail step 1750/4000: train bpb s6 0.0703; quiet KL s6 0.00054; 0.68 s/step; the job: 68.4 min spent, about 110.4 min of steps left; card peak 14.1 GB", "[ 4137.5s] stage 6/8 s6_tryfail step 1800/4000: train bpb s6 0.0715; quiet KL s6 0.00053; 0.68 s/step; the job: 69.0 min spent, about 110.0 min of steps left; card peak 14.1 GB", "[ 4171.3s] stage 6/8 s6_tryfail step 1850/4000: train bpb s6 0.0684; quiet KL s6 0.00053; 0.68 s/step; the job: 69.5 min spent, about 109.5 min of steps left; card peak 14.1 GB", "[ 4204.8s] stage 6/8 s6_tryfail step 1900/4000: train bpb s6 0.0682; quiet KL s6 0.00055; 0.68 s/step; the job: 70.1 min spent, about 108.9 min of steps left; card peak 14.1 GB", "[ 4238.7s] stage 6/8 s6_tryfail step 1950/4000: train bpb s6 0.0702; quiet KL s6 0.00054; 0.68 s/step; the job: 70.6 min spent, about 108.5 min of steps left; card peak 14.1 GB", "[ 4272.8s] stage 6/8 s6_tryfail step 2000/4000: train bpb s6 0.0696; quiet KL s6 0.00049; 0.68 s/step; the job: 71.2 min spent, about 108.0 min of steps left; card peak 14.1 GB", "[ 4306.5s] stage 6/8 s6_tryfail step 2050/4000: train bpb s6 0.0674; quiet KL s6 0.00048; 0.68 s/step; the job: 71.8 min spent, about 107.5 min of steps left; card peak 14.1 GB", "[ 4340.4s] stage 6/8 s6_tryfail step 2100/4000: train bpb s6 0.0692; quiet KL s6 0.00048; 0.68 s/step; the job: 72.3 min spent, about 107.0 min of steps left; card peak 14.1 GB", "[ 4373.9s] stage 6/8 s6_tryfail step 2150/4000: train bpb s6 0.0685; quiet KL s6 0.00054; 0.68 s/step; the job: 72.9 min spent, about 106.4 min of steps left; card peak 14.1 GB", "[ 4407.4s] stage 6/8 s6_tryfail step 2200/4000: train bpb s6 0.0655; quiet KL s6 0.00045; 0.68 s/step; the job: 73.5 min spent, about 105.9 min of steps left; card peak 14.1 GB", "[ 4440.7s] stage 6/8 s6_tryfail step 2250/4000: train bpb s6 0.0660; quiet KL s6 0.00047; 0.68 s/step; the job: 74.0 min spent, about 105.4 min of steps left; card peak 14.1 GB", "[ 4474.3s] stage 6/8 s6_tryfail step 2300/4000: train bpb s6 0.0644; quiet KL s6 0.00051; 0.68 s/step; the job: 74.6 min spent, about 104.9 min of steps left; card peak 14.1 GB", "[ 4508.1s] stage 6/8 s6_tryfail step 2350/4000: train bpb s6 0.0658; quiet KL s6 0.00047; 0.68 s/step; the job: 75.1 min spent, about 104.4 min of steps left; card peak 14.1 GB", "[ 4541.6s] stage 6/8 s6_tryfail step 2400/4000: train bpb s6 0.0619; quiet KL s6 0.00042; 0.68 s/step; the job: 75.7 min spent, about 103.9 min of steps left; card peak 14.1 GB", "[ 4575.4s] stage 6/8 s6_tryfail step 2450/4000: train bpb s6 0.0624; quiet KL s6 0.00047; 0.68 s/step; the job: 76.3 min spent, about 103.4 min of steps left; card peak 14.1 GB", "[ 4609.1s] stage 6/8 s6_tryfail step 2500/4000: train bpb s6 0.0640; quiet KL s6 0.00049; 0.68 s/step; the job: 76.8 min spent, about 102.9 min of steps left; card peak 14.1 GB", "[ 4642.6s] stage 6/8 s6_tryfail step 2550/4000: train bpb s6 0.0639; quiet KL s6 0.00041; 0.68 s/step; the job: 77.4 min spent, about 102.3 min of steps left; card peak 14.1 GB", "[ 4676.4s] stage 6/8 s6_tryfail step 2600/4000: train bpb s6 0.0632; quiet KL s6 0.00045; 0.68 s/step; the job: 77.9 min spent, about 101.8 min of steps left; card peak 14.1 GB", "[ 4710.2s] stage 6/8 s6_tryfail step 2650/4000: train bpb s6 0.0632; quiet KL s6 0.00042; 0.68 s/step; the job: 78.5 min spent, about 101.3 min of steps left; card peak 14.1 GB", "[ 4744.0s] stage 6/8 s6_tryfail step 2700/4000: train bpb s6 0.0637; quiet KL s6 0.00045; 0.68 s/step; the job: 79.1 min spent, about 100.8 min of steps left; card peak 14.1 GB", "[ 4777.7s] stage 6/8 s6_tryfail step 2750/4000: train bpb s6 0.0623; quiet KL s6 0.00041; 0.68 s/step; the job: 79.6 min spent, about 100.3 min of steps left; card peak 14.1 GB", "[ 4811.5s] stage 6/8 s6_tryfail step 2800/4000: train bpb s6 0.0623; quiet KL s6 0.00044; 0.68 s/step; the job: 80.2 min spent, about 99.8 min of steps left; card peak 14.1 GB", "[ 4845.3s] stage 6/8 s6_tryfail step 2850/4000: train bpb s6 0.0602; quiet KL s6 0.00043; 0.68 s/step; the job: 80.8 min spent, about 99.3 min of steps left; card peak 14.1 GB", "[ 4878.9s] stage 6/8 s6_tryfail step 2900/4000: train bpb s6 0.0633; quiet KL s6 0.00043; 0.68 s/step; the job: 81.3 min spent, about 98.8 min of steps left; card peak 14.1 GB", "[ 4912.7s] stage 6/8 s6_tryfail step 2950/4000: train bpb s6 0.0627; quiet KL s6 0.00048; 0.68 s/step; the job: 81.9 min spent, about 98.2 min of steps left; card peak 14.1 GB", "[ 4946.5s] stage 6/8 s6_tryfail step 3000/4000: train bpb s6 0.0591; quiet KL s6 0.00041; 0.68 s/step; the job: 82.4 min spent, about 97.7 min of steps left; card peak 14.1 GB", "[ 4980.0s] stage 6/8 s6_tryfail step 3050/4000: train bpb s6 0.0617; quiet KL s6 0.00047; 0.68 s/step; the job: 83.0 min spent, about 97.2 min of steps left; card peak 14.1 GB", "[ 5013.5s] stage 6/8 s6_tryfail step 3100/4000: train bpb s6 0.0607; quiet KL s6 0.00043; 0.68 s/step; the job: 83.6 min spent, about 96.7 min of steps left; card peak 14.1 GB", "[ 5047.0s] stage 6/8 s6_tryfail step 3150/4000: train bpb s6 0.0619; quiet KL s6 0.00047; 0.67 s/step; the job: 84.1 min spent, about 96.1 min of steps left; card peak 14.1 GB", "[ 5080.2s] stage 6/8 s6_tryfail step 3200/4000: train bpb s6 0.0608; quiet KL s6 0.00043; 0.67 s/step; the job: 84.7 min spent, about 95.6 min of steps left; card peak 14.1 GB", "[ 5113.7s] stage 6/8 s6_tryfail step 3250/4000: train bpb s6 0.0588; quiet KL s6 0.00043; 0.67 s/step; the job: 85.2 min spent, about 95.1 min of steps left; card peak 14.1 GB", "[ 5147.2s] stage 6/8 s6_tryfail step 3300/4000: train bpb s6 0.0625; quiet KL s6 0.00041; 0.67 s/step; the job: 85.8 min spent, about 94.6 min of steps left; card peak 14.1 GB", "[ 5180.5s] stage 6/8 s6_tryfail step 3350/4000: train bpb s6 0.0616; quiet KL s6 0.00044; 0.67 s/step; the job: 86.3 min spent, about 94.0 min of steps left; card peak 14.1 GB", "[ 5214.0s] stage 6/8 s6_tryfail step 3400/4000: train bpb s6 0.0623; quiet KL s6 0.00039; 0.67 s/step; the job: 86.9 min spent, about 93.5 min of steps left; card peak 14.1 GB", "[ 5247.6s] stage 6/8 s6_tryfail step 3450/4000: train bpb s6 0.0585; quiet KL s6 0.00043; 0.67 s/step; the job: 87.5 min spent, about 93.0 min of steps left; card peak 14.1 GB", "[ 5281.2s] stage 6/8 s6_tryfail step 3500/4000: train bpb s6 0.0624; quiet KL s6 0.00044; 0.67 s/step; the job: 88.0 min spent, about 92.5 min of steps left; card peak 14.1 GB", "[ 5314.6s] stage 6/8 s6_tryfail step 3550/4000: train bpb s6 0.0609; quiet KL s6 0.00043; 0.67 s/step; the job: 88.6 min spent, about 91.9 min of steps left; card peak 14.1 GB", "[ 5348.1s] stage 6/8 s6_tryfail step 3600/4000: train bpb s6 0.0610; quiet KL s6 0.00039; 0.67 s/step; the job: 89.1 min spent, about 91.4 min of steps left; card peak 14.1 GB", "[ 5381.6s] stage 6/8 s6_tryfail step 3650/4000: train bpb s6 0.0584; quiet KL s6 0.00039; 0.67 s/step; the job: 89.7 min spent, about 90.9 min of steps left; card peak 14.1 GB", "[ 5414.8s] stage 6/8 s6_tryfail step 3700/4000: train bpb s6 0.0598; quiet KL s6 0.00046; 0.67 s/step; the job: 90.2 min spent, about 90.3 min of steps left; card peak 14.1 GB", "[ 5448.3s] stage 6/8 s6_tryfail step 3750/4000: train bpb s6 0.0603; quiet KL s6 0.00038; 0.67 s/step; the job: 90.8 min spent, about 89.8 min of steps left; card peak 14.1 GB", "[ 5482.0s] stage 6/8 s6_tryfail step 3800/4000: train bpb s6 0.0596; quiet KL s6 0.00048; 0.67 s/step; the job: 91.4 min spent, about 89.3 min of steps left; card peak 14.1 GB", "[ 5515.6s] stage 6/8 s6_tryfail step 3850/4000: train bpb s6 0.0616; quiet KL s6 0.00040; 0.67 s/step; the job: 91.9 min spent, about 88.8 min of steps left; card peak 14.1 GB", "[ 5549.4s] stage 6/8 s6_tryfail step 3900/4000: train bpb s6 0.0593; quiet KL s6 0.00047; 0.67 s/step; the job: 92.5 min spent, about 88.2 min of steps left; card peak 14.1 GB", "[ 5583.2s] stage 6/8 s6_tryfail step 3950/4000: train bpb s6 0.0593; quiet KL s6 0.00044; 0.67 s/step; the job: 93.1 min spent, about 87.7 min of steps left; card peak 14.1 GB", "[ 5616.8s] stage 6/8 s6_tryfail step 4000/4000: train bpb s6 0.0589; quiet KL s6 0.00040; 0.67 s/step; the job: 93.6 min spent, about 87.2 min of steps left; card peak 14.1 GB", "[ 5617.8s] anchors saved: /root/refit/group/anchors/gXA_close6_.safetensors for ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail']", "[ 5798.1s] CLOSE 6/8: the group ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail'] all on, against arms off:", "[ 5798.1s] s1_perspective rows: off 0.5538 -> all on 0.0488 (-0.5050 +- 0.0048); with s1_perspective masked 0.5048 (its share -0.4560); s1_perspective alone 0.0525 (-0.5013); gate mean 0.210 / widest 0.571", "[ 5798.1s] s2_concept rows: off 0.7212 -> all on 0.1119 (-0.6093 +- 0.0038); with s2_concept masked 0.2370 (its share -0.1252); s2_concept alone 0.3503 (-0.3709); gate mean 0.277 / widest 0.498", "[ 5798.1s] s3_rules rows: off 0.9296 -> all on 0.2241 (-0.7055 +- 0.0053); with s3_rules masked 0.3166 (its share -0.0925); s3_rules alone 0.8236 (-0.1060); gate mean 0.120 / widest 0.174", "[ 5798.1s] s4_arith rows: off 1.2339 -> all on 0.3379 (-0.8960 +- 0.0058); with s4_arith masked 0.4861 (its share -0.1482); s4_arith alone 1.1267 (-0.1072); gate mean 0.108 / widest 0.159", "[ 5798.1s] s5_causal rows: off 0.6604 -> all on 0.0430 (-0.6174 +- 0.0069); with s5_causal masked 0.4749 (its share -0.4319); s5_causal alone 0.0911 (-0.5693); gate mean 0.185 / widest 0.368", "[ 5798.1s] s6_tryfail rows: off 0.6728 -> all on 0.0549 (-0.6179 +- 0.0036); with s6_tryfail masked 0.4832 (its share -0.4283); s6_tryfail alone 0.1638 (-0.5090); gate mean 0.230 / widest 0.537", "[ 5798.1s] web holdout: off 0.9506 -> all on 0.9527 (+0.00214 +- 0.00043; the quiet bar +.012); each member's share: s1_perspective +0.00050, s2_concept -0.00014, s3_rules +0.00032, s4_arith +0.00033, s5_causal +0.00021, s6_tryfail +0.00003; alone: s1_perspective +0.00143, s2_concept +0.00105, s3_rules +0.00053, s4_arith +0.00011, s5_causal +0.00028, s6_tryfail +0.00013", "[ 5798.1s] a member's share on a PARTNER's rows (all on minus that member masked; near 0 = it stays out): s1_perspective on s2_concept -0.2334; s1_perspective on s3_rules -0.1806; s1_perspective on s4_arith -0.1488; s1_perspective on s5_causal -0.0177; s1_perspective on s6_tryfail -0.0385; s2_concept on s1_perspective -0.0055; s2_concept on s3_rules -0.1067; s2_concept on s4_arith -0.1187; s2_concept on s5_causal -0.0017; s2_concept on s6_tryfail -0.0033; s3_rules on s1_perspective +0.0000; s3_rules on s2_concept +0.0008; s3_rules on s4_arith -0.0001; s3_rules on s5_causal +0.0000; s3_rules on s6_tryfail -0.0002; s4_arith on s1_perspective +0.0001; s4_arith on s2_concept -0.0005; s4_arith on s3_rules +0.0001; s4_arith on s5_causal -0.0001; s4_arith on s6_tryfail -0.0044; s5_causal on s1_perspective +0.0006; s5_causal on s2_concept -0.0007; s5_causal on s3_rules +0.0003; s5_causal on s4_arith +0.0002; s5_causal on s6_tryfail -0.0008; s6_tryfail on s1_perspective -0.0001; s6_tryfail on s2_concept +0.0002; s6_tryfail on s3_rules -0.0001; s6_tryfail on s4_arith -0.0006; s6_tryfail on s5_causal -0.0002", "[ 5798.1s] a member ALONE on a partner's rows (alone minus off): s1_perspective on s2_concept -0.4817; s1_perspective on s3_rules -0.4209; s1_perspective on s4_arith -0.5113; s1_perspective on s5_causal -0.1978; s1_perspective on s6_tryfail -0.1782; s2_concept on s1_perspective -0.0273; s2_concept on s3_rules -0.3108; s2_concept on s4_arith -0.3537; s2_concept on s5_causal -0.0015; s2_concept on s6_tryfail +0.0004; s3_rules on s1_perspective -0.0003; s3_rules on s2_concept +0.0031; s3_rules on s4_arith +0.0001; s3_rules on s5_causal +0.0071; s3_rules on s6_tryfail +0.0010; s4_arith on s1_perspective -0.0009; s4_arith on s2_concept -0.0004; s4_arith on s3_rules -0.0016; s4_arith on s5_causal -0.0033; s4_arith on s6_tryfail -0.0135; s5_causal on s1_perspective -0.0183; s5_causal on s2_concept -0.0109; s5_causal on s3_rules -0.0122; s5_causal on s4_arith -0.0091; s5_causal on s6_tryfail -0.0065; s6_tryfail on s1_perspective -0.0020; s6_tryfail on s2_concept +0.0012; s6_tryfail on s3_rules +0.0006; s6_tryfail on s4_arith +0.0036; s6_tryfail on s5_causal -0.0098", "[ 5798.1s] the group on the rows of stages not yet attached: s7_mixed -0.0082; s8_register -0.1128", "[ 5798.1s] s1_perspective battery accuracy, off -> all on: in 0.550 -> 0.887, transfer 0.463 -> 0.425; alone: in 0.838, transfer 0.463", "[ 5798.1s] s2_concept battery accuracy, off -> all on: in 0.298 -> 0.976, transfer 0.678 -> 0.712; alone: in 0.964, transfer 0.720", "[ 5798.1s] s3_rules battery accuracy, off -> all on: in 0.155 -> 0.870, transfer 0.070 -> 0.320, held-out 0.050 -> 0.475; alone: in 0.320, transfer 0.080, held-out 0.108", "[ 5798.1s] s4_arith battery accuracy, off -> all on: in 0.344 -> 0.637, transfer 0.242 -> 0.242, held-out 0.350 -> 0.325; alone: in 0.381, transfer 0.246, held-out 0.388", "[ 5798.1s] probes mean: off 0.411 -> all on 0.433", "[ 5798.1s] compiled chain: 18 cache entries (limit 256), 0 limit warnings; card peak 14.1 GB; host RSS 5.4 GB", "[ 5802.2s] STAGE 7/8: s7_mixed attached over ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail'] (on, frozen); at birth the group reads exactly as before on its rows and on web text; 4000 steps of 1 task + 1 abstention chunks", "[ 5803.9s] stage 7/8 s7_mixed step 1/4000: train bpb s7 0.3642; quiet KL s7 0.00017; 1.69 s/step; the job: 96.7 min spent, about 87.2 min of steps left; card peak 14.2 GB", "[ 5866.4s] stage 7/8 s7_mixed step 50/4000: train bpb s7 0.4690; quiet KL s7 0.00015; 1.28 s/step; the job: 97.8 min spent, about 87.2 min of steps left; card peak 14.3 GB", "[ 5902.2s] stage 7/8 s7_mixed step 100/4000: train bpb s7 0.5390; quiet KL s7 0.00015; 1.00 s/step; the job: 98.4 min spent, about 86.7 min of steps left; card peak 14.3 GB", "[ 5938.1s] stage 7/8 s7_mixed step 150/4000: train bpb s7 0.4871; quiet KL s7 0.00017; 0.91 s/step; the job: 99.0 min spent, about 86.1 min of steps left; card peak 14.3 GB", "[ 5974.0s] stage 7/8 s7_mixed step 200/4000: train bpb s7 0.3962; quiet KL s7 0.00021; 0.86 s/step; the job: 99.6 min spent, about 85.6 min of steps left; card peak 14.3 GB", "[ 6009.8s] stage 7/8 s7_mixed step 250/4000: train bpb s7 0.4321; quiet KL s7 0.00027; 0.83 s/step; the job: 100.2 min spent, about 85.1 min of steps left; card peak 14.3 GB", "[ 6045.6s] stage 7/8 s7_mixed step 300/4000: train bpb s7 0.5527; quiet KL s7 0.00029; 0.81 s/step; the job: 100.8 min spent, about 84.6 min of steps left; card peak 14.3 GB", "[ 6081.6s] stage 7/8 s7_mixed step 350/4000: train bpb s7 0.4444; quiet KL s7 0.00040; 0.80 s/step; the job: 101.4 min spent, about 84.1 min of steps left; card peak 14.3 GB", "[ 6117.4s] stage 7/8 s7_mixed step 400/4000: train bpb s7 0.4818; quiet KL s7 0.00049; 0.79 s/step; the job: 102.0 min spent, about 83.6 min of steps left; card peak 14.3 GB", "[ 6153.3s] stage 7/8 s7_mixed step 450/4000: train bpb s7 0.4726; quiet KL s7 0.00053; 0.78 s/step; the job: 102.6 min spent, about 83.1 min of steps left; card peak 14.3 GB", "[ 6189.1s] stage 7/8 s7_mixed step 500/4000: train bpb s7 0.4682; quiet KL s7 0.00043; 0.77 s/step; the job: 103.2 min spent, about 82.6 min of steps left; card peak 14.3 GB", "[ 6225.1s] stage 7/8 s7_mixed step 550/4000: train bpb s7 0.4382; quiet KL s7 0.00046; 0.77 s/step; the job: 103.8 min spent, about 82.1 min of steps left; card peak 14.3 GB", "[ 6261.0s] stage 7/8 s7_mixed step 600/4000: train bpb s7 0.3918; quiet KL s7 0.00034; 0.76 s/step; the job: 104.4 min spent, about 81.6 min of steps left; card peak 14.3 GB", "[ 6297.0s] stage 7/8 s7_mixed step 650/4000: train bpb s7 0.4019; quiet KL s7 0.00050; 0.76 s/step; the job: 104.9 min spent, about 81.1 min of steps left; card peak 14.3 GB", "[ 6332.9s] stage 7/8 s7_mixed step 700/4000: train bpb s7 0.4465; quiet KL s7 0.00047; 0.76 s/step; the job: 105.5 min spent, about 80.6 min of steps left; card peak 14.3 GB", "[ 6368.8s] stage 7/8 s7_mixed step 750/4000: train bpb s7 0.4083; quiet KL s7 0.00060; 0.76 s/step; the job: 106.1 min spent, about 80.1 min of steps left; card peak 14.3 GB", "[ 6404.8s] stage 7/8 s7_mixed step 800/4000: train bpb s7 0.4267; quiet KL s7 0.00043; 0.75 s/step; the job: 106.7 min spent, about 79.5 min of steps left; card peak 14.3 GB", "[ 6440.7s] stage 7/8 s7_mixed step 850/4000: train bpb s7 0.3861; quiet KL s7 0.00065; 0.75 s/step; the job: 107.3 min spent, about 79.0 min of steps left; card peak 14.3 GB", "[ 6476.6s] stage 7/8 s7_mixed step 900/4000: train bpb s7 0.4337; quiet KL s7 0.00052; 0.75 s/step; the job: 107.9 min spent, about 78.5 min of steps left; card peak 14.3 GB", "[ 6512.6s] stage 7/8 s7_mixed step 950/4000: train bpb s7 0.5111; quiet KL s7 0.00066; 0.75 s/step; the job: 108.5 min spent, about 78.0 min of steps left; card peak 14.3 GB", "[ 6548.5s] stage 7/8 s7_mixed step 1000/4000: train bpb s7 0.4723; quiet KL s7 0.00064; 0.75 s/step; the job: 109.1 min spent, about 77.5 min of steps left; card peak 14.3 GB", "[ 6584.5s] stage 7/8 s7_mixed step 1050/4000: train bpb s7 0.5203; quiet KL s7 0.00074; 0.75 s/step; the job: 109.7 min spent, about 77.0 min of steps left; card peak 14.3 GB", "[ 6620.5s] stage 7/8 s7_mixed step 1100/4000: train bpb s7 0.3957; quiet KL s7 0.00068; 0.74 s/step; the job: 110.3 min spent, about 76.4 min of steps left; card peak 14.3 GB", "[ 6656.4s] stage 7/8 s7_mixed step 1150/4000: train bpb s7 0.4222; quiet KL s7 0.00083; 0.74 s/step; the job: 110.9 min spent, about 75.9 min of steps left; card peak 14.3 GB", "[ 6692.2s] stage 7/8 s7_mixed step 1200/4000: train bpb s7 0.4063; quiet KL s7 0.00053; 0.74 s/step; the job: 111.5 min spent, about 75.4 min of steps left; card peak 14.3 GB", "[ 6727.9s] stage 7/8 s7_mixed step 1250/4000: train bpb s7 0.4492; quiet KL s7 0.00059; 0.74 s/step; the job: 112.1 min spent, about 74.9 min of steps left; card peak 14.3 GB", "[ 6763.6s] stage 7/8 s7_mixed step 1300/4000: train bpb s7 0.5419; quiet KL s7 0.00079; 0.74 s/step; the job: 112.7 min spent, about 74.3 min of steps left; card peak 14.3 GB", "[ 6799.4s] stage 7/8 s7_mixed step 1350/4000: train bpb s7 0.3960; quiet KL s7 0.00090; 0.74 s/step; the job: 113.3 min spent, about 73.8 min of steps left; card peak 14.3 GB", "[ 6835.1s] stage 7/8 s7_mixed step 1400/4000: train bpb s7 0.3689; quiet KL s7 0.00061; 0.74 s/step; the job: 113.9 min spent, about 73.3 min of steps left; card peak 14.3 GB", "[ 6870.9s] stage 7/8 s7_mixed step 1450/4000: train bpb s7 0.3995; quiet KL s7 0.00059; 0.74 s/step; the job: 114.5 min spent, about 72.8 min of steps left; card peak 14.3 GB", "[ 6906.6s] stage 7/8 s7_mixed step 1500/4000: train bpb s7 0.4601; quiet KL s7 0.00056; 0.74 s/step; the job: 115.1 min spent, about 72.2 min of steps left; card peak 14.3 GB", "[ 6942.4s] stage 7/8 s7_mixed step 1550/4000: train bpb s7 0.4147; quiet KL s7 0.00057; 0.74 s/step; the job: 115.7 min spent, about 71.7 min of steps left; card peak 14.3 GB", "[ 6978.2s] stage 7/8 s7_mixed step 1600/4000: train bpb s7 0.3934; quiet KL s7 0.00057; 0.74 s/step; the job: 116.3 min spent, about 71.2 min of steps left; card peak 14.3 GB", "[ 7013.9s] stage 7/8 s7_mixed step 1650/4000: train bpb s7 0.3674; quiet KL s7 0.00062; 0.73 s/step; the job: 116.9 min spent, about 70.7 min of steps left; card peak 14.3 GB", "[ 7049.8s] stage 7/8 s7_mixed step 1700/4000: train bpb s7 0.4999; quiet KL s7 0.00057; 0.73 s/step; the job: 117.5 min spent, about 70.1 min of steps left; card peak 14.3 GB", "[ 7085.6s] stage 7/8 s7_mixed step 1750/4000: train bpb s7 0.4460; quiet KL s7 0.00056; 0.73 s/step; the job: 118.1 min spent, about 69.6 min of steps left; card peak 14.3 GB", "[ 7121.5s] stage 7/8 s7_mixed step 1800/4000: train bpb s7 0.5317; quiet KL s7 0.00065; 0.73 s/step; the job: 118.7 min spent, about 69.1 min of steps left; card peak 14.3 GB", "[ 7157.5s] stage 7/8 s7_mixed step 1850/4000: train bpb s7 0.4875; quiet KL s7 0.00065; 0.73 s/step; the job: 119.3 min spent, about 68.5 min of steps left; card peak 14.3 GB", "[ 7193.4s] stage 7/8 s7_mixed step 1900/4000: train bpb s7 0.4753; quiet KL s7 0.00065; 0.73 s/step; the job: 119.9 min spent, about 68.0 min of steps left; card peak 14.3 GB", "[ 7229.1s] stage 7/8 s7_mixed step 1950/4000: train bpb s7 0.4237; quiet KL s7 0.00068; 0.73 s/step; the job: 120.5 min spent, about 67.5 min of steps left; card peak 14.3 GB", "[ 7265.0s] stage 7/8 s7_mixed step 2000/4000: train bpb s7 0.3462; quiet KL s7 0.00062; 0.73 s/step; the job: 121.1 min spent, about 66.9 min of steps left; card peak 14.3 GB", "[ 7300.8s] stage 7/8 s7_mixed step 2050/4000: train bpb s7 0.4922; quiet KL s7 0.00057; 0.73 s/step; the job: 121.7 min spent, about 66.4 min of steps left; card peak 14.3 GB", "[ 7336.8s] stage 7/8 s7_mixed step 2100/4000: train bpb s7 0.3943; quiet KL s7 0.00059; 0.73 s/step; the job: 122.3 min spent, about 65.9 min of steps left; card peak 14.3 GB", "[ 7373.1s] stage 7/8 s7_mixed step 2150/4000: train bpb s7 0.4072; quiet KL s7 0.00056; 0.73 s/step; the job: 122.9 min spent, about 65.3 min of steps left; card peak 14.3 GB", "[ 7409.3s] stage 7/8 s7_mixed step 2200/4000: train bpb s7 0.3955; quiet KL s7 0.00061; 0.73 s/step; the job: 123.5 min spent, about 64.8 min of steps left; card peak 14.3 GB", "[ 7445.6s] stage 7/8 s7_mixed step 2250/4000: train bpb s7 0.2781; quiet KL s7 0.00052; 0.73 s/step; the job: 124.1 min spent, about 64.3 min of steps left; card peak 14.3 GB", "[ 7481.8s] stage 7/8 s7_mixed step 2300/4000: train bpb s7 0.3987; quiet KL s7 0.00059; 0.73 s/step; the job: 124.7 min spent, about 63.7 min of steps left; card peak 14.3 GB", "[ 7518.0s] stage 7/8 s7_mixed step 2350/4000: train bpb s7 0.3842; quiet KL s7 0.00067; 0.73 s/step; the job: 125.3 min spent, about 63.2 min of steps left; card peak 14.3 GB", "[ 7554.2s] stage 7/8 s7_mixed step 2400/4000: train bpb s7 0.4196; quiet KL s7 0.00062; 0.73 s/step; the job: 125.9 min spent, about 62.7 min of steps left; card peak 14.3 GB", "[ 7590.3s] stage 7/8 s7_mixed step 2450/4000: train bpb s7 0.4184; quiet KL s7 0.00073; 0.73 s/step; the job: 126.5 min spent, about 62.1 min of steps left; card peak 14.3 GB", "[ 7626.3s] stage 7/8 s7_mixed step 2500/4000: train bpb s7 0.4866; quiet KL s7 0.00063; 0.73 s/step; the job: 127.1 min spent, about 61.6 min of steps left; card peak 14.3 GB", "[ 7662.2s] stage 7/8 s7_mixed step 2550/4000: train bpb s7 0.3877; quiet KL s7 0.00070; 0.73 s/step; the job: 127.7 min spent, about 61.0 min of steps left; card peak 14.3 GB", "[ 7698.1s] stage 7/8 s7_mixed step 2600/4000: train bpb s7 0.4071; quiet KL s7 0.00067; 0.73 s/step; the job: 128.3 min spent, about 60.5 min of steps left; card peak 14.3 GB", "[ 7734.0s] stage 7/8 s7_mixed step 2650/4000: train bpb s7 0.3705; quiet KL s7 0.00061; 0.73 s/step; the job: 128.9 min spent, about 60.0 min of steps left; card peak 14.3 GB", "[ 7769.9s] stage 7/8 s7_mixed step 2700/4000: train bpb s7 0.4671; quiet KL s7 0.00066; 0.73 s/step; the job: 129.5 min spent, about 59.4 min of steps left; card peak 14.3 GB", "[ 7805.8s] stage 7/8 s7_mixed step 2750/4000: train bpb s7 0.3768; quiet KL s7 0.00065; 0.73 s/step; the job: 130.1 min spent, about 58.9 min of steps left; card peak 14.3 GB", "[ 7841.7s] stage 7/8 s7_mixed step 2800/4000: train bpb s7 0.4549; quiet KL s7 0.00063; 0.73 s/step; the job: 130.7 min spent, about 58.3 min of steps left; card peak 14.3 GB", "[ 7877.6s] stage 7/8 s7_mixed step 2850/4000: train bpb s7 0.4400; quiet KL s7 0.00060; 0.73 s/step; the job: 131.3 min spent, about 57.8 min of steps left; card peak 14.3 GB", "[ 7913.6s] stage 7/8 s7_mixed step 2900/4000: train bpb s7 0.4118; quiet KL s7 0.00063; 0.73 s/step; the job: 131.9 min spent, about 57.3 min of steps left; card peak 14.3 GB", "[ 7949.5s] stage 7/8 s7_mixed step 2950/4000: train bpb s7 0.3883; quiet KL s7 0.00064; 0.73 s/step; the job: 132.5 min spent, about 56.7 min of steps left; card peak 14.3 GB", "[ 7985.5s] stage 7/8 s7_mixed step 3000/4000: train bpb s7 0.4967; quiet KL s7 0.00061; 0.73 s/step; the job: 133.1 min spent, about 56.2 min of steps left; card peak 14.3 GB", "[ 8021.5s] stage 7/8 s7_mixed step 3050/4000: train bpb s7 0.4448; quiet KL s7 0.00069; 0.73 s/step; the job: 133.7 min spent, about 55.6 min of steps left; card peak 14.3 GB", "[ 8057.5s] stage 7/8 s7_mixed step 3100/4000: train bpb s7 0.3773; quiet KL s7 0.00074; 0.73 s/step; the job: 134.3 min spent, about 55.1 min of steps left; card peak 14.3 GB", "[ 8093.4s] stage 7/8 s7_mixed step 3150/4000: train bpb s7 0.3895; quiet KL s7 0.00067; 0.73 s/step; the job: 134.9 min spent, about 54.5 min of steps left; card peak 14.3 GB", "[ 8129.4s] stage 7/8 s7_mixed step 3200/4000: train bpb s7 0.3850; quiet KL s7 0.00074; 0.73 s/step; the job: 135.5 min spent, about 54.0 min of steps left; card peak 14.3 GB", "[ 8165.3s] stage 7/8 s7_mixed step 3250/4000: train bpb s7 0.4035; quiet KL s7 0.00067; 0.73 s/step; the job: 136.1 min spent, about 53.4 min of steps left; card peak 14.3 GB", "[ 8201.2s] stage 7/8 s7_mixed step 3300/4000: train bpb s7 0.4375; quiet KL s7 0.00062; 0.73 s/step; the job: 136.7 min spent, about 52.9 min of steps left; card peak 14.3 GB", "[ 8237.1s] stage 7/8 s7_mixed step 3350/4000: train bpb s7 0.4425; quiet KL s7 0.00064; 0.73 s/step; the job: 137.3 min spent, about 52.3 min of steps left; card peak 14.3 GB", "[ 8273.1s] stage 7/8 s7_mixed step 3400/4000: train bpb s7 0.3765; quiet KL s7 0.00062; 0.73 s/step; the job: 137.9 min spent, about 51.8 min of steps left; card peak 14.3 GB", "[ 8309.0s] stage 7/8 s7_mixed step 3450/4000: train bpb s7 0.3871; quiet KL s7 0.00067; 0.73 s/step; the job: 138.5 min spent, about 51.2 min of steps left; card peak 14.3 GB", "[ 8345.0s] stage 7/8 s7_mixed step 3500/4000: train bpb s7 0.4083; quiet KL s7 0.00075; 0.73 s/step; the job: 139.1 min spent, about 50.7 min of steps left; card peak 14.3 GB", "[ 8381.0s] stage 7/8 s7_mixed step 3550/4000: train bpb s7 0.3569; quiet KL s7 0.00068; 0.73 s/step; the job: 139.7 min spent, about 50.1 min of steps left; card peak 14.3 GB", "[ 8417.0s] stage 7/8 s7_mixed step 3600/4000: train bpb s7 0.3614; quiet KL s7 0.00076; 0.73 s/step; the job: 140.3 min spent, about 49.6 min of steps left; card peak 14.3 GB", "[ 8452.9s] stage 7/8 s7_mixed step 3650/4000: train bpb s7 0.4565; quiet KL s7 0.00065; 0.73 s/step; the job: 140.9 min spent, about 49.0 min of steps left; card peak 14.3 GB", "[ 8488.9s] stage 7/8 s7_mixed step 3700/4000: train bpb s7 0.3751; quiet KL s7 0.00064; 0.73 s/step; the job: 141.5 min spent, about 48.5 min of steps left; card peak 14.3 GB", "[ 8524.9s] stage 7/8 s7_mixed step 3750/4000: train bpb s7 0.4692; quiet KL s7 0.00061; 0.73 s/step; the job: 142.1 min spent, about 47.9 min of steps left; card peak 14.3 GB", "[ 8560.9s] stage 7/8 s7_mixed step 3800/4000: train bpb s7 0.4175; quiet KL s7 0.00115; 0.73 s/step; the job: 142.7 min spent, about 47.4 min of steps left; card peak 14.3 GB", "[ 8596.9s] stage 7/8 s7_mixed step 3850/4000: train bpb s7 0.4314; quiet KL s7 0.00083; 0.73 s/step; the job: 143.3 min spent, about 46.8 min of steps left; card peak 14.3 GB", "[ 8632.9s] stage 7/8 s7_mixed step 3900/4000: train bpb s7 0.3549; quiet KL s7 0.00071; 0.73 s/step; the job: 143.9 min spent, about 46.3 min of steps left; card peak 14.3 GB", "[ 8668.6s] stage 7/8 s7_mixed step 3950/4000: train bpb s7 0.4113; quiet KL s7 0.00073; 0.73 s/step; the job: 144.5 min spent, about 45.7 min of steps left; card peak 14.3 GB", "[ 8704.6s] stage 7/8 s7_mixed step 4000/4000: train bpb s7 0.4132; quiet KL s7 0.00065; 0.73 s/step; the job: 145.1 min spent, about 45.2 min of steps left; card peak 14.3 GB", "[ 8706.0s] anchors saved: /root/refit/group/anchors/gXA_close7_.safetensors for ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail', 's7_mixed']", "[ 8915.9s] CLOSE 7/8: the group ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail', 's7_mixed'] all on, against arms off:", "[ 8915.9s] s1_perspective rows: off 0.5538 -> all on 0.0485 (-0.5053 +- 0.0048); with s1_perspective masked 0.4463 (its share -0.3978); s1_perspective alone 0.0525 (-0.5013); gate mean 0.210 / widest 0.571", "[ 8915.9s] s2_concept rows: off 0.7212 -> all on 0.1112 (-0.6099 +- 0.0038); with s2_concept masked 0.2362 (its share -0.1250); s2_concept alone 0.3503 (-0.3709); gate mean 0.277 / widest 0.498", "[ 8915.9s] s3_rules rows: off 0.9296 -> all on 0.2232 (-0.7064 +- 0.0054); with s3_rules masked 0.3144 (its share -0.0912); s3_rules alone 0.8236 (-0.1060); gate mean 0.120 / widest 0.174", "[ 8915.9s] s4_arith rows: off 1.2339 -> all on 0.3279 (-0.9060 +- 0.0062); with s4_arith masked 0.4726 (its share -0.1447); s4_arith alone 1.1267 (-0.1072); gate mean 0.108 / widest 0.159", "[ 8915.9s] s5_causal rows: off 0.6604 -> all on 0.0429 (-0.6176 +- 0.0069); with s5_causal masked 0.4558 (its share -0.4130); s5_causal alone 0.0911 (-0.5693); gate mean 0.185 / widest 0.368", "[ 8915.9s] s6_tryfail rows: off 0.6728 -> all on 0.0546 (-0.6182 +- 0.0036); with s6_tryfail masked 0.4697 (its share -0.4151); s6_tryfail alone 0.1638 (-0.5090); gate mean 0.230 / widest 0.537", "[ 8915.9s] s7_mixed rows: off 0.7830 -> all on 0.7188 (-0.0642 +- 0.0392); with s7_mixed masked 0.7748 (its share -0.0560); s7_mixed alone 0.7242 (-0.0588); gate mean 0.251 / widest 0.464", "[ 8915.9s] web holdout: off 0.9506 -> all on 0.9520 (+0.00144 +- 0.00030; the quiet bar +.012); each member's share: s1_perspective +0.00030, s2_concept -0.00018, s3_rules +0.00036, s4_arith +0.00025, s5_causal +0.00031, s6_tryfail -0.00010, s7_mixed -0.00070; alone: s1_perspective +0.00143, s2_concept +0.00105, s3_rules +0.00053, s4_arith +0.00011, s5_causal +0.00028, s6_tryfail +0.00013, s7_mixed -0.00019", "[ 8915.9s] a member's share on a PARTNER's rows (all on minus that member masked; near 0 = it stays out): s1_perspective on s2_concept -0.2226; s1_perspective on s3_rules -0.1821; s1_perspective on s4_arith -0.1268; s1_perspective on s5_causal -0.0149; s1_perspective on s6_tryfail -0.0344; s1_perspective on s7_mixed -0.0038; s2_concept on s1_perspective -0.0049; s2_concept on s3_rules -0.1078; s2_concept on s4_arith -0.1064; s2_concept on s5_causal -0.0014; s2_concept on s6_tryfail -0.0035; s2_concept on s7_mixed -0.0009; s3_rules on s1_perspective +0.0000; s3_rules on s2_concept +0.0007; s3_rules on s4_arith -0.0001; s3_rules on s5_causal -0.0000; s3_rules on s6_tryfail -0.0002; s3_rules on s7_mixed +0.0001; s4_arith on s1_perspective -0.0000; s4_arith on s2_concept -0.0006; s4_arith on s3_rules -0.0000; s4_arith on s5_causal -0.0000; s4_arith on s6_tryfail -0.0040; s4_arith on s7_mixed +0.0000; s5_causal on s1_perspective +0.0002; s5_causal on s2_concept -0.0007; s5_causal on s3_rules +0.0002; s5_causal on s4_arith +0.0001; s5_causal on s6_tryfail -0.0009; s5_causal on s7_mixed -0.0001; s6_tryfail on s1_perspective +0.0000; s6_tryfail on s2_concept +0.0004; s6_tryfail on s3_rules +0.0000; s6_tryfail on s4_arith -0.0008; s6_tryfail on s5_causal -0.0001; s6_tryfail on s7_mixed -0.0001; s7_mixed on s1_perspective -0.0003; s7_mixed on s2_concept -0.0006; s7_mixed on s3_rules -0.0009; s7_mixed on s4_arith -0.0100; s7_mixed on s5_causal -0.0001; s7_mixed on s6_tryfail -0.0003", "[ 8915.9s] a member ALONE on a partner's rows (alone minus off): s1_perspective on s2_concept -0.4817; s1_perspective on s3_rules -0.4209; s1_perspective on s4_arith -0.5113; s1_perspective on s5_causal -0.1978; s1_perspective on s6_tryfail -0.1782; s1_perspective on s7_mixed -0.0101; s2_concept on s1_perspective -0.0273; s2_concept on s3_rules -0.3108; s2_concept on s4_arith -0.3537; s2_concept on s5_causal -0.0015; s2_concept on s6_tryfail +0.0004; s2_concept on s7_mixed -0.0000; s3_rules on s1_perspective -0.0003; s3_rules on s2_concept +0.0031; s3_rules on s4_arith +0.0001; s3_rules on s5_causal +0.0071; s3_rules on s6_tryfail +0.0010; s3_rules on s7_mixed -0.0005; s4_arith on s1_perspective -0.0009; s4_arith on s2_concept -0.0004; s4_arith on s3_rules -0.0016; s4_arith on s5_causal -0.0033; s4_arith on s6_tryfail -0.0135; s4_arith on s7_mixed -0.0002; s5_causal on s1_perspective -0.0183; s5_causal on s2_concept -0.0109; s5_causal on s3_rules -0.0122; s5_causal on s4_arith -0.0091; s5_causal on s6_tryfail -0.0065; s5_causal on s7_mixed -0.0011; s6_tryfail on s1_perspective -0.0020; s6_tryfail on s2_concept +0.0012; s6_tryfail on s3_rules +0.0006; s6_tryfail on s4_arith +0.0036; s6_tryfail on s5_causal -0.0098; s6_tryfail on s7_mixed -0.0000; s7_mixed on s1_perspective -0.0568; s7_mixed on s2_concept -0.0058; s7_mixed on s3_rules -0.0617; s7_mixed on s4_arith -0.0385; s7_mixed on s5_causal -0.1028; s7_mixed on s6_tryfail -0.0802", "[ 8915.9s] the group on the rows of stages not yet attached: s8_register -0.1149", "[ 8915.9s] s1_perspective battery accuracy, off -> all on: in 0.550 -> 0.900, transfer 0.463 -> 0.412; alone: in 0.838, transfer 0.463", "[ 8915.9s] s2_concept battery accuracy, off -> all on: in 0.298 -> 0.988, transfer 0.678 -> 0.686; alone: in 0.964, transfer 0.720", "[ 8915.9s] s3_rules battery accuracy, off -> all on: in 0.155 -> 0.870, transfer 0.070 -> 0.320, held-out 0.050 -> 0.442; alone: in 0.320, transfer 0.080, held-out 0.108", "[ 8915.9s] s4_arith battery accuracy, off -> all on: in 0.344 -> 0.644, transfer 0.242 -> 0.242, held-out 0.350 -> 0.325; alone: in 0.381, transfer 0.246, held-out 0.388", "[ 8915.9s] probes mean: off 0.411 -> all on 0.419", "[ 8915.9s] compiled chain: 23 cache entries (limit 256), 0 limit warnings; card peak 14.3 GB; host RSS 11.5 GB", "[ 8919.9s] STAGE 8/8: s8_register attached over ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail', 's7_mixed'] (on, frozen); at birth the group reads exactly as before on its rows and on web text; 4000 steps of 1 task + 1 abstention chunks", "[ 8921.9s] stage 8/8 s8_register step 1/4000: train bpb s8 0.3285; quiet KL s8 0.00012; 1.94 s/step; the job: 148.7 min spent, about 45.2 min of steps left; card peak 14.4 GB", "[ 8959.0s] stage 8/8 s8_register step 50/4000: train bpb s8 0.3734; quiet KL s8 0.00015; 0.78 s/step; the job: 149.3 min spent, about 44.6 min of steps left; card peak 14.5 GB", "[ 8997.1s] stage 8/8 s8_register step 100/4000: train bpb s8 0.3524; quiet KL s8 0.00029; 0.77 s/step; the job: 150.0 min spent, about 44.1 min of steps left; card peak 14.5 GB", "[ 9035.0s] stage 8/8 s8_register step 150/4000: train bpb s8 0.3352; quiet KL s8 0.00045; 0.77 s/step; the job: 150.6 min spent, about 43.6 min of steps left; card peak 14.5 GB", "[ 9073.4s] stage 8/8 s8_register step 200/4000: train bpb s8 0.3001; quiet KL s8 0.00057; 0.77 s/step; the job: 151.2 min spent, about 43.0 min of steps left; card peak 14.5 GB", "[ 9111.4s] stage 8/8 s8_register step 250/4000: train bpb s8 0.2690; quiet KL s8 0.00075; 0.77 s/step; the job: 151.9 min spent, about 42.5 min of steps left; card peak 14.5 GB", "[ 9149.8s] stage 8/8 s8_register step 300/4000: train bpb s8 0.2341; quiet KL s8 0.00092; 0.77 s/step; the job: 152.5 min spent, about 41.9 min of steps left; card peak 14.5 GB", "[ 9187.8s] stage 8/8 s8_register step 350/4000: train bpb s8 0.1983; quiet KL s8 0.00088; 0.77 s/step; the job: 153.1 min spent, about 41.4 min of steps left; card peak 14.5 GB", "[ 9226.0s] stage 8/8 s8_register step 400/4000: train bpb s8 0.1662; quiet KL s8 0.00080; 0.77 s/step; the job: 153.8 min spent, about 40.8 min of steps left; card peak 14.5 GB", "[ 9263.8s] stage 8/8 s8_register step 450/4000: train bpb s8 0.1412; quiet KL s8 0.00079; 0.76 s/step; the job: 154.4 min spent, about 40.3 min of steps left; card peak 14.5 GB", "[ 9301.9s] stage 8/8 s8_register step 500/4000: train bpb s8 0.1173; quiet KL s8 0.00079; 0.76 s/step; the job: 155.0 min spent, about 39.7 min of steps left; card peak 14.5 GB", "[ 9339.8s] stage 8/8 s8_register step 550/4000: train bpb s8 0.0982; quiet KL s8 0.00075; 0.76 s/step; the job: 155.7 min spent, about 39.2 min of steps left; card peak 14.5 GB", "[ 9378.1s] stage 8/8 s8_register step 600/4000: train bpb s8 0.0812; quiet KL s8 0.00078; 0.76 s/step; the job: 156.3 min spent, about 38.6 min of steps left; card peak 14.5 GB", "[ 9416.2s] stage 8/8 s8_register step 650/4000: train bpb s8 0.0693; quiet KL s8 0.00076; 0.76 s/step; the job: 156.9 min spent, about 38.1 min of steps left; card peak 14.5 GB", "[ 9454.6s] stage 8/8 s8_register step 700/4000: train bpb s8 0.0593; quiet KL s8 0.00077; 0.76 s/step; the job: 157.6 min spent, about 37.5 min of steps left; card peak 14.5 GB", "[ 9492.7s] stage 8/8 s8_register step 750/4000: train bpb s8 0.0497; quiet KL s8 0.00076; 0.76 s/step; the job: 158.2 min spent, about 37.0 min of steps left; card peak 14.5 GB", "[ 9531.2s] stage 8/8 s8_register step 800/4000: train bpb s8 0.0437; quiet KL s8 0.00075; 0.76 s/step; the job: 158.9 min spent, about 36.4 min of steps left; card peak 14.5 GB", "[ 9569.4s] stage 8/8 s8_register step 850/4000: train bpb s8 0.0404; quiet KL s8 0.00069; 0.76 s/step; the job: 159.5 min spent, about 35.9 min of steps left; card peak 14.5 GB", "[ 9608.0s] stage 8/8 s8_register step 900/4000: train bpb s8 0.0361; quiet KL s8 0.00074; 0.76 s/step; the job: 160.1 min spent, about 35.3 min of steps left; card peak 14.5 GB", "[ 9646.2s] stage 8/8 s8_register step 950/4000: train bpb s8 0.0334; quiet KL s8 0.00060; 0.76 s/step; the job: 160.8 min spent, about 34.8 min of steps left; card peak 14.5 GB", "[ 9684.8s] stage 8/8 s8_register step 1000/4000: train bpb s8 0.0326; quiet KL s8 0.00065; 0.76 s/step; the job: 161.4 min spent, about 34.2 min of steps left; card peak 14.5 GB", "[ 9722.9s] stage 8/8 s8_register step 1050/4000: train bpb s8 0.0301; quiet KL s8 0.00058; 0.76 s/step; the job: 162.0 min spent, about 33.7 min of steps left; card peak 14.5 GB", "[ 9761.4s] stage 8/8 s8_register step 1100/4000: train bpb s8 0.0295; quiet KL s8 0.00055; 0.76 s/step; the job: 162.7 min spent, about 33.1 min of steps left; card peak 14.5 GB", "[ 9799.6s] stage 8/8 s8_register step 1150/4000: train bpb s8 0.0290; quiet KL s8 0.00057; 0.76 s/step; the job: 163.3 min spent, about 32.6 min of steps left; card peak 14.5 GB", "[ 9838.1s] stage 8/8 s8_register step 1200/4000: train bpb s8 0.0286; quiet KL s8 0.00055; 0.77 s/step; the job: 164.0 min spent, about 32.0 min of steps left; card peak 14.5 GB", "[ 9876.3s] stage 8/8 s8_register step 1250/4000: train bpb s8 0.0293; quiet KL s8 0.00068; 0.77 s/step; the job: 164.6 min spent, about 31.4 min of steps left; card peak 14.5 GB", "[ 9914.7s] stage 8/8 s8_register step 1300/4000: train bpb s8 0.0272; quiet KL s8 0.00054; 0.77 s/step; the job: 165.2 min spent, about 30.9 min of steps left; card peak 14.5 GB", "[ 9953.0s] stage 8/8 s8_register step 1350/4000: train bpb s8 0.0264; quiet KL s8 0.00050; 0.77 s/step; the job: 165.9 min spent, about 30.3 min of steps left; card peak 14.5 GB", "[ 9991.5s] stage 8/8 s8_register step 1400/4000: train bpb s8 0.0263; quiet KL s8 0.00048; 0.77 s/step; the job: 166.5 min spent, about 29.8 min of steps left; card peak 14.5 GB", "[10029.7s] stage 8/8 s8_register step 1450/4000: train bpb s8 0.0255; quiet KL s8 0.00047; 0.77 s/step; the job: 167.2 min spent, about 29.2 min of steps left; card peak 14.5 GB", "[10068.1s] stage 8/8 s8_register step 1500/4000: train bpb s8 0.0251; quiet KL s8 0.00046; 0.77 s/step; the job: 167.8 min spent, about 28.6 min of steps left; card peak 14.5 GB", "[10106.4s] stage 8/8 s8_register step 1550/4000: train bpb s8 0.0258; quiet KL s8 0.00045; 0.77 s/step; the job: 168.4 min spent, about 28.1 min of steps left; card peak 14.5 GB", "[10144.6s] stage 8/8 s8_register step 1600/4000: train bpb s8 0.0254; quiet KL s8 0.00044; 0.77 s/step; the job: 169.1 min spent, about 27.5 min of steps left; card peak 14.5 GB", "[10183.1s] stage 8/8 s8_register step 1650/4000: train bpb s8 0.0253; quiet KL s8 0.00042; 0.77 s/step; the job: 169.7 min spent, about 27.0 min of steps left; card peak 14.5 GB", "[10221.3s] stage 8/8 s8_register step 1700/4000: train bpb s8 0.0252; quiet KL s8 0.00038; 0.77 s/step; the job: 170.4 min spent, about 26.4 min of steps left; card peak 14.5 GB", "[10259.9s] stage 8/8 s8_register step 1750/4000: train bpb s8 0.0249; quiet KL s8 0.00040; 0.77 s/step; the job: 171.0 min spent, about 25.8 min of steps left; card peak 14.5 GB", "[10298.1s] stage 8/8 s8_register step 1800/4000: train bpb s8 0.0243; quiet KL s8 0.00041; 0.77 s/step; the job: 171.6 min spent, about 25.3 min of steps left; card peak 14.5 GB", "[10336.5s] stage 8/8 s8_register step 1850/4000: train bpb s8 0.0244; quiet KL s8 0.00037; 0.77 s/step; the job: 172.3 min spent, about 24.7 min of steps left; card peak 14.5 GB", "[10374.6s] stage 8/8 s8_register step 1900/4000: train bpb s8 0.0246; quiet KL s8 0.00037; 0.77 s/step; the job: 172.9 min spent, about 24.1 min of steps left; card peak 14.5 GB", "[10413.2s] stage 8/8 s8_register step 1950/4000: train bpb s8 0.0241; quiet KL s8 0.00041; 0.77 s/step; the job: 173.6 min spent, about 23.6 min of steps left; card peak 14.5 GB", "[10451.5s] stage 8/8 s8_register step 2000/4000: train bpb s8 0.0244; quiet KL s8 0.00039; 0.77 s/step; the job: 174.2 min spent, about 23.0 min of steps left; card peak 14.5 GB", "[10490.1s] stage 8/8 s8_register step 2050/4000: train bpb s8 0.0241; quiet KL s8 0.00044; 0.77 s/step; the job: 174.8 min spent, about 22.4 min of steps left; card peak 14.5 GB", "[10528.3s] stage 8/8 s8_register step 2100/4000: train bpb s8 0.0240; quiet KL s8 0.00038; 0.77 s/step; the job: 175.5 min spent, about 21.9 min of steps left; card peak 14.5 GB", "[10566.9s] stage 8/8 s8_register step 2150/4000: train bpb s8 0.0237; quiet KL s8 0.00038; 0.77 s/step; the job: 176.1 min spent, about 21.3 min of steps left; card peak 14.5 GB", "[10605.2s] stage 8/8 s8_register step 2200/4000: train bpb s8 0.0244; quiet KL s8 0.00039; 0.77 s/step; the job: 176.8 min spent, about 20.7 min of steps left; card peak 14.5 GB", "[10643.7s] stage 8/8 s8_register step 2250/4000: train bpb s8 0.0236; quiet KL s8 0.00035; 0.77 s/step; the job: 177.4 min spent, about 20.2 min of steps left; card peak 14.5 GB", "[10682.0s] stage 8/8 s8_register step 2300/4000: train bpb s8 0.0239; quiet KL s8 0.00041; 0.77 s/step; the job: 178.0 min spent, about 19.6 min of steps left; card peak 14.5 GB", "[10720.2s] stage 8/8 s8_register step 2350/4000: train bpb s8 0.0236; quiet KL s8 0.00038; 0.77 s/step; the job: 178.7 min spent, about 19.0 min of steps left; card peak 14.5 GB", "[10758.1s] stage 8/8 s8_register step 2400/4000: train bpb s8 0.0236; quiet KL s8 0.00035; 0.77 s/step; the job: 179.3 min spent, about 18.5 min of steps left; card peak 14.5 GB", "[10796.3s] stage 8/8 s8_register step 2450/4000: train bpb s8 0.0232; quiet KL s8 0.00044; 0.77 s/step; the job: 179.9 min spent, about 17.9 min of steps left; card peak 14.5 GB", "[10834.3s] stage 8/8 s8_register step 2500/4000: train bpb s8 0.0238; quiet KL s8 0.00035; 0.77 s/step; the job: 180.6 min spent, about 17.3 min of steps left; card peak 14.5 GB", "[10872.6s] stage 8/8 s8_register step 2550/4000: train bpb s8 0.0235; quiet KL s8 0.00039; 0.77 s/step; the job: 181.2 min spent, about 16.8 min of steps left; card peak 14.5 GB", "[10910.5s] stage 8/8 s8_register step 2600/4000: train bpb s8 0.0236; quiet KL s8 0.00033; 0.77 s/step; the job: 181.8 min spent, about 16.2 min of steps left; card peak 14.5 GB", "[10948.7s] stage 8/8 s8_register step 2650/4000: train bpb s8 0.0230; quiet KL s8 0.00037; 0.77 s/step; the job: 182.5 min spent, about 15.6 min of steps left; card peak 14.5 GB", "[10986.6s] stage 8/8 s8_register step 2700/4000: train bpb s8 0.0229; quiet KL s8 0.00033; 0.77 s/step; the job: 183.1 min spent, about 15.0 min of steps left; card peak 14.5 GB", "[11024.8s] stage 8/8 s8_register step 2750/4000: train bpb s8 0.0234; quiet KL s8 0.00033; 0.77 s/step; the job: 183.7 min spent, about 14.5 min of steps left; card peak 14.5 GB", "[11062.8s] stage 8/8 s8_register step 2800/4000: train bpb s8 0.0231; quiet KL s8 0.00034; 0.77 s/step; the job: 184.4 min spent, about 13.9 min of steps left; card peak 14.5 GB", "[11100.7s] stage 8/8 s8_register step 2850/4000: train bpb s8 0.0234; quiet KL s8 0.00035; 0.77 s/step; the job: 185.0 min spent, about 13.3 min of steps left; card peak 14.5 GB", "[11138.9s] stage 8/8 s8_register step 2900/4000: train bpb s8 0.0230; quiet KL s8 0.00035; 0.77 s/step; the job: 185.6 min spent, about 12.7 min of steps left; card peak 14.5 GB", "[11177.0s] stage 8/8 s8_register step 2950/4000: train bpb s8 0.0231; quiet KL s8 0.00034; 0.77 s/step; the job: 186.3 min spent, about 12.2 min of steps left; card peak 14.5 GB", "[11215.2s] stage 8/8 s8_register step 3000/4000: train bpb s8 0.0230; quiet KL s8 0.00032; 0.77 s/step; the job: 186.9 min spent, about 11.6 min of steps left; card peak 14.5 GB", "[11253.1s] stage 8/8 s8_register step 3050/4000: train bpb s8 0.0230; quiet KL s8 0.00035; 0.76 s/step; the job: 187.6 min spent, about 11.0 min of steps left; card peak 14.5 GB", "[11291.6s] stage 8/8 s8_register step 3100/4000: train bpb s8 0.0228; quiet KL s8 0.00036; 0.77 s/step; the job: 188.2 min spent, about 10.4 min of steps left; card peak 14.5 GB", "[11329.9s] stage 8/8 s8_register step 3150/4000: train bpb s8 0.0235; quiet KL s8 0.00035; 0.77 s/step; the job: 188.8 min spent, about 9.9 min of steps left; card peak 14.5 GB", "[11368.0s] stage 8/8 s8_register step 3200/4000: train bpb s8 0.0232; quiet KL s8 0.00036; 0.77 s/step; the job: 189.5 min spent, about 9.3 min of steps left; card peak 14.5 GB", "[11405.8s] stage 8/8 s8_register step 3250/4000: train bpb s8 0.0226; quiet KL s8 0.00036; 0.76 s/step; the job: 190.1 min spent, about 8.7 min of steps left; card peak 14.5 GB", "[11443.8s] stage 8/8 s8_register step 3300/4000: train bpb s8 0.0227; quiet KL s8 0.00034; 0.76 s/step; the job: 190.7 min spent, about 8.1 min of steps left; card peak 14.5 GB", "[11481.8s] stage 8/8 s8_register step 3350/4000: train bpb s8 0.0230; quiet KL s8 0.00034; 0.76 s/step; the job: 191.4 min spent, about 7.5 min of steps left; card peak 14.5 GB", "[11520.0s] stage 8/8 s8_register step 3400/4000: train bpb s8 0.0231; quiet KL s8 0.00033; 0.76 s/step; the job: 192.0 min spent, about 7.0 min of steps left; card peak 14.5 GB", "[11557.9s] stage 8/8 s8_register step 3450/4000: train bpb s8 0.0225; quiet KL s8 0.00037; 0.76 s/step; the job: 192.6 min spent, about 6.4 min of steps left; card peak 14.5 GB", "[11596.1s] stage 8/8 s8_register step 3500/4000: train bpb s8 0.0229; quiet KL s8 0.00032; 0.76 s/step; the job: 193.3 min spent, about 5.8 min of steps left; card peak 14.5 GB", "[11634.0s] stage 8/8 s8_register step 3550/4000: train bpb s8 0.0228; quiet KL s8 0.00032; 0.76 s/step; the job: 193.9 min spent, about 5.2 min of steps left; card peak 14.5 GB", "[11672.3s] stage 8/8 s8_register step 3600/4000: train bpb s8 0.0223; quiet KL s8 0.00032; 0.76 s/step; the job: 194.5 min spent, about 4.7 min of steps left; card peak 14.5 GB", "[11710.5s] stage 8/8 s8_register step 3650/4000: train bpb s8 0.0220; quiet KL s8 0.00032; 0.76 s/step; the job: 195.2 min spent, about 4.1 min of steps left; card peak 14.5 GB", "[11749.0s] stage 8/8 s8_register step 3700/4000: train bpb s8 0.0222; quiet KL s8 0.00033; 0.76 s/step; the job: 195.8 min spent, about 3.5 min of steps left; card peak 14.5 GB", "[11787.0s] stage 8/8 s8_register step 3750/4000: train bpb s8 0.0226; quiet KL s8 0.00031; 0.76 s/step; the job: 196.5 min spent, about 2.9 min of steps left; card peak 14.5 GB", "[11825.0s] stage 8/8 s8_register step 3800/4000: train bpb s8 0.0221; quiet KL s8 0.00035; 0.76 s/step; the job: 197.1 min spent, about 2.3 min of steps left; card peak 14.5 GB", "[11863.2s] stage 8/8 s8_register step 3850/4000: train bpb s8 0.0221; quiet KL s8 0.00031; 0.76 s/step; the job: 197.7 min spent, about 1.7 min of steps left; card peak 14.5 GB", "[11901.2s] stage 8/8 s8_register step 3900/4000: train bpb s8 0.0225; quiet KL s8 0.00030; 0.76 s/step; the job: 198.4 min spent, about 1.2 min of steps left; card peak 14.5 GB", "[11939.4s] stage 8/8 s8_register step 3950/4000: train bpb s8 0.0228; quiet KL s8 0.00029; 0.76 s/step; the job: 199.0 min spent, about 0.6 min of steps left; card peak 14.5 GB", "[11977.4s] stage 8/8 s8_register step 4000/4000: train bpb s8 0.0226; quiet KL s8 0.00031; 0.76 s/step; the job: 199.6 min spent, about 0.0 min of steps left; card peak 14.5 GB", "[11978.7s] anchors saved: /root/refit/group/anchors/gXA_close8_.safetensors for ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail', 's7_mixed', 's8_register']", "[12220.2s] CLOSE 8/8: the group ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail', 's7_mixed', 's8_register'] all on, against arms off:", "[12220.2s] s1_perspective rows: off 0.5538 -> all on 0.0492 (-0.5046 +- 0.0048); with s1_perspective masked 0.4324 (its share -0.3832); s1_perspective alone 0.0525 (-0.5013); gate mean 0.210 / widest 0.571", "[12220.2s] s2_concept rows: off 0.7212 -> all on 0.1105 (-0.6107 +- 0.0038); with s2_concept masked 0.2307 (its share -0.1202); s2_concept alone 0.3503 (-0.3709); gate mean 0.277 / widest 0.498", "[12220.2s] s3_rules rows: off 0.9296 -> all on 0.2237 (-0.7059 +- 0.0053); with s3_rules masked 0.3137 (its share -0.0900); s3_rules alone 0.8236 (-0.1060); gate mean 0.120 / widest 0.174", "[12220.2s] s4_arith rows: off 1.2339 -> all on 0.3279 (-0.9059 +- 0.0062); with s4_arith masked 0.4711 (its share -0.1431); s4_arith alone 1.1267 (-0.1072); gate mean 0.108 / widest 0.159", "[12220.2s] s5_causal rows: off 0.6604 -> all on 0.0431 (-0.6174 +- 0.0069); with s5_causal masked 0.4581 (its share -0.4150); s5_causal alone 0.0911 (-0.5693); gate mean 0.185 / widest 0.368", "[12220.2s] s6_tryfail rows: off 0.6728 -> all on 0.0548 (-0.6180 +- 0.0037); with s6_tryfail masked 0.4686 (its share -0.4138); s6_tryfail alone 0.1638 (-0.5090); gate mean 0.230 / widest 0.537", "[12220.2s] s7_mixed rows: off 0.7830 -> all on 0.7200 (-0.0630 +- 0.0390); with s7_mixed masked 0.7768 (its share -0.0568); s7_mixed alone 0.7242 (-0.0588); gate mean 0.251 / widest 0.464", "[12220.2s] s8_register rows: off 0.4907 -> all on 0.0212 (-0.4695 +- 0.0085); with s8_register masked 0.3757 (its share -0.3546); s8_register alone 0.0678 (-0.4228); gate mean 0.203 / widest 0.452", "[12220.2s] web holdout: off 0.9506 -> all on 0.9525 (+0.00189 +- 0.00032; the quiet bar +.012); each member's share: s1_perspective +0.00025, s2_concept -0.00043, s3_rules +0.00009, s4_arith +0.00020, s5_causal +0.00035, s6_tryfail -0.00025, s7_mixed -0.00096, s8_register +0.00045; alone: s1_perspective +0.00143, s2_concept +0.00105, s3_rules +0.00053, s4_arith +0.00011, s5_causal +0.00028, s6_tryfail +0.00013, s7_mixed -0.00019, s8_register +0.00048", "[12220.2s] a member's share on a PARTNER's rows (all on minus that member masked; near 0 = it stays out): s1_perspective on s2_concept -0.2174; s1_perspective on s3_rules -0.1832; s1_perspective on s4_arith -0.1280; s1_perspective on s5_causal -0.0128; s1_perspective on s6_tryfail -0.0280; s1_perspective on s7_mixed -0.0029; s1_perspective on s8_register -0.0272; s2_concept on s1_perspective -0.0057; s2_concept on s3_rules -0.1086; s2_concept on s4_arith -0.1060; s2_concept on s5_causal -0.0013; s2_concept on s6_tryfail -0.0035; s2_concept on s7_mixed -0.0008; s2_concept on s8_register -0.0007; s3_rules on s1_perspective +0.0001; s3_rules on s2_concept +0.0005; s3_rules on s4_arith -0.0002; s3_rules on s5_causal -0.0001; s3_rules on s6_tryfail -0.0003; s3_rules on s7_mixed +0.0001; s3_rules on s8_register -0.0001; s4_arith on s1_perspective +0.0000; s4_arith on s2_concept -0.0006; s4_arith on s3_rules -0.0001; s4_arith on s5_causal -0.0000; s4_arith on s6_tryfail -0.0036; s4_arith on s7_mixed +0.0001; s4_arith on s8_register -0.0000; s5_causal on s1_perspective +0.0007; s5_causal on s2_concept -0.0005; s5_causal on s3_rules +0.0005; s5_causal on s4_arith +0.0002; s5_causal on s6_tryfail -0.0005; s5_causal on s7_mixed +0.0004; s5_causal on s8_register -0.0003; s6_tryfail on s1_perspective -0.0001; s6_tryfail on s2_concept +0.0002; s6_tryfail on s3_rules -0.0002; s6_tryfail on s4_arith -0.0009; s6_tryfail on s5_causal -0.0001; s6_tryfail on s7_mixed +0.0001; s6_tryfail on s8_register -0.0006; s7_mixed on s1_perspective -0.0004; s7_mixed on s2_concept -0.0005; s7_mixed on s3_rules -0.0010; s7_mixed on s4_arith -0.0103; s7_mixed on s5_causal -0.0002; s7_mixed on s6_tryfail -0.0004; s7_mixed on s8_register -0.0001; s8_register on s1_perspective +0.0007; s8_register on s2_concept -0.0008; s8_register on s3_rules +0.0005; s8_register on s4_arith +0.0001; s8_register on s5_causal +0.0002; s8_register on s6_tryfail +0.0002; s8_register on s7_mixed +0.0012", "[12220.2s] a member ALONE on a partner's rows (alone minus off): s1_perspective on s2_concept -0.4817; s1_perspective on s3_rules -0.4209; s1_perspective on s4_arith -0.5113; s1_perspective on s5_causal -0.1978; s1_perspective on s6_tryfail -0.1782; s1_perspective on s7_mixed -0.0101; s1_perspective on s8_register -0.1119; s2_concept on s1_perspective -0.0273; s2_concept on s3_rules -0.3108; s2_concept on s4_arith -0.3537; s2_concept on s5_causal -0.0015; s2_concept on s6_tryfail +0.0004; s2_concept on s7_mixed -0.0000; s2_concept on s8_register -0.0215; s3_rules on s1_perspective -0.0003; s3_rules on s2_concept +0.0031; s3_rules on s4_arith +0.0001; s3_rules on s5_causal +0.0071; s3_rules on s6_tryfail +0.0010; s3_rules on s7_mixed -0.0005; s3_rules on s8_register +0.0002; s4_arith on s1_perspective -0.0009; s4_arith on s2_concept -0.0004; s4_arith on s3_rules -0.0016; s4_arith on s5_causal -0.0033; s4_arith on s6_tryfail -0.0135; s4_arith on s7_mixed -0.0002; s4_arith on s8_register -0.0015; s5_causal on s1_perspective -0.0183; s5_causal on s2_concept -0.0109; s5_causal on s3_rules -0.0122; s5_causal on s4_arith -0.0091; s5_causal on s6_tryfail -0.0065; s5_causal on s7_mixed -0.0011; s5_causal on s8_register -0.0064; s6_tryfail on s1_perspective -0.0020; s6_tryfail on s2_concept +0.0012; s6_tryfail on s3_rules +0.0006; s6_tryfail on s4_arith +0.0036; s6_tryfail on s5_causal -0.0098; s6_tryfail on s7_mixed -0.0000; s6_tryfail on s8_register -0.0003; s7_mixed on s1_perspective -0.0568; s7_mixed on s2_concept -0.0058; s7_mixed on s3_rules -0.0617; s7_mixed on s4_arith -0.0385; s7_mixed on s5_causal -0.1028; s7_mixed on s6_tryfail -0.0802; s7_mixed on s8_register -0.0080; s8_register on s1_perspective -0.0172; s8_register on s2_concept -0.0079; s8_register on s3_rules -0.0048; s8_register on s4_arith -0.0065; s8_register on s5_causal -0.0189; s8_register on s6_tryfail -0.0220; s8_register on s7_mixed -0.0002", "[12220.2s] s1_perspective battery accuracy, off -> all on: in 0.550 -> 0.887, transfer 0.463 -> 0.438; alone: in 0.838, transfer 0.463", "[12220.2s] s2_concept battery accuracy, off -> all on: in 0.298 -> 0.976, transfer 0.678 -> 0.720; alone: in 0.964, transfer 0.720", "[12220.2s] s3_rules battery accuracy, off -> all on: in 0.155 -> 0.855, transfer 0.070 -> 0.310, held-out 0.050 -> 0.433; alone: in 0.320, transfer 0.080, held-out 0.108", "[12220.2s] s4_arith battery accuracy, off -> all on: in 0.344 -> 0.644, transfer 0.242 -> 0.242, held-out 0.350 -> 0.425; alone: in 0.381, transfer 0.246, held-out 0.388", "[12220.2s] probes mean: off 0.411 -> all on 0.422", "[12220.2s] compiled chain: 28 cache entries (limit 256), 0 limit warnings; card peak 14.5 GB; host RSS 11.6 GB", "[12223.7s] the write of s1_perspective (mean |delta| / |x|, all members live): own rows 1.63e-02; web 6.65e-03 (own over web 2.44x); partners' rows s2_concept 1.27e-02, s3_rules 1.43e-02, s4_arith 1.51e-02, s5_causal 1.39e-02, s6_tryfail 1.39e-02, s7_mixed 6.62e-03, s8_register 1.28e-02", "[12223.7s] the write of s2_concept (mean |delta| / |x|, all members live): own rows 1.14e-02; web 4.36e-03 (own over web 2.61x); partners' rows s1_perspective 8.64e-03, s3_rules 1.11e-02, s4_arith 1.17e-02, s5_causal 8.99e-03, s6_tryfail 9.33e-03, s7_mixed 4.44e-03, s8_register 8.74e-03", "[12223.7s] the write of s3_rules (mean |delta| / |x|, all members live): own rows 8.02e-03; web 1.41e-03 (own over web 5.69x); partners' rows s1_perspective 1.77e-03, s2_concept 1.77e-03, s4_arith 1.33e-03, s5_causal 1.98e-03, s6_tryfail 1.81e-03, s7_mixed 1.40e-03, s8_register 1.72e-03", "[12223.7s] the write of s4_arith (mean |delta| / |x|, all members live): own rows 7.78e-03; web 1.06e-03 (own over web 7.37x); partners' rows s1_perspective 1.50e-03, s2_concept 1.30e-03, s3_rules 1.41e-03, s5_causal 1.95e-03, s6_tryfail 3.49e-03, s7_mixed 1.07e-03, s8_register 1.60e-03", "[12223.7s] the write of s5_causal (mean |delta| / |x|, all members live): own rows 1.62e-02; web 3.10e-03 (own over web 5.23x); partners' rows s1_perspective 5.62e-03, s2_concept 3.61e-03, s3_rules 4.00e-03, s4_arith 3.83e-03, s6_tryfail 4.88e-03, s7_mixed 3.13e-03, s8_register 5.08e-03", "[12223.7s] the write of s6_tryfail (mean |delta| / |x|, all members live): own rows 1.91e-02; web 4.97e-03 (own over web 3.84x); partners' rows s1_perspective 6.86e-03, s2_concept 5.46e-03, s3_rules 6.16e-03, s4_arith 7.01e-03, s5_causal 7.47e-03, s7_mixed 4.95e-03, s8_register 7.23e-03", "[12223.7s] the write of s7_mixed (mean |delta| / |x|, all members live): own rows 4.20e-03; web 4.29e-03 (own over web 0.98x); partners' rows s1_perspective 7.05e-03, s2_concept 5.80e-03, s3_rules 8.15e-03, s4_arith 6.69e-03, s5_causal 6.33e-03, s6_tryfail 6.51e-03, s8_register 5.64e-03", "[12223.7s] the write of s8_register (mean |delta| / |x|, all members live): own rows 1.58e-02; web 3.65e-03 (own over web 4.35x); partners' rows s1_perspective 5.72e-03, s2_concept 4.29e-03, s3_rules 4.27e-03, s4_arith 4.54e-03, s5_causal 6.14e-03, s6_tryfail 5.95e-03, s7_mixed 3.60e-03"], "write_matrix": {"s1_perspective": {"s1_perspective": 0.016259906124105328, "s2_concept": 0.012682219906764658, "s3_rules": 0.014342284986923914, "s4_arith": 0.015080223842232954, "s5_causal": 0.01386075758637162, "s6_tryfail": 0.013927268690167693, "s7_mixed": 0.006622602681090939, "s8_register": 0.012757788877934217, "web": 0.0066530740668895305}, "s2_concept": {"s1_perspective": 0.008635623464670061, "s2_concept": 0.011405108652638773, "s3_rules": 0.011130451526923935, "s4_arith": 0.011651348579334808, "s5_causal": 0.008991965344193886, "s6_tryfail": 0.009325220954906399, "s7_mixed": 0.004444046860385242, "s8_register": 0.008743224692352669, "web": 0.004364331670785759}, "s3_rules": {"s1_perspective": 0.001768741619798675, "s2_concept": 0.0017706412018014817, "s3_rules": 0.008017134199462816, "s4_arith": 0.0013297268965288822, "s5_causal": 0.0019758822627409245, "s6_tryfail": 0.001807058964004682, "s7_mixed": 0.0013973523007280164, "s8_register": 0.0017157856054836884, "web": 0.0014097344737820094}, "s4_arith": {"s1_perspective": 0.0014966278313295334, "s2_concept": 0.0012962376918039809, "s3_rules": 0.0014111982163740322, "s4_arith": 0.007781411268297234, "s5_causal": 0.0019459262330201454, "s6_tryfail": 0.003492681495117722, "s7_mixed": 0.0010707704118431138, "s8_register": 0.0015968359939506627, "web": 0.0010562455399849568}, "s5_causal": {"s1_perspective": 0.005616438273136737, "s2_concept": 0.0036103186757827643, "s3_rules": 0.003996874850599852, "s4_arith": 0.0038280392454907997, "s5_causal": 0.016218368175032083, "s6_tryfail": 0.004876768260146491, "s7_mixed": 0.0031346824907814153, "s8_register": 0.0050818327435990795, "web": 0.0031035011031690374}, "s6_tryfail": {"s1_perspective": 0.006858849967102287, "s2_concept": 0.005455568631987262, "s3_rules": 0.0061557409089800785, "s4_arith": 0.0070096770468808245, "s5_causal": 0.00746873772732215, "s6_tryfail": 0.019121937890304253, "s7_mixed": 0.004953494948495063, "s8_register": 0.007229927094158484, "web": 0.00497367834122997}, "s7_mixed": {"s1_perspective": 0.0070506989541172516, "s2_concept": 0.005799024595944502, "s3_rules": 0.0081541325962462, "s4_arith": 0.006685453878162662, "s5_causal": 0.006333926487059216, "s6_tryfail": 0.006509710477985209, "s7_mixed": 0.0041998932656497345, "s8_register": 0.005640076406052685, "web": 0.00428722212654975}, "s8_register": {"s1_perspective": 0.005716469840990612, "s2_concept": 0.00429300589939885, "s3_rules": 0.004270069468475413, "s4_arith": 0.004537898972557741, "s5_causal": 0.006136119322036393, "s6_tryfail": 0.005953655916528078, "s7_mixed": 0.0035969439509244694, "s8_register": 0.01584133654978359, "web": 0.0036454277942539193}}, "finished": true} \ No newline at end of file diff --git a/mini-beatrix-3/arm_refit/group/results/group_gXB.json b/mini-beatrix-3/arm_refit/group/results/group_gXB.json new file mode 100644 index 0000000000000000000000000000000000000000..9717a8091ba45f7a9bf72a29621948b39727626c --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/results/group_gXB.json @@ -0,0 +1 @@ +{"run": "gXB", "stages": ["s1_perspective", "s2_concept", "s3_rules", "s4_arith", "s5_causal", "s6_tryfail", "s7_mixed", "s8_register"], "seed_offset": 100, "data_seed": 20261012, "steps_per_stage": 4000, "ctx": 4096, "chunk_rows": 2, "lambda": 2.0, "lr": 0.001, "spec": {"n_slots": 16, "K": 16, "D": 8, "hidden": 256}, "trunk_step": 245674, "smoke": false, "resume": "gRB:4", "no_partner_rows_of": ["s7_mixed"], "freeze_earlier": true, "resume_members": 4, "offdomain": "roster", "options": "the first 4 members loaded from gRB:4 (the later members attach fresh); a member's partner rows come from every other member of the roster, attached or not", "form": "staged group, earlier arms frozen: each stage's arm attaches fresh over the arms already attached, which stay on and stop training; per step the newest member takes one task chunk of its own stage text (all on) and one abstention chunk (one web row + one row of a partner's stage text; KL from this member masked, partners as is, to all on); step loss = task + lambda * KL", "off": {"rows": {"s1_perspective": {"mean": 0.5538333013185592, "se": 0.005481125510495537, "per_batch": [0.5580802623967968, 0.5727959149222103, 0.5683326211670187, 0.5592622991561769, 0.5403289870966665, 0.5581498723747335, 0.5248631032525267, 0.5488533501823444]}, "s2_concept": {"mean": 0.7211839326967642, "se": 0.0036750359411895426, "per_batch": [0.7178970755158839, 0.7206461751942632, 0.7084789615893348, 0.7429370185552426, 0.7197164369836988, 0.7196987227706599, 0.7128483098214926, 0.7272487611435373]}, "s3_rules": {"mean": 0.9296131361815035, "se": 0.006694211546016496, "per_batch": [0.9087232204987897, 0.9265088063377671, 0.9502096494618454, 0.9083145037290143, 0.9115083935381875, 0.9565801448237801, 0.9359078301900705, 0.9391525408725736]}, "s4_arith": {"mean": 1.2338748670942346, "se": 0.013664467393095502, "per_batch": [1.2484452802709944, 1.2007697976426643, 1.2552932854630496, 1.2180985975609502, 1.162178696606884, 1.2806971007959715, 1.2668025364414397, 1.2387136419719225]}, "s5_causal": {"mean": 0.6604376826115352, "se": 0.007017957355969411, "per_batch": [0.625201307410232, 0.6701150926310071, 0.6715778910679207, 0.6723486313177623, 0.6862873952135659, 0.6580370090348883, 0.6384860212848043, 0.6614481129321013]}, "s6_tryfail": {"mean": 0.6728212611401647, "se": 0.004824708124390516, "per_batch": [0.6802316281027101, 0.6945571809803041, 0.6730548350778678, 0.6813304682547353, 0.6768184174180241, 0.6671325694997442, 0.6543597189898025, 0.65508527079813]}, "s7_mixed": {"mean": 0.7829759286650222, "se": 0.08891676616480938, "per_batch": [1.0719286507834895, 1.0487072092986913, 0.9167929044512491, 0.8095931944633957, 0.8840941010166795, 0.6359224908867059, 0.36390978578087546, 0.5328590926390914]}, "s8_register": {"mean": 0.4906504977761471, "se": 0.009077126336788649, "per_batch": [0.4430003073464096, 0.5108862451995572, 0.478870513858552, 0.5101862758105455, 0.4840021751906592, 0.5198495510058996, 0.5052149023090424, 0.4731940114885114]}, "web": {"mean": 0.9505967609111062, "se": 0.038152120989285714, "per_batch": [1.1091405369506755, 0.8791101297860008, 0.8925084382014679, 0.8542297436509799, 1.1341640986425394, 0.9075658632498085, 0.9278510449364259, 0.9002042318709514]}}, "battery_acc": {"s1_perspective": {"in": 0.55, "transfer": 0.4625}, "s2_concept": {"in": 0.2976190476190476, "transfer": 0.6779661016949152}, "s3_rules": {"in": 0.155, "transfer": 0.07, "held-out": 0.05}, "s4_arith": {"in": 0.34375, "transfer": 0.24166666666666667, "held-out": 0.35}}, "probes": {"p0_grounding": {"acc": 0.36666666666666664, "n": 30, "families": {"count": 0.125, "permanence": 0.7142857142857143, "sequence": 0.25, "attribute": 0.42857142857142855}}, "p1_self_other": {"acc": 0.5, "n": 30, "families": {"belief": 0.375, "reality": 0.6666666666666666, "perspective": 0.6, "reference": 0.7142857142857143, "second_order": 0.0}}, "p2_difference": {"acc": 0.5666666666666667, "n": 30, "families": {"category": 0.8333333333333334, "antonym": 0.6666666666666666, "odd_one_out": 0.6666666666666666, "property": 0.5, "analogy": 0.16666666666666666}}, "p3_logic": {"acc": 0.16666666666666666, "n": 30, "families": {"d1": 0.5, "d2": 0.0, "d3": 0.5, "d4": 0.0, "d5": 0.0, "d6": 0.0, "negation": 0.0, "contrapositive": 0.25}}, "p4_arithmetic": {"acc": 0.3, "n": 30, "families": {"add2": 0.0, "add3": 0.0, "sub2": 0.5, "sub3": 0.6, "mul1": 0.25, "mul2": 0.3333333333333333, "word": 0.4}}, "p5_causal": {"acc": 0.5666666666666667, "n": 30, "families": {"effect": 0.7142857142857143, "cause": 0.42857142857142855, "counterfactual": 0.5, "prevention": 0.4, "chain": 0.8}}, "p6_learning": {"acc": 0.36666666666666664, "n": 30, "families": {"error_detect": 0.25, "correction": 0.25, "strategy": 0.7142857142857143, "check": 0.2857142857142857}}, "p7_composite": {"acc": 0.3333333333333333, "n": 30, "families": {"logic_arith": 0.25, "causal_category": 0.375, "belief_causal": 0.2857142857142857, "three_step": 0.42857142857142855}}, "p8_articulation": {"acc": 0.5333333333333333, "n": 30, "families": {"definition": 0.5, "register": 0.5714285714285714, "grammar": 0.875, "paraphrase": 0.14285714285714285}}}, "probes_mean": 0.41111111111111115}, "closes": [{"close": 4, "attached": ["s1_perspective", "s2_concept", "s3_rules", "s4_arith"], "rows": {"s1_perspective": {"off": 0.5538333013185592, "all_on": 0.047862855039216154, "all_on_minus_off": {"mean": -0.5059704462793431, "se": 0.004823672924823277, "per_batch": [-0.5076920060506244, -0.525351802703869, -0.517005817873883, -0.5117583799709982, -0.4928881855443545, -0.5080673743095059, -0.48176060706389173, -0.5032393967176176]}}, "s2_concept": {"off": 0.7211839326967642, "all_on": 0.12117190301141822, "all_on_minus_off": {"mean": -0.600012029685346, "se": 0.003718090090662156, "per_batch": [-0.5987170218221803, -0.5979946839396408, -0.5873827954377903, -0.622315975974538, -0.5987677889509319, -0.5986096509034635, -0.5910468965633381, -0.6052614238908844]}}, "s3_rules": {"off": 0.9296131361815035, "all_on": 0.22223902876251628, "all_on_minus_off": {"mean": -0.7073741074189872, "se": 0.005422662904359449, "per_batch": [-0.6880081246311109, -0.7060808999991979, -0.7221312453844636, -0.6910756071919001, -0.6935751600438773, -0.7298150432640493, -0.7135861583893568, -0.7147206204479419]}}, "s4_arith": {"off": 1.2338748670942346, "all_on": 0.33787596037663836, "all_on_minus_off": {"mean": -0.8959989067175962, "se": 0.005857920648093876, "per_batch": [-0.9124769568321842, -0.8875824466219613, -0.9114362038204877, -0.8715034016318329, -0.8760851484378062, -0.9143963692071386, -0.8984319076581474, -0.8960788195312113]}}, "s5_causal": {"off": 0.6604376826115352, "all_on": 0.48561756523455085, "all_on_minus_off": {"mean": -0.17482011737698436, "se": 0.0029811798753642225, "per_batch": [-0.16453666199370337, -0.18111075904434037, -0.186119237802767, -0.17376662690020855, -0.17715545904413477, -0.16182922511249553, -0.17256696191911503, -0.18147600719911022]}}, "s6_tryfail": {"off": 0.6728212611401647, "all_on": 0.4878494485882916, "all_on_minus_off": {"mean": -0.18497181255187314, "se": 0.003714955467444921, "per_batch": [-0.18539660969950428, -0.2051020808657259, -0.17944803077579874, -0.1946919710090001, -0.18664885837609912, -0.17727077341153824, -0.17403152317820003, -0.17718465309911874]}}, "s7_mixed": {"off": 0.7829759286650222, "all_on": 0.7769350810491288, "all_on_minus_off": {"mean": -0.006040847615893442, "se": 0.00794014237493403, "per_batch": [0.000961555000974057, 0.0014185988956418871, 0.001284882384596009, 0.001317559088260034, -0.0007160497668673438, -0.0597048951342547, 0.014178980663396124, -0.007067412058893607]}}, "s8_register": {"off": 0.4906504977761471, "all_on": 0.38026790218154694, "all_on_minus_off": {"mean": -0.1103825955946002, "se": 0.0022388773591315125, "per_batch": [-0.09819450490819442, -0.11392999305539303, -0.10747948276514435, -0.11145615160971967, -0.10897405499680485, -0.12020658588995664, -0.11357609575558003, -0.10924389577600863]}}, "web": {"off": 0.9505967609111062, "all_on": 0.951700233845789, "all_on_minus_off": {"mean": 0.0011034729346828115, "se": 0.0003934523248249499, "per_batch": [0.0002704427184816449, 0.0001944263868003393, 3.989997500020159e-05, 0.00035652003523867215, 0.0026083748743108703, 0.0028633391542148834, 0.0010795350994659003, 0.00141524523394998]}}}, "member_in_group": {"s1_perspective": {"s1_perspective": {"masked": 0.5211330414059152, "all_on_minus_masked": {"mean": -0.47327018636669904, "se": 0.004689062698412304, "per_batch": [-0.4788197306064269, -0.4904067768620432, -0.4859485458461364, -0.4772327768189639, -0.4547830644842498, -0.47405535434861074, -0.4536777319871546, -0.4712375099800069]}}, "s2_concept": {"masked": 0.359174009700452, "all_on_minus_masked": {"mean": -0.23800210668903377, "se": 0.0015381030335393337, "per_batch": [-0.23638749862875777, -0.2387091246795019, -0.2313158054972233, -0.24544863031669412, -0.2366284355737009, -0.23781975939622757, -0.23525884098463923, -0.2424487584355255]}}, "s3_rules": {"masked": 0.4173434410405745, "all_on_minus_masked": {"mean": -0.19510441227805825, "se": 0.0029604857516389417, "per_batch": [-0.193632428892697, -0.19958605980677166, -0.19552226025170297, -0.18527153531666427, -0.19749051569950238, -0.21194901116621329, -0.1887918481971751, -0.18859163889373948]}}, "s4_arith": {"masked": 0.49698961167875727, "all_on_minus_masked": {"mean": -0.15911365130211888, "se": 0.004948485212035152, "per_batch": [-0.1680660469590472, -0.17766317332731313, -0.1307859696707942, -0.16126733629429546, -0.16889111222873016, -0.15636752907338264, -0.15590760446931273, -0.15396043839407542]}}, "web": {"masked": 0.9516641927315175, "all_on_minus_masked": {"mean": 3.604111427141232e-05, "se": 0.0003809923053268703, "per_batch": [0.00029323041972095787, -0.0007857887317922474, -0.0012215927690785167, -0.0013614146642300096, 0.0014377749612131119, 0.0011326777385823883, 0.0008576774798528808, -6.423552009726663e-05]}}}, "s2_concept": {"s1_perspective": {"masked": 0.05194227344057353, "all_on_minus_masked": {"mean": -0.004079418401357371, "se": 0.0003573019997286657, "per_batch": [-0.004327749928480361, -0.004211113444448335, -0.004391770470264039, -0.0030456676364186697, -0.006157591589082252, -0.003741450322773364, -0.0028837083494264656, -0.003876295469965485]}}, "s2_concept": {"masked": 0.24241949451258962, "all_on_minus_masked": {"mean": -0.1212475915011714, "se": 0.0014385494084546994, "per_batch": [-0.12685179128989768, -0.12412858574938891, -0.11767001377238108, -0.12602099684815785, -0.12119677600229928, -0.11988682714633998, -0.11560455588009715, -0.11862118532080919]}}, "s3_rules": {"masked": 0.3226985881782366, "all_on_minus_masked": {"mean": -0.10045955941572031, "se": 0.001013638599279675, "per_batch": [-0.1011096753338131, -0.09923835360760622, -0.1023408561356772, -0.10069659450427196, -0.09543669860811521, -0.10170838993712922, -0.10494641479408012, -0.09819949240506942]}}, "s4_arith": {"masked": 0.4517032421682614, "all_on_minus_masked": {"mean": -0.11382728179162305, "se": 0.0031806138203151002, "per_batch": [-0.1254835295663802, -0.11203949526361495, -0.10012180698725714, -0.11858833663449442, -0.10577231847918606, -0.1254753173948015, -0.1098109225759033, -0.11332652743134675]}}, "web": {"masked": 0.9522468807015565, "all_on_minus_masked": {"mean": -0.0005466468557674548, "se": 0.00016270460116765277, "per_batch": [-0.0010037767417607757, 0.00012606328308240045, -0.0005770017936447269, -0.00044887471875187934, -0.0008285264205316611, -0.0009810750318469674, -0.0008035029448310027, 0.00014351952214497476]}}}, "s3_rules": {"s1_perspective": {"masked": 0.04775003945880438, "all_on_minus_masked": {"mean": 0.00011281558041177728, "se": 3.900813527960418e-05, "per_batch": [0.0001760242431580808, -3.564340439131397e-05, 0.0001392844493675119, 0.00019313114246109764, 4.632782657616741e-05, 9.82343403896982e-05, -8.4459604947143e-06, 0.0002936120062276906]}}, "s2_concept": {"masked": 0.12176321295544607, "all_on_minus_masked": {"mean": -0.0005913099440278333, "se": 7.182195605793311e-05, "per_batch": [-0.0006947454159916822, -0.0006234908539560768, -0.00034145005545675733, -0.0009288138038159449, -0.0004376958464458458, -0.0007906257438476949, -0.00048234684217610124, -0.00043131099053256294]}}, "s3_rules": {"masked": 0.3180865401789643, "all_on_minus_masked": {"mean": -0.09584751141644804, "se": 0.0012728916345571442, "per_batch": [-0.09422374796724253, -0.0942421286130535, -0.09694262169473264, -0.10054316448187098, -0.09025894592557676, -0.10102720965272438, -0.09574411759653245, -0.09379815539985115]}}, "s4_arith": {"masked": 0.33816040087056304, "all_on_minus_masked": {"mean": -0.00028444049392464116, "se": 0.00015344607551703374, "per_batch": [-0.0007324741100246945, -8.960296109955523e-05, -0.00015190367637457625, -0.0004033423119360058, 0.0004619454002174894, -0.0005632861772383868, -0.0008385014142816005, 4.1641299340200444e-05]}}, "web": {"masked": 0.9513241507840148, "all_on_minus_masked": {"mean": 0.0003760830617742239, "se": 0.00021309106519225247, "per_batch": [0.000573562140627315, 0.00027766598981793233, 0.00010353355581937951, -8.616330808219885e-05, 0.001633835183196597, 0.0007242189427832946, -0.000236390153610877, 1.840214364234871e-05]}}}, "s4_arith": {"s1_perspective": {"masked": 0.04784311532737903, "all_on_minus_masked": {"mean": 1.9739711837117417e-05, "se": 7.317546250761771e-05, "per_batch": [7.612919529598888e-05, -4.426403476581048e-05, -0.00012592892163645203, 0.000166807047963409, -0.0003273743503751994, 0.000368612565377395, -3.937327813190172e-05, 8.330947096951008e-05]}}, "s2_concept": {"masked": 0.12121096725822438, "all_on_minus_masked": {"mean": -3.9064246806136405e-05, "se": 3.919354443192451e-05, "per_batch": [-1.7595974966408412e-05, 2.4948233290786326e-05, -0.0002383357073488246, 2.6506826064240374e-05, 8.614181025090595e-05, 4.7155493083442246e-05, -9.799786424476153e-05, -0.0001433367905784716]}}, "s3_rules": {"masked": 0.22219209899666212, "all_on_minus_masked": {"mean": 4.6929765854186734e-05, "se": 0.00013377785726005438, "per_batch": [0.00029260698261171503, -9.772914135278143e-05, 0.0004029123553088432, 0.0006557483499085304, -0.00016179267879917747, -0.0004922573424320786, -0.0002511591637537891, 2.7108765342231944e-05]}}, "s4_arith": {"masked": 0.5115781636507253, "all_on_minus_masked": {"mean": -0.17370220327408692, "se": 0.001970804996200162, "per_batch": [-0.17963785662668857, -0.1729364021509997, -0.1762955887854834, -0.16951123867207873, -0.16224763740421344, -0.17824097054071802, -0.1754108240381197, -0.17533710797439367]}}, "web": {"masked": 0.9517140999470148, "all_on_minus_masked": {"mean": -1.3866101225806782e-05, "se": 0.00013826493575802825, "per_batch": [0.00014635723588418692, 2.65713195582995e-05, -0.0005515483613169891, -5.1250829957050215e-05, 0.00016682317133698277, 0.00038248941551888294, -0.0006359918428906486, 0.00040562108205988157]}}}}, "member_alone": {"s1_perspective": {"s1_perspective": {"alone": 0.05217516482213097, "alone_minus_off": {"mean": -0.5016581364964282, "se": 0.004888586924071733, "per_batch": [-0.503264758784162, -0.5212149750156778, -0.5118989327924157, -0.509032675307594, -0.4859917833594548, -0.5041102201213458, -0.4784889810965285, -0.4992627654942472]}}, "s2_concept": {"alone": 0.24394271260453448, "alone_minus_off": {"mean": -0.47724122009222963, "se": 0.0030459897360259366, "per_batch": [-0.47044852344580024, -0.4716034084441754, -0.4687537634085354, -0.4938985943670891, -0.47575242540274654, -0.47682052365593, -0.47438017583151226, -0.4862723461820483]}}, "s3_rules": {"alone": 0.5034592174177095, "alone_minus_off": {"mean": -0.42615391876379405, "se": 0.003854800114533744, "per_batch": [-0.4117213926773634, -0.4236871716024687, -0.442290362963648, -0.410341145912864, -0.4260619510412451, -0.4339105083003576, -0.4286801719223179, -0.4325386456900876]}}, "s4_arith": {"alone": 0.7531657780810914, "alone_minus_off": {"mean": -0.480709089013143, "se": 0.003514095864453405, "per_batch": [-0.4824103037742731, -0.4703906082881222, -0.490546115055989, -0.47029558787352066, -0.4725516132918761, -0.49690809727670615, -0.47641963209676885, -0.48615075444788836]}}, "web": {"alone": 0.9519326898962615, "alone_minus_off": {"mean": 0.001335928985155277, "se": 0.0004171900864352844, "per_batch": [0.0008697162654132118, -0.0001570201602377752, 0.0007944738556607822, 0.00027680607656377365, 0.003323134771296754, 0.002440089850441529, 0.002121320007067329, 0.0010189112150366109]}}}, "s2_concept": {"s1_perspective": {"alone": 0.5237967463236832, "alone_minus_off": {"mean": -0.03003655499487594, "se": 0.0008929364404090443, "per_batch": [-0.02682404806374683, -0.03162864138957566, -0.02918588580804582, -0.032230279697956554, -0.03334249150108626, -0.031257674811664526, -0.026360984776298668, -0.02946243391063319]}}, "s2_concept": {"alone": 0.36033454325401637, "alone_minus_off": {"mean": -0.36084938944274775, "se": 0.0021975474547336524, "per_batch": [-0.36082329913716604, -0.3578469673813343, -0.3548674109507988, -0.37462086527661115, -0.36102178861409306, -0.36006399573360676, -0.3551193225386502, -0.36243146590972153]}}, "s3_rules": {"alone": 0.6157804256378956, "alone_minus_off": {"mean": -0.31383271054360795, "se": 0.0028684777591963, "per_batch": [-0.3051689393960323, -0.3130032220951231, -0.3208357849677054, -0.3010766122187518, -0.30994988510736754, -0.3192230176592393, -0.3165332949910704, -0.32487092791357386]}}, "s4_arith": {"alone": 0.8761517594682684, "alone_minus_off": {"mean": -0.35772310762596604, "se": 0.004197781721024297, "per_batch": [-0.3465006699646368, -0.3508452096994077, -0.37445335417467074, -0.3532194301945688, -0.34704009354906795, -0.37528652412677577, -0.36393756097898977, -0.35050201831961103]}}, "web": {"alone": 0.9514616616622953, "alone_minus_off": {"mean": 0.0008649007511890849, "se": 0.00018835522783446044, "per_batch": [5.219673453682461e-05, 0.0011502199689704673, 0.0014158471732280686, 0.00112304671013419, 0.0007925820465013445, 0.0004966858956918507, 0.00034379331907474775, 0.001544834161375186]}}}, "s3_rules": {"s1_perspective": {"alone": 0.5535631380718851, "alone_minus_off": {"mean": -0.00027016324667410163, "se": 0.00023913782920347164, "per_batch": [0.0002715606057124287, -0.0004794876306054574, -0.0007019041938338688, 0.0007255088126647546, -0.0014021745524844809, -0.0003248322318171182, 0.00029976576045387393, -0.000549742543482945]}}, "s2_concept": {"alone": 0.7227973879358336, "alone_minus_off": {"mean": 0.0016134552390694584, "se": 0.0002752200095664509, "per_batch": [0.000995564570182128, 0.0010287142261358184, 0.0016847850435147693, 0.0009734218038836007, 0.0022790280979085864, 0.0019874315133706633, 0.0009180433903056562, 0.003040653267254445]}}, "s3_rules": {"alone": 0.8701872076558641, "alone_minus_off": {"mean": -0.05942592852563948, "se": 0.010330913371419828, "per_batch": [-0.06883390622470331, -0.052931358430022946, -0.047376318807154205, -0.08934343927937138, -0.029561753891506393, -0.06865332444129746, -0.10310746880189825, -0.015599858329161886]}}, "s4_arith": {"alone": 1.2291967885059467, "alone_minus_off": {"mean": -0.0046780785882879206, "se": 0.0004241556239844928, "per_batch": [-0.0034991590144548557, -0.00447249482701273, -0.00568497251559541, -0.003813715282882768, -0.004839591795279485, -0.0037889497811585127, -0.00709909986231505, -0.004226645627604553]}}, "web": {"alone": 0.9509403070051203, "alone_minus_off": {"mean": 0.0003435460940141577, "se": 0.0001181112928609609, "per_batch": [0.00013586629418149698, 0.00014833503636912937, 0.001022264876728518, 0.0005037371843772398, 0.0003148142424045375, 0.0005318563477932914, 4.4629497898962533e-05, 4.6865272360085974e-05]}}}, "s4_arith": {"s1_perspective": {"alone": 0.5525393683469615, "alone_minus_off": {"mean": -0.0012939329715976888, "se": 0.00033731606119881077, "per_batch": [-0.0005440671160045207, -0.002287799213102515, 5.077787766727404e-05, -0.002374908425764488, -0.002254219600521523, -0.0003551871696944042, -0.00143721601759772, -0.0011488441077636136]}}, "s2_concept": {"alone": 0.7197303675784188, "alone_minus_off": {"mean": -0.001453565118345393, "se": 0.00024222923191314095, "per_batch": [-0.0015742001990100007, -0.001012590852617512, -0.0006298004674595292, -0.0029371197114349545, -0.0012482930756249067, -0.0015436732784818163, -0.001588259780718082, -0.0010945835814163418]}}, "s3_rules": {"alone": 0.9300652463238748, "alone_minus_off": {"mean": 0.000452110142371287, "se": 0.0003056447991096855, "per_batch": [0.002266817329697246, -0.0006020252693452743, 0.00024163562446222198, 0.0009758295609955114, 0.00042324930377346526, 0.00028738300959174623, 0.00022779102106773585, -0.0002037994412723565]}}, "s4_arith": {"alone": 1.1550061816504822, "alone_minus_off": {"mean": -0.07886868544375247, "se": 0.023194415891436188, "per_batch": [-0.07496078816169094, 0.045132117196045485, -0.145121196572499, -0.12688088860464064, -0.013288755466867475, -0.09220961614732559, -0.08542928612838474, -0.13819106966465688]}}, "web": {"alone": 0.9506567613584259, "alone_minus_off": {"mean": 6.000044731975518e-05, "se": 9.631581610844272e-05, "per_batch": [-0.0005266108769419464, -4.480148054974986e-05, 0.00030853687564791343, 0.00030509722263061256, -7.223271336176396e-06, 0.0002469670866389606, 9.467644929994634e-05, 0.00010336157316848116]}}}}, "battery": {"s1_perspective": {"acc_off": {"in": 0.55, "transfer": 0.4625}, "acc_all_on": {"in": 1.0, "transfer": 0.4625}, "acc_member_masked": {"in": 0.5625, "transfer": 0.4875}, "acc_member_alone": {"in": 0.925, "transfer": 0.4625}, "margin_all_on_minus_off": {"in": {"mean": 3.0999846355989575, "se": 0.1651246719249035, "per_batch": null}, "transfer": {"mean": -0.17582505494356154, "se": 0.11536113102244777, "per_batch": null}}}, "s2_concept": {"acc_off": {"in": 0.2976190476190476, "transfer": 0.6779661016949152}, "acc_all_on": {"in": 0.9523809523809523, "transfer": 0.7542372881355932}, "acc_member_masked": {"in": 0.8928571428571429, "transfer": 0.7288135593220338}, "acc_member_alone": {"in": 0.9047619047619048, "transfer": 0.7033898305084746}, "margin_all_on_minus_off": {"in": {"mean": 4.35635296530312, "se": 0.3367575048310155, "per_batch": null}, "transfer": {"mean": 0.12476909312151246, "se": 0.1689522993127803, "per_batch": null}}}, "s3_rules": {"acc_off": {"in": 0.155, "transfer": 0.07, "held-out": 0.05}, "acc_all_on": {"in": 0.885, "transfer": 0.33, "held-out": 0.48333333333333334}, "acc_member_masked": {"in": 0.8, "transfer": 0.21, "held-out": 0.4166666666666667}, "acc_member_alone": {"in": 0.335, "transfer": 0.1, "held-out": 0.14166666666666666}, "margin_all_on_minus_off": {"in": {"mean": 3.013909998992458, "se": 0.13655973860275505, "per_batch": null}, "transfer": {"mean": 1.0933739161491394, "se": 0.14863747380240996, "per_batch": null}, "held-out": {"mean": 1.6468284621213873, "se": 0.20772868702034944, "per_batch": null}}}, "s4_arith": {"acc_off": {"in": 0.34375, "transfer": 0.24166666666666667, "held-out": 0.35}, "acc_all_on": {"in": 0.59375, "transfer": 0.2708333333333333, "held-out": 0.2875}, "acc_member_masked": {"in": 0.48125, "transfer": 0.25833333333333336, "held-out": 0.2875}, "acc_member_alone": {"in": 0.4125, "transfer": 0.2375, "held-out": 0.4}, "margin_all_on_minus_off": {"in": {"mean": 1.5344162943940318, "se": 0.22434896998022902, "per_batch": null}, "transfer": {"mean": -0.15970706542332966, "se": 0.057600734283105026, "per_batch": null}, "held-out": {"mean": -0.12023597359657287, "se": 0.09775862709405657, "per_batch": null}}}}, "probes_all_on": {"p0_grounding": {"acc": 0.3333333333333333, "n": 30, "families": {"count": 0.125, "permanence": 0.5714285714285714, "sequence": 0.25, "attribute": 0.42857142857142855}}, "p1_self_other": {"acc": 0.5, "n": 30, "families": {"belief": 0.125, "reality": 0.6666666666666666, "perspective": 0.8, "reference": 0.8571428571428571, "second_order": 0.0}}, "p2_difference": {"acc": 0.5333333333333333, "n": 30, "families": {"category": 1.0, "antonym": 0.5, "odd_one_out": 0.5, "property": 0.5, "analogy": 0.16666666666666666}}, "p3_logic": {"acc": 0.26666666666666666, "n": 30, "families": {"d1": 0.5, "d2": 0.25, "d3": 0.5, "d4": 0.5, "d5": 0.0, "d6": 0.0, "negation": 0.0, "contrapositive": 0.25}}, "p4_arithmetic": {"acc": 0.3, "n": 30, "families": {"add2": 0.4, "add3": 0.0, "sub2": 0.25, "sub3": 0.4, "mul1": 0.25, "mul2": 0.3333333333333333, "word": 0.4}}, "p5_causal": {"acc": 0.5666666666666667, "n": 30, "families": {"effect": 0.7142857142857143, "cause": 0.42857142857142855, "counterfactual": 0.5, "prevention": 0.4, "chain": 0.8}}, "p6_learning": {"acc": 0.36666666666666664, "n": 30, "families": {"error_detect": 0.25, "correction": 0.25, "strategy": 0.7142857142857143, "check": 0.2857142857142857}}, "p7_composite": {"acc": 0.36666666666666664, "n": 30, "families": {"logic_arith": 0.25, "causal_category": 0.5, "belief_causal": 0.2857142857142857, "three_step": 0.42857142857142855}}, "p8_articulation": {"acc": 0.5, "n": 30, "families": {"definition": 0.375, "register": 0.5714285714285714, "grammar": 0.875, "paraphrase": 0.14285714285714285}}}, "probes_mean_off": 0.41111111111111115, "probes_mean_all_on": 0.4148148148148148, "gates": {"s1_perspective": {"mean": 0.21163099743248942, "max": 0.554569513244894}, "s2_concept": {"mean": 0.27282737975382454, "max": 0.48724048599867004}, "s3_rules": {"mean": 0.12374002504886239, "max": 0.17456351278495152}, "s4_arith": {"mean": 0.10761241921992425, "max": 0.16724209369972712}}, "chain_proof": "compiled chain: 8 cache entries (limit 256), 0 limit warnings", "resumed_read": true}, {"close": 5, "attached": ["s1_perspective", "s2_concept", "s3_rules", "s4_arith", "s5_causal"], "rows": {"s1_perspective": {"off": 0.5538333013185592, "all_on": 0.047946781229222746, "all_on_minus_off": {"mean": -0.5058865200893364, "se": 0.004828306773977591, "per_batch": [-0.5076058481169999, -0.525296768255593, -0.5168950234255225, -0.5116196437163303, -0.4927038953850393, -0.5082228466258859, -0.48171291949948214, -0.5030352156898386]}}, "s2_concept": {"off": 0.7211839326967642, "all_on": 0.12118336001191751, "all_on_minus_off": {"mean": -0.6000005726848465, "se": 0.0037192467114687544, "per_batch": [-0.5989980307247744, -0.5978078462873098, -0.5874249956807457, -0.6222191604910179, -0.5988008203688132, -0.5980649388525183, -0.5910525934886479, -0.6056361955849459]}}, "s3_rules": {"off": 0.9296131361815035, "all_on": 0.22262603003539144, "all_on_minus_off": {"mean": -0.706987106146112, "se": 0.005431843747466948, "per_batch": [-0.6875137605012058, -0.7054556355741548, -0.7217293219293972, -0.6910979649365122, -0.6927777624831519, -0.729362470918304, -0.713451302493249, -0.7145086303329221]}}, "s4_arith": {"off": 1.2338748670942346, "all_on": 0.3379316531959998, "all_on_minus_off": {"mean": -0.8959432138982347, "se": 0.005804015701286444, "per_batch": [-0.9122737808280211, -0.8874999379452098, -0.9112647156197462, -0.871532789167299, -0.8762336554568262, -0.9138974905326482, -0.8985752551976416, -0.8962680864384858]}}, "s5_causal": {"off": 0.6604376826115352, "all_on": 0.04257019342493747, "all_on_minus_off": {"mean": -0.6178674891865977, "se": 0.006917613128820046, "per_batch": [-0.5836138762609995, -0.6274796314541635, -0.6286477859889442, -0.6293462102213271, -0.6435690330245486, -0.6159303864477611, -0.5954890391397025, -0.6188639509553354]}}, "s6_tryfail": {"off": 0.6728212611401647, "all_on": 0.487750171603081, "all_on_minus_off": {"mean": -0.1850710895370837, "se": 0.0036888785570146697, "per_batch": [-0.18505793286429273, -0.20526090684379772, -0.17884385572331768, -0.1948563004318995, -0.18653053431230548, -0.17803476334233354, -0.17469056069630634, -0.17729386208241665]}}, "s7_mixed": {"off": 0.7829759286650222, "all_on": 0.776759180418501, "all_on_minus_off": {"mean": -0.006216748246521228, "se": 0.008146963933420282, "per_batch": [0.0010176213451551508, 0.001067324331254671, 0.0012357813377746885, 0.0023836795409585276, 7.558637505422627e-05, -0.061341954992155, 0.013796663230528083, -0.007968687140740172]}}, "s8_register": {"off": 0.4906504977761471, "all_on": 0.3793570928114932, "all_on_minus_off": {"mean": -0.11129340496465395, "se": 0.002203283938365805, "per_batch": [-0.09948833039063532, -0.11439069158139176, -0.10743829292026269, -0.11210564409070306, -0.11017144120777445, -0.12104246456881262, -0.11461499995377955, -0.11109537500387218]}}, "web": {"off": 0.9505967609111062, "all_on": 0.9519801786057309, "all_on_minus_off": {"mean": 0.0013834176946247723, "se": 0.0004164309813143485, "per_batch": [0.0007523381061993462, 0.00013294258911689116, 9.51064059271367e-05, 0.0008782294066309593, 0.003214269753300636, 0.002877269748934763, 0.001866269735837811, 0.0012509158110506347]}}}, "member_in_group": {"s1_perspective": {"s1_perspective": {"masked": 0.5187216834070835, "all_on_minus_masked": {"mean": -0.4707749021778608, "se": 0.0043812322752412244, "per_batch": [-0.4755617396386025, -0.4881699705049215, -0.48139690137852176, -0.47476883512463086, -0.45404666702000235, -0.4712784794721171, -0.4524591435400084, -0.46851748074408134]}}, "s2_concept": {"masked": 0.3581652992063946, "all_on_minus_masked": {"mean": -0.2369819391944771, "se": 0.0015361789103154927, "per_batch": [-0.23506341494868072, -0.23735757751986514, -0.23039857902450922, -0.2445294367924108, -0.23612817028908803, -0.23626802443098655, -0.23455644234068418, -0.24155386820959224]}}, "s3_rules": {"masked": 0.4193903248065653, "all_on_minus_masked": {"mean": -0.1967642947711739, "se": 0.003064286459530783, "per_batch": [-0.195639466428266, -0.20031722254908352, -0.19751884984123202, -0.18679715041680553, -0.1992022590235399, -0.21453636116145608, -0.1898094050488363, -0.19029364370017185]}}, "s4_arith": {"masked": 0.4976382442464864, "all_on_minus_masked": {"mean": -0.1597065910504865, "se": 0.0053152756830371455, "per_batch": [-0.16896381789438242, -0.18001215536846005, -0.13032709846046087, -0.16069921310500151, -0.17217305715514514, -0.156229856961367, -0.15525540326157827, -0.15399212619749691]}}, "s5_causal": {"masked": 0.08532192595858355, "all_on_minus_masked": {"mean": -0.04275173253364608, "se": 0.002220324010700502, "per_batch": [-0.04416140949337083, -0.05257142561405624, -0.0389490506768041, -0.040628732672600316, -0.04004718140039448, -0.047993658594113, -0.04565602203346516, -0.032006379784364485]}}, "web": {"masked": 0.9517370703798207, "all_on_minus_masked": {"mean": 0.00024310822591026848, "se": 0.0005153176377177189, "per_batch": [0.000933521828883288, -0.001280668809649943, -0.0015991806790478513, -0.0005263529029656544, 0.002310758896992837, 0.0008143378518353783, 0.0019031600144479022, -0.0006107103932138092]}}}, "s2_concept": {"s1_perspective": {"masked": 0.052145148475097575, "all_on_minus_masked": {"mean": -0.004198367245874822, "se": 0.0003508577771089566, "per_batch": [-0.00439394175123118, -0.004156213357618226, -0.004185122566336699, -0.0028337796360978565, -0.006183587841651725, -0.004036056603701356, -0.003249187605883455, -0.004549048604478079]}}, "s2_concept": {"masked": 0.24209673682149851, "all_on_minus_masked": {"mean": -0.12091337680958102, "se": 0.001456951686715796, "per_batch": [-0.12658080037671626, -0.12409019062258382, -0.11644169218208754, -0.12530796752659634, -0.1214665630369246, -0.11970065592678097, -0.11569594316119937, -0.11802320164375922]}}, "s3_rules": {"masked": 0.32337599290600927, "all_on_minus_masked": {"mean": -0.10074996287061785, "se": 0.0009653436987328162, "per_batch": [-0.10130289784205745, -0.09914219380794254, -0.10282406289110763, -0.10126602906127877, -0.0961135148327544, -0.10208976146541754, -0.10478371920636392, -0.09847752385802053]}}, "s4_arith": {"masked": 0.45148088472406933, "all_on_minus_masked": {"mean": -0.11354923152806952, "se": 0.003144956349553675, "per_batch": [-0.12426397909128067, -0.11218082200696144, -0.0997514208507927, -0.1191835040956365, -0.10503868348626805, -0.12487892455727201, -0.1097762680717545, -0.11332025006459029]}}, "s5_causal": {"masked": 0.04338749495436772, "all_on_minus_masked": {"mean": -0.0008173015294302482, "se": 0.0001609407107024218, "per_batch": [-0.0004936547014703363, -0.00051597482487762, -0.001448134228698647, -0.0009546837566262109, -0.0008960994789471433, -0.000910838929571875, -0.0012813594274822745, -3.766688776787874e-05]}}, "web": {"masked": 0.9523372253378378, "all_on_minus_masked": {"mean": -0.0003570467321068471, "se": 0.00012591817450675132, "per_batch": [-0.0004935042166507309, -3.5944374030316695e-05, -0.0006715922516192796, 0.00017894794822270743, -0.0009584593132583308, -0.00023415437914975357, -0.0003370000043656507, -0.0003046672660034222]}}}, "s3_rules": {"s1_perspective": {"masked": 0.047842492226173275, "all_on_minus_masked": {"mean": 0.00010428900304947, "se": 4.7836885101168255e-05, "per_batch": [0.0001665544484449541, -2.65713195582995e-05, 8.693185555330718e-05, 0.00015189292745890204, -9.672949219464988e-05, 0.00019386206872725603, 2.930423136966731e-05, 0.0003290673045946227]}}, "s2_concept": {"masked": 0.12177191554630262, "all_on_minus_masked": {"mean": -0.0005885555343851084, "se": 0.00011600491600937184, "per_batch": [-0.0010567796449735684, -0.0007444484020910358, -0.0001552035934880014, -0.0010234472574532222, -0.0005856761685976153, -0.00043114975679738043, -0.0003500169412528481, -0.0003617225104271954]}}, "s3_rules": {"masked": 0.317781934719226, "all_on_minus_masked": {"mean": -0.09515590468383461, "se": 0.0012477811649909657, "per_batch": [-0.09345732878150378, -0.09349338064469087, -0.0961676033764508, -0.1000193913186683, -0.08970735456859699, -0.09992755258310759, -0.09540361344565546, -0.09307101275200316]}}, "s4_arith": {"masked": 0.33813486682136773, "all_on_minus_masked": {"mean": -0.00020321362536793175, "se": 8.817625610815621e-05, "per_batch": [-0.0001000294093080889, -0.00035834735090400915, -1.9498533041562016e-05, -0.00018870796365921816, 1.2189270379880934e-05, -0.0003888527736007896, -0.0006589085311181631, 7.644628830849598e-05]}}, "s5_causal": {"masked": 0.04267627010702579, "all_on_minus_masked": {"mean": -0.00010607668208832254, "se": 0.00010445587509891845, "per_batch": [-0.000127412272000145, 0.00032031231277414307, 4.847223525411559e-05, -0.00039090312926656767, 0.00011768987776855644, -0.00016450678000811542, -0.0006205886467227709, -3.16770545057965e-05]}}, "web": {"masked": 0.9515811036133195, "all_on_minus_masked": {"mean": 0.0003990749924114717, "se": 0.00014871141756800163, "per_batch": [0.0003277129412191382, 0.00015702016023766419, -0.00014850701902002772, 0.0003593577489777733, 0.0011297540335177825, 0.0008747037622882647, 0.00042866675727559755, 6.389155479558095e-05]}}}, "s4_arith": {"s1_perspective": {"masked": 0.047889463308172106, "all_on_minus_masked": {"mean": 5.7317921050640576e-05, "se": 6.268801864596545e-05, "per_batch": [4.287205018539686e-05, -0.00016948890242529657, -4.815514224161543e-06, 0.0003429817759409981, -0.000140972029129105, 0.00026385363317057986, 8.365343627123739e-05, 4.0458918615475514e-05]}}, "s2_concept": {"masked": 0.12131966969887088, "all_on_minus_masked": {"mean": -0.00013630968695337552, "se": 3.304084187280858e-05, "per_batch": [-0.00010403875485633096, -7.640329264581303e-05, -0.0002383034606017742, -0.00019822075403506945, -0.00017852874051127454, 5.299215429710458e-05, -0.0002040896619957544, -0.00014388498527809213]}}, "s3_rules": {"masked": 0.222453899586938, "all_on_minus_masked": {"mean": 0.00017213044845339198, "se": 8.221039243400085e-05, "per_batch": [0.00039990265891873267, -2.0250957139089643e-05, 0.0002892963165825879, 0.00016781207157939848, 7.442549216088445e-05, 0.00011411048884746311, -0.00018668716751157888, 0.0005384346841887377]}}, "s4_arith": {"masked": 0.510733454737639, "all_on_minus_masked": {"mean": -0.17280180154163927, "se": 0.0019236685506066754, "per_batch": [-0.17891546499957067, -0.17148719034350424, -0.1751953082787573, -0.16906793189164843, -0.16168985467180264, -0.17725950854790679, -0.17439715829393432, -0.17440199530598988]}}, "s5_causal": {"masked": 0.042695655776452734, "all_on_minus_masked": {"mean": -0.0001254623515152646, "se": 6.817579859730018e-05, "per_batch": [-7.998536879580409e-05, 2.960251377975842e-05, -0.00024182373048663164, -0.00019772630391383383, -6.638530323303804e-05, -0.00031367217011348675, -0.0003580813152409684, 0.00022437286588188748]}}, "web": {"masked": 0.9517662966815517, "all_on_minus_masked": {"mean": 0.00021388192417927387, "se": 0.00014278809535310343, "per_batch": [3.1214851131666776e-05, 0.0002983898992469092, -0.0005634151642265328, 0.0005819032991942752, -7.515641842692489e-05, 0.0006877586208002828, 0.00034800689402092466, 0.00040235341169359007]}}}, "s5_causal": {"s1_perspective": {"masked": 0.047862855039216154, "all_on_minus_masked": {"mean": 8.392619000659098e-05, "se": 3.957921422000305e-05, "per_batch": [8.615793362443114e-05, 5.503444827609921e-05, 0.00011079444836052665, 0.00013873625466789136, 0.00018429015931518272, -0.0001554723163799815, 4.7687564409551464e-05, 0.0002041810277790268]}}, "s2_concept": {"masked": 0.12117190301141822, "all_on_minus_masked": {"mean": 1.1457000499275408e-05, "se": 0.00010020219538663027, "per_batch": [-0.00028100890259415146, 0.0001868376523310733, -4.220024295545355e-05, 9.68154835200713e-05, -3.303141788132957e-05, 0.0005447120509452369, -5.696925309828238e-06, -0.0003747716940614154]}}, "s3_rules": {"masked": 0.22223902876251628, "all_on_minus_masked": {"mean": 0.0003870012728751128, "se": 9.477647241836384e-05, "per_batch": [0.0004943641299051393, 0.0006252644250430983, 0.00040192345506637195, -2.23577446121781e-05, 0.0007973975607254158, 0.0004525723457454722, 0.00013485589610778836, 0.00021199011501979448]}}, "s4_arith": {"masked": 0.33787596037663836, "all_on_minus_masked": {"mean": 5.569281936142895e-05, "se": 8.246765875195317e-05, "per_batch": [0.0002031760041630304, 8.250867675146933e-05, 0.00017148820074158744, -2.938753546616324e-05, -0.00014850701902002772, 0.0004988786744902773, -0.00014334753949418744, -0.00018926690727455453]}}, "s5_causal": {"masked": 0.48561756523455085, "all_on_minus_masked": {"mean": -0.4430473718096134, "se": 0.005744793601154896, "per_batch": [-0.41907721426729605, -0.4463688724098232, -0.44252854818617726, -0.45557958332111853, -0.46641357398041383, -0.45410116133526557, -0.4229220772205875, -0.4373879437562252]}}, "web": {"masked": 0.951700233845789, "all_on_minus_masked": {"mean": 0.0002799447599419608, "se": 0.0001265995870350301, "per_batch": [0.00048189538771770124, -6.148379768344814e-05, 5.520643092693511e-05, 0.0005217093713922871, 0.0006058948789897656, 1.3930594719879785e-05, 0.0007867346363719108, -0.00016432942289934527]}}}}, "member_alone": {"s1_perspective": {"s1_perspective": {"alone": 0.05217516482213097, "alone_minus_off": {"mean": -0.5016581364964282, "se": 0.004888586924071733, "per_batch": [-0.503264758784162, -0.5212149750156778, -0.5118989327924157, -0.509032675307594, -0.4859917833594548, -0.5041102201213458, -0.4784889810965285, -0.4992627654942472]}}, "s2_concept": {"alone": 0.24394271260453448, "alone_minus_off": {"mean": -0.47724122009222963, "se": 0.0030459897360259366, "per_batch": [-0.47044852344580024, -0.4716034084441754, -0.4687537634085354, -0.4938985943670891, -0.47575242540274654, -0.47682052365593, -0.47438017583151226, -0.4862723461820483]}}, "s3_rules": {"alone": 0.5034592174177095, "alone_minus_off": {"mean": -0.42615391876379405, "se": 0.003854800114533744, "per_batch": [-0.4117213926773634, -0.4236871716024687, -0.442290362963648, -0.410341145912864, -0.4260619510412451, -0.4339105083003576, -0.4286801719223179, -0.4325386456900876]}}, "s4_arith": {"alone": 0.7531657780810914, "alone_minus_off": {"mean": -0.480709089013143, "se": 0.003514095864453405, "per_batch": [-0.4824103037742731, -0.4703906082881222, -0.490546115055989, -0.47029558787352066, -0.4725516132918761, -0.49690809727670615, -0.47641963209676885, -0.48615075444788836]}}, "s5_causal": {"alone": 0.468494725289589, "alone_minus_off": {"mean": -0.19194295732194622, "se": 0.002210700040436461, "per_batch": [-0.1875164678540393, -0.1992422879855283, -0.19983549914401677, -0.1947394812163009, -0.19148918109764468, -0.18138197568475095, -0.1878665385398705, -0.19347222705341832]}}, "web": {"alone": 0.9519326898962615, "alone_minus_off": {"mean": 0.001335928985155277, "se": 0.0004171900864352844, "per_batch": [0.0008697162654132118, -0.0001570201602377752, 0.0007944738556607822, 0.00027680607656377365, 0.003323134771296754, 0.002440089850441529, 0.002121320007067329, 0.0010189112150366109]}}}, "s2_concept": {"s1_perspective": {"alone": 0.5237967463236832, "alone_minus_off": {"mean": -0.03003655499487594, "se": 0.0008929364404090443, "per_batch": [-0.02682404806374683, -0.03162864138957566, -0.02918588580804582, -0.032230279697956554, -0.03334249150108626, -0.031257674811664526, -0.026360984776298668, -0.02946243391063319]}}, "s2_concept": {"alone": 0.36033454325401637, "alone_minus_off": {"mean": -0.36084938944274775, "se": 0.0021975474547336524, "per_batch": [-0.36082329913716604, -0.3578469673813343, -0.3548674109507988, -0.37462086527661115, -0.36102178861409306, -0.36006399573360676, -0.3551193225386502, -0.36243146590972153]}}, "s3_rules": {"alone": 0.6157804256378956, "alone_minus_off": {"mean": -0.31383271054360795, "se": 0.0028684777591963, "per_batch": [-0.3051689393960323, -0.3130032220951231, -0.3208357849677054, -0.3010766122187518, -0.30994988510736754, -0.3192230176592393, -0.3165332949910704, -0.32487092791357386]}}, "s4_arith": {"alone": 0.8761517594682684, "alone_minus_off": {"mean": -0.35772310762596604, "se": 0.004197781721024297, "per_batch": [-0.3465006699646368, -0.3508452096994077, -0.37445335417467074, -0.3532194301945688, -0.34704009354906795, -0.37528652412677577, -0.36393756097898977, -0.35050201831961103]}}, "s5_causal": {"alone": 0.6477866226906935, "alone_minus_off": {"mean": -0.012651059920841742, "se": 0.0009352908521067744, "per_batch": [-0.007499647456174596, -0.011507402164893321, -0.015175620125146061, -0.01262455846923527, -0.013140119460859356, -0.011161158093043633, -0.015682152027599816, -0.014417821569781886]}}, "web": {"alone": 0.9514616616622953, "alone_minus_off": {"mean": 0.0008649007511890849, "se": 0.00018835522783446044, "per_batch": [5.219673453682461e-05, 0.0011502199689704673, 0.0014158471732280686, 0.00112304671013419, 0.0007925820465013445, 0.0004966858956918507, 0.00034379331907474775, 0.001544834161375186]}}}, "s3_rules": {"s1_perspective": {"alone": 0.5535631380718851, "alone_minus_off": {"mean": -0.00027016324667410163, "se": 0.00023913782920347164, "per_batch": [0.0002715606057124287, -0.0004794876306054574, -0.0007019041938338688, 0.0007255088126647546, -0.0014021745524844809, -0.0003248322318171182, 0.00029976576045387393, -0.000549742543482945]}}, "s2_concept": {"alone": 0.7227973879358336, "alone_minus_off": {"mean": 0.0016134552390694584, "se": 0.0002752200095664509, "per_batch": [0.000995564570182128, 0.0010287142261358184, 0.0016847850435147693, 0.0009734218038836007, 0.0022790280979085864, 0.0019874315133706633, 0.0009180433903056562, 0.003040653267254445]}}, "s3_rules": {"alone": 0.8701872076558641, "alone_minus_off": {"mean": -0.05942592852563948, "se": 0.010330913371419828, "per_batch": [-0.06883390622470331, -0.052931358430022946, -0.047376318807154205, -0.08934343927937138, -0.029561753891506393, -0.06865332444129746, -0.10310746880189825, -0.015599858329161886]}}, "s4_arith": {"alone": 1.2291967885059467, "alone_minus_off": {"mean": -0.0046780785882879206, "se": 0.0004241556239844928, "per_batch": [-0.0034991590144548557, -0.00447249482701273, -0.00568497251559541, -0.003813715282882768, -0.004839591795279485, -0.0037889497811585127, -0.00709909986231505, -0.004226645627604553]}}, "s5_causal": {"alone": 0.666408248442262, "alone_minus_off": {"mean": 0.005970565830726776, "se": 0.0006340753987558168, "per_batch": [0.006426389723465098, 0.007101077662799882, 0.001980165246371679, 0.007604900838502515, 0.006885497409943375, 0.006719362169209986, 0.004959076741966384, 0.00608805685355529]}}, "web": {"alone": 0.9509403070051203, "alone_minus_off": {"mean": 0.0003435460940141577, "se": 0.0001181112928609609, "per_batch": [0.00013586629418149698, 0.00014833503636912937, 0.001022264876728518, 0.0005037371843772398, 0.0003148142424045375, 0.0005318563477932914, 4.4629497898962533e-05, 4.6865272360085974e-05]}}}, "s4_arith": {"s1_perspective": {"alone": 0.5525393683469615, "alone_minus_off": {"mean": -0.0012939329715976888, "se": 0.00033731606119881077, "per_batch": [-0.0005440671160045207, -0.002287799213102515, 5.077787766727404e-05, -0.002374908425764488, -0.002254219600521523, -0.0003551871696944042, -0.00143721601759772, -0.0011488441077636136]}}, "s2_concept": {"alone": 0.7197303675784188, "alone_minus_off": {"mean": -0.001453565118345393, "se": 0.00024222923191314095, "per_batch": [-0.0015742001990100007, -0.001012590852617512, -0.0006298004674595292, -0.0029371197114349545, -0.0012482930756249067, -0.0015436732784818163, -0.001588259780718082, -0.0010945835814163418]}}, "s3_rules": {"alone": 0.9300652463238748, "alone_minus_off": {"mean": 0.000452110142371287, "se": 0.0003056447991096855, "per_batch": [0.002266817329697246, -0.0006020252693452743, 0.00024163562446222198, 0.0009758295609955114, 0.00042324930377346526, 0.00028738300959174623, 0.00022779102106773585, -0.0002037994412723565]}}, "s4_arith": {"alone": 1.1550061816504822, "alone_minus_off": {"mean": -0.07886868544375247, "se": 0.023194415891436188, "per_batch": [-0.07496078816169094, 0.045132117196045485, -0.145121196572499, -0.12688088860464064, -0.013288755466867475, -0.09220961614732559, -0.08542928612838474, -0.13819106966465688]}}, "s5_causal": {"alone": 0.6593072245240404, "alone_minus_off": {"mean": -0.0011304580874947762, "se": 0.000321404023591384, "per_batch": [-0.0015519714413860797, -0.0018272296740919902, -0.0005069188634181376, -0.0016303525345167103, -0.00010731717413836606, -9.652526279668727e-05, -0.0007122661485483572, -0.002611083601061881]}}, "web": {"alone": 0.9506567613584259, "alone_minus_off": {"mean": 6.000044731975518e-05, "se": 9.631581610844272e-05, "per_batch": [-0.0005266108769419464, -4.480148054974986e-05, 0.00030853687564791343, 0.00030509722263061256, -7.223271336176396e-06, 0.0002469670866389606, 9.467644929994634e-05, 0.00010336157316848116]}}}, "s5_causal": {"s1_perspective": {"alone": 0.5516690125210654, "alone_minus_off": {"mean": -0.0021642887974937935, "se": 0.0003518279348647084, "per_batch": [-0.0022480712207532116, -0.002950921319166744, -0.0034509608765503907, -0.002130435087563054, -0.0009878683465559535, -0.0030761676846575714, -0.000640592378801208, -0.0018292934659022153]}}, "s2_concept": {"alone": 0.7201691168186011, "alone_minus_off": {"mean": -0.0010148158781630306, "se": 0.0002551944996123987, "per_batch": [-0.0015721364071996646, -0.0003884228169737103, -0.0008650297381771477, -0.0023907308296438057, -0.0003562620612622691, -0.00037191248249079933, -0.0007946458383116806, -0.001379386851245168]}}, "s3_rules": {"alone": 0.9277824453565631, "alone_minus_off": {"mean": -0.0018306908249405285, "se": 0.00022613039944255578, "per_batch": [-0.0009550196602411409, -0.00278887066639133, -0.002538291944084281, -0.002192434833199086, -0.0011808328808240454, -0.0015964289566340328, -0.0015663319927330388, -0.0018273156654172729]}}, "s4_arith": {"alone": 1.2316764666128335, "alone_minus_off": {"mean": -0.002198400481400936, "se": 0.00021403025089921716, "per_batch": [-0.003272657863268469, -0.002493060506907341, -0.0025671850294290977, -0.0021541686933821413, -0.001368895909542589, -0.0017796764711284219, -0.001641144445858389, -0.0023104149316910405]}}, "s5_causal": {"alone": 0.1417724646021276, "alone_minus_off": {"mean": -0.5186652180094076, "se": 0.007470902026256088, "per_batch": [-0.49118512734418557, -0.5035505832009803, -0.5335864502652528, -0.5333256923197978, -0.5513792916223135, -0.5048105495990326, -0.5011616904359175, -0.5303223592877806]}}, "web": {"alone": 0.9506011034730404, "alone_minus_off": {"mean": 4.342561934281286e-06, "se": 0.00010631213097956605, "per_batch": [-0.00025565221050771747, -0.00029426231562623695, 0.0002472250606152526, 0.0005403694890109279, 9.476244062556205e-05, -0.000294778263578821, -0.00012821306621824125, 0.0001252893611535244]}}}}, "battery": {"s1_perspective": {"acc_off": {"in": 0.55, "transfer": 0.4625}, "acc_all_on": {"in": 1.0, "transfer": 0.4625}, "acc_member_masked": {"in": 0.55, "transfer": 0.45}, "acc_member_alone": {"in": 0.925, "transfer": 0.4625}, "margin_all_on_minus_off": {"in": {"mean": 3.24611360373965, "se": 0.17265411528343902, "per_batch": null}, "transfer": {"mean": -0.18288378417491913, "se": 0.13469547115003136, "per_batch": null}}}, "s2_concept": {"acc_off": {"in": 0.2976190476190476, "transfer": 0.6779661016949152}, "acc_all_on": {"in": 0.9523809523809523, "transfer": 0.7796610169491526}, "acc_member_masked": {"in": 0.9166666666666666, "transfer": 0.7372881355932204}, "acc_member_alone": {"in": 0.9047619047619048, "transfer": 0.7033898305084746}, "margin_all_on_minus_off": {"in": {"mean": 4.397694960946128, "se": 0.3377279155440735, "per_batch": null}, "transfer": {"mean": 0.12031523201425197, "se": 0.17704083117187203, "per_batch": null}}}, "s3_rules": {"acc_off": {"in": 0.155, "transfer": 0.07, "held-out": 0.05}, "acc_all_on": {"in": 0.88, "transfer": 0.33, "held-out": 0.475}, "acc_member_masked": {"in": 0.82, "transfer": 0.19, "held-out": 0.4166666666666667}, "acc_member_alone": {"in": 0.335, "transfer": 0.1, "held-out": 0.14166666666666666}, "margin_all_on_minus_off": {"in": {"mean": 3.0421862694621087, "se": 0.13812059963185946, "per_batch": null}, "transfer": {"mean": 1.0796322298049927, "se": 0.1499289340066482, "per_batch": null}, "held-out": {"mean": 1.6696617346877853, "se": 0.20846516395396075, "per_batch": null}}}, "s4_arith": {"acc_off": {"in": 0.34375, "transfer": 0.24166666666666667, "held-out": 0.35}, "acc_all_on": {"in": 0.59375, "transfer": 0.275, "held-out": 0.2875}, "acc_member_masked": {"in": 0.46875, "transfer": 0.26666666666666666, "held-out": 0.2875}, "acc_member_alone": {"in": 0.4125, "transfer": 0.2375, "held-out": 0.4}, "margin_all_on_minus_off": {"in": {"mean": 1.554516810149653, "se": 0.22773014690092566, "per_batch": null}, "transfer": {"mean": -0.1676435301701228, "se": 0.05868681520972574, "per_batch": null}, "held-out": {"mean": -0.12690404057502747, "se": 0.10126189715688672, "per_batch": null}}}}, "probes_all_on": {"p0_grounding": {"acc": 0.3333333333333333, "n": 30, "families": {"count": 0.125, "permanence": 0.5714285714285714, "sequence": 0.25, "attribute": 0.42857142857142855}}, "p1_self_other": {"acc": 0.5, "n": 30, "families": {"belief": 0.125, "reality": 0.6666666666666666, "perspective": 0.8, "reference": 0.8571428571428571, "second_order": 0.0}}, "p2_difference": {"acc": 0.5666666666666667, "n": 30, "families": {"category": 1.0, "antonym": 0.5, "odd_one_out": 0.5, "property": 0.5, "analogy": 0.3333333333333333}}, "p3_logic": {"acc": 0.3, "n": 30, "families": {"d1": 0.5, "d2": 0.25, "d3": 0.5, "d4": 0.5, "d5": 0.0, "d6": 0.0, "negation": 0.25, "contrapositive": 0.25}}, "p4_arithmetic": {"acc": 0.26666666666666666, "n": 30, "families": {"add2": 0.2, "add3": 0.0, "sub2": 0.25, "sub3": 0.4, "mul1": 0.25, "mul2": 0.3333333333333333, "word": 0.4}}, "p5_causal": {"acc": 0.5333333333333333, "n": 30, "families": {"effect": 0.7142857142857143, "cause": 0.42857142857142855, "counterfactual": 0.5, "prevention": 0.4, "chain": 0.6}}, "p6_learning": {"acc": 0.36666666666666664, "n": 30, "families": {"error_detect": 0.25, "correction": 0.25, "strategy": 0.7142857142857143, "check": 0.2857142857142857}}, "p7_composite": {"acc": 0.36666666666666664, "n": 30, "families": {"logic_arith": 0.25, "causal_category": 0.5, "belief_causal": 0.2857142857142857, "three_step": 0.42857142857142855}}, "p8_articulation": {"acc": 0.5, "n": 30, "families": {"definition": 0.375, "register": 0.5714285714285714, "grammar": 0.875, "paraphrase": 0.14285714285714285}}}, "probes_mean_off": 0.41111111111111115, "probes_mean_all_on": 0.4148148148148148, "gates": {"s1_perspective": {"mean": 0.21163099743248942, "max": 0.554569513244894}, "s2_concept": {"mean": 0.27282737975382454, "max": 0.48724048599867004}, "s3_rules": {"mean": 0.12374002504886239, "max": 0.17456351278495152}, "s4_arith": {"mean": 0.10761241921992425, "max": 0.16724209369972712}, "s5_causal": {"mean": 0.18640250725757604, "max": 0.3602895342887907}}, "chain_proof": "compiled chain: 13 cache entries (limit 256), 0 limit warnings", "curve": [{"step": 1, "train_bpb": {"s5_causal": 0.5002587062764781}, "quiet_kl": {"s5_causal": 0.00014205803745426238}}, {"step": 50, "train_bpb": {"s5_causal": 0.47479200806700805}, "quiet_kl": {"s5_causal": 0.00020960591034963728}}, {"step": 100, "train_bpb": {"s5_causal": 0.4432783216010956}, "quiet_kl": {"s5_causal": 0.0005776302301092073}}, {"step": 150, "train_bpb": {"s5_causal": 0.41218470104008903}, "quiet_kl": {"s5_causal": 0.0005999575663008728}}, {"step": 200, "train_bpb": {"s5_causal": 0.375140439483393}, "quiet_kl": {"s5_causal": 0.0006078194914152846}}, {"step": 250, "train_bpb": {"s5_causal": 0.33462434695546456}, "quiet_kl": {"s5_causal": 0.0006896784494165331}}, {"step": 300, "train_bpb": {"s5_causal": 0.2950140694214086}, "quiet_kl": {"s5_causal": 0.0008424301532795653}}, {"step": 350, "train_bpb": {"s5_causal": 0.25384222983344146}, "quiet_kl": {"s5_causal": 0.0009000897168880328}}, {"step": 400, "train_bpb": {"s5_causal": 0.2116303479520826}, "quiet_kl": {"s5_causal": 0.0008479387074476108}}, {"step": 450, "train_bpb": {"s5_causal": 0.17664345185901253}, "quiet_kl": {"s5_causal": 0.0008970295614562928}}, {"step": 500, "train_bpb": {"s5_causal": 0.1449180584045193}, "quiet_kl": {"s5_causal": 0.0008805390662746504}}, {"step": 550, "train_bpb": {"s5_causal": 0.11754954207523483}, "quiet_kl": {"s5_causal": 0.0008769319846760481}}, {"step": 600, "train_bpb": {"s5_causal": 0.09994425661305094}, "quiet_kl": {"s5_causal": 0.0008256271295249462}}, {"step": 650, "train_bpb": {"s5_causal": 0.08376807861643326}, "quiet_kl": {"s5_causal": 0.000822398122982122}}, {"step": 700, "train_bpb": {"s5_causal": 0.07323257448913689}, "quiet_kl": {"s5_causal": 0.0007506253628525882}}, {"step": 750, "train_bpb": {"s5_causal": 0.06620126664064982}, "quiet_kl": {"s5_causal": 0.0007284318481106311}}, {"step": 800, "train_bpb": {"s5_causal": 0.061003752291614005}, "quiet_kl": {"s5_causal": 0.0006977543700486422}}, {"step": 850, "train_bpb": {"s5_causal": 0.057040138757554856}, "quiet_kl": {"s5_causal": 0.0006314711138838903}}, {"step": 900, "train_bpb": {"s5_causal": 0.054913176967115146}, "quiet_kl": {"s5_causal": 0.0006655895477160811}}, {"step": 950, "train_bpb": {"s5_causal": 0.053277182541736884}, "quiet_kl": {"s5_causal": 0.0006091089290566743}}, {"step": 1000, "train_bpb": {"s5_causal": 0.051342804558971895}, "quiet_kl": {"s5_causal": 0.0005620899662608281}}, {"step": 1050, "train_bpb": {"s5_causal": 0.050292103103347294}, "quiet_kl": {"s5_causal": 0.0005693897540913895}}, {"step": 1100, "train_bpb": {"s5_causal": 0.04907071886323294}, "quiet_kl": {"s5_causal": 0.0005052202951628715}}, {"step": 1150, "train_bpb": {"s5_causal": 0.048908725609667185}, "quiet_kl": {"s5_causal": 0.0004889974341494962}}, {"step": 1200, "train_bpb": {"s5_causal": 0.048053565955822976}, "quiet_kl": {"s5_causal": 0.0004733337758807465}}, {"step": 1250, "train_bpb": {"s5_causal": 0.0479133836495288}, "quiet_kl": {"s5_causal": 0.0004842422466026619}}, {"step": 1300, "train_bpb": {"s5_causal": 0.047709994924324456}, "quiet_kl": {"s5_causal": 0.0004992673854576423}}, {"step": 1350, "train_bpb": {"s5_causal": 0.047162081900034625}, "quiet_kl": {"s5_causal": 0.00045933669025544077}}, {"step": 1400, "train_bpb": {"s5_causal": 0.04675838111176074}, "quiet_kl": {"s5_causal": 0.00044424480700399726}}, {"step": 1450, "train_bpb": {"s5_causal": 0.04637059484557359}, "quiet_kl": {"s5_causal": 0.00043485293921548875}}, {"step": 1500, "train_bpb": {"s5_causal": 0.045869359148853234}, "quiet_kl": {"s5_causal": 0.0004135572590166703}}, {"step": 1550, "train_bpb": {"s5_causal": 0.04566729297023403}, "quiet_kl": {"s5_causal": 0.0003929612482897937}}, {"step": 1600, "train_bpb": {"s5_causal": 0.045591114437415065}, "quiet_kl": {"s5_causal": 0.00038184704666491596}}, {"step": 1650, "train_bpb": {"s5_causal": 0.04532819047796156}, "quiet_kl": {"s5_causal": 0.000387379300664179}}, {"step": 1700, "train_bpb": {"s5_causal": 0.04531057596111561}, "quiet_kl": {"s5_causal": 0.00041130212775897235}}, {"step": 1750, "train_bpb": {"s5_causal": 0.0453029528301161}, "quiet_kl": {"s5_causal": 0.00037816873897099867}}, {"step": 1800, "train_bpb": {"s5_causal": 0.04487343997194289}, "quiet_kl": {"s5_causal": 0.0003597767755854875}}, {"step": 1850, "train_bpb": {"s5_causal": 0.04524810664909812}, "quiet_kl": {"s5_causal": 0.0004809653048869222}}, {"step": 1900, "train_bpb": {"s5_causal": 0.04470151380663197}, "quiet_kl": {"s5_causal": 0.00040219156537204983}}, {"step": 1950, "train_bpb": {"s5_causal": 0.04444344491741231}, "quiet_kl": {"s5_causal": 0.000377581731590908}}, {"step": 2000, "train_bpb": {"s5_causal": 0.04480685688866136}, "quiet_kl": {"s5_causal": 0.000341146384307649}}, {"step": 2050, "train_bpb": {"s5_causal": 0.04417530316556535}, "quiet_kl": {"s5_causal": 0.0003392991560394876}}, {"step": 2100, "train_bpb": {"s5_causal": 0.04453701003632025}, "quiet_kl": {"s5_causal": 0.00037088638258865104}}, {"step": 2150, "train_bpb": {"s5_causal": 0.044109676305822625}, "quiet_kl": {"s5_causal": 0.00034736187139060347}}, {"step": 2200, "train_bpb": {"s5_causal": 0.04388423892107647}, "quiet_kl": {"s5_causal": 0.0003461908106692135}}, {"step": 2250, "train_bpb": {"s5_causal": 0.04488547709142136}, "quiet_kl": {"s5_causal": 0.0003610755910631269}}, {"step": 2300, "train_bpb": {"s5_causal": 0.044453946231818586}, "quiet_kl": {"s5_causal": 0.00036063317384105176}}, {"step": 2350, "train_bpb": {"s5_causal": 0.04401741132201064}, "quiet_kl": {"s5_causal": 0.00035092675738269464}}, {"step": 2400, "train_bpb": {"s5_causal": 0.04412462095324766}, "quiet_kl": {"s5_causal": 0.00036637546436395495}}, {"step": 2450, "train_bpb": {"s5_causal": 0.043844493945440384}, "quiet_kl": {"s5_causal": 0.00031527867686236276}}, {"step": 2500, "train_bpb": {"s5_causal": 0.043894696809863396}, "quiet_kl": {"s5_causal": 0.0003275712367030792}}, {"step": 2550, "train_bpb": {"s5_causal": 0.0436980059489969}, "quiet_kl": {"s5_causal": 0.0003616753814276308}}, {"step": 2600, "train_bpb": {"s5_causal": 0.04377718321431197}, "quiet_kl": {"s5_causal": 0.0003302946028998122}}, {"step": 2650, "train_bpb": {"s5_causal": 0.043987670147218626}, "quiet_kl": {"s5_causal": 0.0003459362892317586}}, {"step": 2700, "train_bpb": {"s5_causal": 0.043827061568970614}, "quiet_kl": {"s5_causal": 0.00037086522148456426}}, {"step": 2750, "train_bpb": {"s5_causal": 0.04347486001809519}, "quiet_kl": {"s5_causal": 0.00032774969935417175}}, {"step": 2800, "train_bpb": {"s5_causal": 0.04351465411627592}, "quiet_kl": {"s5_causal": 0.0003318996904999949}}, {"step": 2850, "train_bpb": {"s5_causal": 0.04360634376437619}, "quiet_kl": {"s5_causal": 0.0003193380776792765}}, {"step": 2900, "train_bpb": {"s5_causal": 0.04331495796607476}, "quiet_kl": {"s5_causal": 0.0003052758570993319}}, {"step": 2950, "train_bpb": {"s5_causal": 0.04328516256300314}, "quiet_kl": {"s5_causal": 0.00031300869071856143}}, {"step": 3000, "train_bpb": {"s5_causal": 0.04325561411626924}, "quiet_kl": {"s5_causal": 0.0003185829348512925}}, {"step": 3050, "train_bpb": {"s5_causal": 0.0431763945002264}, "quiet_kl": {"s5_causal": 0.0003293832784402184}}, {"step": 3100, "train_bpb": {"s5_causal": 0.04366199477045563}, "quiet_kl": {"s5_causal": 0.00040295566810527817}}, {"step": 3150, "train_bpb": {"s5_causal": 0.04358666341342591}, "quiet_kl": {"s5_causal": 0.00035175218421500175}}, {"step": 3200, "train_bpb": {"s5_causal": 0.043320469472589135}, "quiet_kl": {"s5_causal": 0.0003443358623189852}}, {"step": 3250, "train_bpb": {"s5_causal": 0.042987033324961385}, "quiet_kl": {"s5_causal": 0.00032035535405157135}}, {"step": 3300, "train_bpb": {"s5_causal": 0.04294536445465251}, "quiet_kl": {"s5_causal": 0.00031660375418141485}}, {"step": 3350, "train_bpb": {"s5_causal": 0.04280402212537826}, "quiet_kl": {"s5_causal": 0.00030007210705662146}}, {"step": 3400, "train_bpb": {"s5_causal": 0.04321648010900036}, "quiet_kl": {"s5_causal": 0.0003129728915519081}}, {"step": 3450, "train_bpb": {"s5_causal": 0.04280023442247132}, "quiet_kl": {"s5_causal": 0.00027893567079445347}}, {"step": 3500, "train_bpb": {"s5_causal": 0.04277529446584561}, "quiet_kl": {"s5_causal": 0.000287664610077627}}, {"step": 3550, "train_bpb": {"s5_causal": 0.04294359002365223}, "quiet_kl": {"s5_causal": 0.00033117909071734176}}, {"step": 3600, "train_bpb": {"s5_causal": 0.04295575693628753}, "quiet_kl": {"s5_causal": 0.00033154026750708}}, {"step": 3650, "train_bpb": {"s5_causal": 0.04301591867108704}, "quiet_kl": {"s5_causal": 0.0003191953044733964}}, {"step": 3700, "train_bpb": {"s5_causal": 0.04285260932028557}, "quiet_kl": {"s5_causal": 0.00032508621108718217}}, {"step": 3750, "train_bpb": {"s5_causal": 0.04276063665323546}, "quiet_kl": {"s5_causal": 0.00029929409531177954}}, {"step": 3800, "train_bpb": {"s5_causal": 0.0428498428181127}, "quiet_kl": {"s5_causal": 0.00029512351466109976}}, {"step": 3850, "train_bpb": {"s5_causal": 0.0425406144647192}, "quiet_kl": {"s5_causal": 0.00029631719371536747}}, {"step": 3900, "train_bpb": {"s5_causal": 0.04262807828410946}, "quiet_kl": {"s5_causal": 0.0003191924002021551}}, {"step": 3950, "train_bpb": {"s5_causal": 0.042581383489552004}, "quiet_kl": {"s5_causal": 0.0003132560817175545}}, {"step": 4000, "train_bpb": {"s5_causal": 0.04247512126774293}, "quiet_kl": {"s5_causal": 0.00029696322890231385}}], "steps_done": {"s1_perspective": 3200, "s2_concept": 2400, "s3_rules": 800, "s4_arith": 800, "s5_causal": 4000, "s6_tryfail": 0, "s7_mixed": 0, "s8_register": 0}, "stage_seconds": 2749.4498569965363, "inert_at_birth": true}, {"close": 6, "attached": ["s1_perspective", "s2_concept", "s3_rules", "s4_arith", "s5_causal", "s6_tryfail"], "rows": {"s1_perspective": {"off": 0.5538333013185592, "all_on": 0.048118676545145664, "all_on_minus_off": {"mean": -0.5057146247734136, "se": 0.004820240645617063, "per_batch": [-0.507592519461558, -0.5249112315224791, -0.5168167122003438, -0.5112517513283034, -0.49226029300942786, -0.5082934025084024, -0.48163300668586717, -0.5029580814709266]}}, "s2_concept": {"off": 0.7211839326967642, "all_on": 0.12151103130108834, "all_on_minus_off": {"mean": -0.5996729013956759, "se": 0.0037475212590685042, "per_batch": [-0.5988467074898466, -0.5971387047884653, -0.5870160209369939, -0.6220082775143131, -0.5981930659274111, -0.5979833223357682, -0.590672683812892, -0.6055244283597164]}}, "s3_rules": {"off": 0.9296131361815035, "all_on": 0.2224989079841146, "all_on_minus_off": {"mean": -0.7071142281973889, "se": 0.005464077448023725, "per_batch": [-0.688091450225454, -0.7052577265386744, -0.7221855489064735, -0.6908666482711018, -0.6926657157861148, -0.7296803378527611, -0.7136468467672801, -0.7145195512312519]}}, "s4_arith": {"off": 1.2338748670942346, "all_on": 0.3363804556765427, "all_on_minus_off": {"mean": -0.8974944114176918, "se": 0.005832828072572422, "per_batch": [-0.913781531230304, -0.8892367262427792, -0.9124813423897811, -0.8729642867617557, -0.8777276472470401, -0.9162305641785904, -0.8992278003706777, -0.8983053929206066]}}, "s5_causal": {"off": 0.6604376826115352, "all_on": 0.04261427976669097, "all_on_minus_off": {"mean": -0.6178234028448443, "se": 0.006924028249021772, "per_batch": [-0.5837280243710518, -0.6273962816747602, -0.6286863504111713, -0.6292703819956701, -0.6433750688411224, -0.6161043764586269, -0.5949296360082531, -0.6190971029980981]}}, "s6_tryfail": {"off": 0.6728212611401647, "all_on": 0.05458008825131443, "all_on_minus_off": {"mean": -0.6182411728888503, "se": 0.003591826211933704, "per_batch": [-0.6275335372662934, -0.6247747796871767, -0.6204394947688053, -0.6271889109062535, -0.6229506242029689, -0.6172677369770149, -0.6055252184050188, -0.6002490808972711]}}, "s7_mixed": {"off": 0.7829759286650222, "all_on": 0.7763963238974697, "all_on_minus_off": {"mean": -0.006579604767552552, "se": 0.00858680870634249, "per_batch": [0.0010782452295843292, 0.0011276042503820527, 0.002170507045214043, 0.0020096172753318875, -0.0001660492494079957, -0.06482873125574762, 0.014021960503158404, -0.008049991938935519]}}, "s8_register": {"off": 0.4906504977761471, "all_on": 0.37942930133979524, "all_on_minus_off": {"mean": -0.11122119643635187, "se": 0.002113655305310244, "per_batch": [-0.09977861560746037, -0.11433664603335814, -0.10767472606953638, -0.11137170812814601, -0.11099528110107004, -0.12054900334782448, -0.11432748795719966, -0.1107361032462198]}}, "web": {"off": 0.9505967609111062, "all_on": 0.9520687174241782, "all_on_minus_off": {"mean": 0.0014719565130720813, "se": 0.00034427219185691194, "per_batch": [0.0010453965432695167, 0.00048146543109039985, 0.0006710763036666956, 0.0009668864631507956, 0.003104544822050137, 0.002838057704538044, 0.0012700918766218594, 0.0013981329601892023]}}}, "member_in_group": {"s1_perspective": {"s1_perspective": {"masked": 0.5211283065085587, "all_on_minus_masked": {"mean": -0.473009629963413, "se": 0.0042184550278582915, "per_batch": [-0.4781080717716144, -0.4882468091286523, -0.48416804170850075, -0.47732335793139025, -0.45534327109714623, -0.47349954941668493, -0.45629961824847376, -0.47108832040484133]}}, "s2_concept": {"masked": 0.3584392783181347, "all_on_minus_masked": {"mean": -0.2369282470170464, "se": 0.0015036758415095223, "per_batch": [-0.2349426401321123, -0.2358342411898478, -0.23079132290558585, -0.24440574901969336, -0.2355921326130957, -0.23675754079992045, -0.23532458135293155, -0.24177776812318424]}}, "s3_rules": {"masked": 0.4201675896482241, "all_on_minus_masked": {"mean": -0.1976686816641095, "se": 0.003142655945218954, "per_batch": [-0.19520039472061324, -0.20349421506280949, -0.19871847182666286, -0.18723127762324596, -0.19808669355871833, -0.2156948792933307, -0.19210636233809725, -0.19081715888939826]}}, "s4_arith": {"masked": 0.4856595827459398, "all_on_minus_masked": {"mean": -0.14927912706939708, "se": 0.0052060855236191915, "per_batch": [-0.1613788025498859, -0.16687493811483406, -0.11814270808827754, -0.15186657111765212, -0.1562410788293358, -0.14900787549501043, -0.1435791141335378, -0.14714192822664307]}}, "s5_causal": {"masked": 0.08852680384680955, "all_on_minus_masked": {"mean": -0.04591252408011858, "se": 0.002775179316447826, "per_batch": [-0.046082590064954396, -0.06253310951925972, -0.041200069787682245, -0.041360223256840405, -0.043164829150657086, -0.05043869025853792, -0.04577507433629601, -0.036745606266720875]}}, "s6_tryfail": {"masked": 0.07867829661840395, "all_on_minus_masked": {"mean": -0.024098208367089514, "se": 0.002347562091448561, "per_batch": [-0.03557417599988697, -0.026186465381335677, -0.019685628667878298, -0.021739837819904254, -0.027626594355082454, -0.025710256170012248, -0.012596358688951668, -0.02366634985366456]}}, "web": {"masked": 0.9520075130983024, "all_on_minus_masked": {"mean": 6.120432587582159e-05, "se": 0.0003520672031210196, "per_batch": [0.00026072569870794204, -0.0006300584414359323, -0.0014090538585189716, -0.0007207792897659937, 0.0016319433740372702, 0.0006713342776429876, 0.000913915806684984, -0.00022839296034571355]}}}, "s2_concept": {"s1_perspective": {"masked": 0.05240656210440904, "all_on_minus_masked": {"mean": -0.004287885559263373, "se": 0.00032587601541498077, "per_batch": [-0.004196623906113146, -0.004145416071818744, -0.004605077327454479, -0.0032324676675448832, -0.00618445312936388, -0.0038440917185914233, -0.0034172898981861682, -0.004677664755034262]}}, "s2_concept": {"masked": 0.24312670867076885, "all_on_minus_masked": {"mean": -0.12161567736968051, "se": 0.0013752552984466323, "per_batch": [-0.126788942379923, -0.12418407165212354, -0.11782082105935641, -0.12631179801293546, -0.12240074054966435, -0.12015308853662264, -0.11673263308276903, -0.11853332368404966]}}, "s3_rules": {"masked": 0.3236720905989877, "all_on_minus_masked": {"mean": -0.10117318261487313, "se": 0.000979548965453322, "per_batch": [-0.1017938653146081, -0.10001749950950886, -0.10304116948999059, -0.10107521431014646, -0.09648084827716602, -0.10342202507032627, -0.10499515037776841, -0.09855968856947023]}}, "s4_arith": {"masked": 0.44821822872684014, "all_on_minus_masked": {"mean": -0.11183777305029749, "se": 0.00298070739832692, "per_batch": [-0.12252547096720262, -0.1100271477637006, -0.09834643008240035, -0.11752877302036002, -0.10404172155704772, -0.12145883456655149, -0.10837740418529912, -0.112396402259818]}}, "s5_causal": {"masked": 0.04340456859921285, "all_on_minus_masked": {"mean": -0.0007902888325218788, "se": 0.00013779430005367099, "per_batch": [-0.000683883636698121, -0.0007203815798859092, -0.0012787635643458153, -0.001046001169776524, -0.000820980681724974, -0.0008737793555398596, -0.0009505427368609043, 5.2022064657077094e-05]}}, "s6_tryfail": {"masked": 0.057036200032382633, "all_on_minus_masked": {"mean": -0.0024561117810681966, "se": 0.0005487683573118232, "per_batch": [-0.002715697422907018, -0.006181905636347976, -0.0019421947017358582, -0.0015636125170662638, -0.002070633495183341, -0.001520031038446057, -0.0016873164131571763, -0.001967503023701883]}}, "web": {"masked": 0.9526002620534169, "all_on_minus_masked": {"mean": -0.0005315446292385451, "se": 0.00020022444522142146, "per_batch": [-0.0010765254030757898, 1.1264863631454958e-05, -0.00035462822607912337, 0.0002851472351306228, -0.0007795113650357344, -0.0007439109563071034, -0.0013918555934326893, -0.00020233758873999808]}}}, "s3_rules": {"s1_perspective": {"masked": 0.04795871319743358, "all_on_minus_masked": {"mean": 0.00015996334771208043, "se": 5.782804536322485e-05, "per_batch": [0.00018473623931585664, 0.0001321310459831901, 8.37877977172205e-05, 6.541790082194299e-05, 0.0004755266551778256, 0.00020815275212239526, -9.500966568602026e-05, 0.0002249640562442326]}}, "s2_concept": {"masked": 0.12200165077749384, "all_on_minus_masked": {"mean": -0.0004906194764054995, "se": 0.00010704578034618784, "per_batch": [-0.0008811101160328821, -0.0006804923538014235, -0.0003156741556586823, -0.0008833781372411392, -0.0002483429478458976, -0.00036336709452607086, -5.098210706515338e-05, -0.0005016088990727474]}}, "s3_rules": {"masked": 0.31812585165083107, "all_on_minus_masked": {"mean": -0.09562694366671651, "se": 0.0013313753211474974, "per_batch": [-0.09457684984729525, -0.09383192849291436, -0.09697740518586967, -0.10034089138662497, -0.08924439877030571, -0.1006879093804034, -0.09584911300488425, -0.09350705326543446]}}, "s4_arith": {"masked": 0.33684335504302665, "all_on_minus_masked": {"mean": -0.00046289936648401253, "se": 0.00012979059091528704, "per_batch": [-0.0004255925673913863, -0.0004372873876500316, -0.00042505512160745385, -0.0008198627944943637, 0.00026442332570153493, -0.0006526741600243469, -0.0009259545922453705, -0.0002811916341606824]}}, "s5_causal": {"masked": 0.04274135042471469, "all_on_minus_masked": {"mean": -0.00012707065802372314, "se": 8.394228721051919e-05, "per_batch": [-0.00020733583453079463, 0.0002089938547742698, -9.539931387938377e-05, 1.9087387016850088e-05, -0.00013297483586398318, -3.602767812684038e-05, -0.0006209352992534098, -0.0001519735443264933]}}, "s6_tryfail": {"masked": 0.054738190692999594, "all_on_minus_masked": {"mean": -0.0001581024416851635, "se": 9.770562463135943e-05, "per_batch": [-8.571722808158738e-05, -0.00048020780795599016, -0.00010251778328775052, 0.0001395854190065285, 0.00017278881953872866, -0.00047619308794991794, 2.502885015837758e-05, -0.0004575867149096968]}}, "web": {"masked": 0.9517037164944689, "all_on_minus_masked": {"mean": 0.00036500092970929987, "se": 0.00017890858600685294, "per_batch": [0.00021970783647717873, 0.000460225573708839, -4.643531573256254e-06, 0.0005819892905197799, 0.001131989807979128, 0.000926728514174191, -0.0003408696140100309, -5.5120439601430427e-05]}}}, "s4_arith": {"s1_perspective": {"masked": 0.04802085435849153, "all_on_minus_masked": {"mean": 9.78221866541366e-05, "se": 5.457332786330251e-05, "per_batch": [0.00014164921081688542, 0.00014129987105731406, -7.153403384328738e-06, 0.00037795874756021974, -1.7241260748993015e-05, 0.00022315823841017463, -9.909694087293502e-05, 2.2003030394755763e-05]}}, "s2_concept": {"masked": 0.12171173102380185, "all_on_minus_masked": {"mean": -0.0002006997227135196, "se": 4.8420076914516294e-05, "per_batch": [-0.000255727452917312, -0.00035890629451933165, -0.00029995924093609294, -0.00016683392025260146, -1.8961087257629572e-05, -1.4704516648769705e-05, -0.0001436592580488505, -0.00034684601112756897]}}, "s3_rules": {"masked": 0.2225592819562542, "all_on_minus_masked": {"mean": -6.037397213961085e-05, "se": 9.72669745423061e-05, "per_batch": [-2.5840393292148045e-05, -1.8058178340718545e-06, -0.00035168302318308586, 0.0005106594860743996, -9.388102953974964e-05, -0.00031014921299971077, -0.000257995474125583, 4.7703687783062776e-05]}}, "s4_arith": {"masked": 0.5118949449446989, "all_on_minus_masked": {"mean": -0.17551448926815622, "se": 0.0018546283723838704, "per_batch": [-0.18157055466142213, -0.17318033804341726, -0.1781877419102761, -0.1725092187440878, -0.1651040327588998, -0.1803674500273113, -0.17645686551633005, -0.17673971248350523]}}, "s5_causal": {"masked": 0.042685293149931025, "all_on_minus_masked": {"mean": -7.101338324005078e-05, "se": 5.9225878861614116e-05, "per_batch": [-5.8097889244594636e-05, 0.0001703703135109702, -0.0001579499414439689, 0.00017677398002668987, -0.0001797568041275871, -0.00020407622585116164, -0.00026833593100869524, -4.7034567782058834e-05]}}, "s6_tryfail": {"masked": 0.0586756750372746, "all_on_minus_masked": {"mean": -0.004095586785960164, "se": 0.0005802648080000451, "per_batch": [-0.005020694901095808, -0.00756470526832119, -0.0022897823880452736, -0.0036670463284438418, -0.00394842606862892, -0.003384220859099951, -0.0026879544715397094, -0.0042018640025066195]}}, "web": {"masked": 0.9519098699482751, "all_on_minus_masked": {"mean": 0.00015884747590316772, "se": 0.00013080205005513813, "per_batch": [0.0002537604013481687, -0.00012477341320094038, -0.00010869303534533081, 0.00032143557446262516, 0.00019846797909561786, 0.00041808982424740293, -0.00043614800258806596, 0.0007486404792058643]}}}, "s5_causal": {"s1_perspective": {"masked": 0.0480574527368602, "all_on_minus_masked": {"mean": 6.122380828547068e-05, "se": 4.6083293019414016e-05, "per_batch": [0.00015873998674637707, 6.0570139850738e-06, 0.00023974381530274952, 0.00016334589711479441, -1.503773303481587e-05, 2.4163562446229137e-05, -0.00017571789906123403, 8.84958227845914e-05]}}, "s2_concept": {"masked": 0.12149394052515884, "all_on_minus_masked": {"mean": 1.7090775929491653e-05, "se": 0.00010358387168329964, "per_batch": [-0.00012482715777936693, 0.0002473755454347887, -0.00022030977575525357, 0.0002675297623327827, 0.0001679625563988929, 0.00027119514257929744, 7.459747481174117e-05, -0.0005467973405869492]}}, "s3_rules": {"masked": 0.22219914222366072, "all_on_minus_masked": {"mean": 0.00029976576045387393, "se": 0.00012127499679584701, "per_batch": [0.00021355945670886722, 0.0005090041530598222, -0.00011015488787763372, 0.00041254338375718014, 0.0010037552439294273, 9.149477025902075e-05, 0.00016095426337622842, 0.00011696970041807919]}}, "s4_arith": {"masked": 0.3364524465393023, "all_on_minus_masked": {"mean": -7.199086275962163e-05, "se": 0.00013391023944769455, "per_batch": [0.00034538415859519667, 0.00036748392923108275, -0.0004447041394685103, -3.6460321982956234e-05, 0.0001571276493945173, 7.760717120186555e-05, -0.00042565706088548705, -0.0006167082881626818]}}, "s5_causal": {"masked": 0.488108991906275, "all_on_minus_masked": {"mean": -0.44549471213958397, "se": 0.005776856023464081, "per_batch": [-0.42210423253501184, -0.4486892381502039, -0.4441040785635025, -0.4570613159730006, -0.4698508084916423, -0.45719571772742096, -0.4251721396779702, -0.4397801659979195]}}, "s6_tryfail": {"masked": 0.05488085097369753, "all_on_minus_masked": {"mean": -0.000300762722383093, "se": 5.7481366367343595e-05, "per_batch": [-0.0005620930475980501, -0.000405545839650176, -0.00030330215371223573, -0.00038923436010741136, -0.00024994453628205543, -0.0003029796862418707, -1.9396418342622346e-05, -0.00017360573713032235]}}, "web": {"masked": 0.9520541848901802, "all_on_minus_masked": {"mean": 1.4532533997926866e-05, "se": 0.00015858429567688703, "per_batch": [0.0005859448914895538, -0.00016622123205889405, -0.0004735542291507411, 0.0006929181003263452, 0.00013070681465587874, -0.0001078331220910611, 7.395253987074746e-06, -0.0005530962051747412]}}}, "s6_tryfail": {"s1_perspective": {"masked": 0.047946781229222746, "all_on_minus_masked": {"mean": 0.00017189531592291807, "se": 6.917971309464499e-05, "per_batch": [1.3328655441867399e-05, 0.0003855367331138601, 7.831122517881067e-05, 0.0003678923880269039, 0.00044360237561141824, -7.055588251646261e-05, 7.991281361497543e-05, 7.713421891197142e-05]}}, "s2_concept": {"masked": 0.12118336001191751, "all_on_minus_masked": {"mean": 0.00032767128917082604, "se": 7.976340449814361e-05, "per_batch": [0.00015132323492791921, 0.0006691414988444777, 0.000408974743751761, 0.00021088297670485145, 0.0006077544414021252, 8.161651675007986e-05, 0.00037990967575593504, 0.00011176722522945881]}}, "s3_rules": {"masked": 0.22262603003539144, "all_on_minus_masked": {"mean": -0.00012712205127681647, "se": 0.0001083326894862117, "per_batch": [-0.0005776897242481815, 0.00019790903548036476, -0.00045622697707631277, 0.0002313166654104859, 0.00011204669703709924, -0.0003178669344571783, -0.00019554427403101204, -1.0920898329797035e-05]}}, "s4_arith": {"masked": 0.3379316531959998, "all_on_minus_masked": {"mean": -0.0015511975194571065, "se": 0.00018033094493480532, "per_batch": [-0.0015077504022829036, -0.0017367882975694582, -0.0012166267700348676, -0.0014314975944565989, -0.0014939917902138666, -0.002333073645942152, -0.0006525451730362009, -0.0020373064821208042]}}, "s5_causal": {"masked": 0.04257019342493747, "all_on_minus_masked": {"mean": 4.408634175349837e-05, "se": 8.932801050887662e-05, "per_batch": [-0.00011414811005235059, 8.334977940330918e-05, -3.856442222706369e-05, 7.582822565697922e-05, 0.00019396418342620958, -0.00017399001086584875, 0.0005594031314494069, -0.0002331520427626549]}}, "s6_tryfail": {"masked": 0.487750171603081, "all_on_minus_masked": {"mean": -0.4331700833517666, "se": 0.0029937285894294893, "per_batch": [-0.4424756044020007, -0.4195138728433789, -0.4415956390454876, -0.432332610474354, -0.4364200898906634, -0.4392329736346814, -0.43083465770871243, -0.4229552188148545]}}, "web": {"masked": 0.9519801786057309, "all_on_minus_masked": {"mean": 8.853881844730904e-05, "se": 0.00012499413632642052, "per_batch": [0.00029305843707017054, 0.0003485228419735087, 0.0005759698977395589, 8.865705651983635e-05, -0.00010972493125049887, -3.9212044396719214e-05, -0.0005961778592159517, 0.00014721714913856765]}}}}, "member_alone": {"s1_perspective": {"s1_perspective": {"alone": 0.05217516482213097, "alone_minus_off": {"mean": -0.5016581364964282, "se": 0.004888586924071733, "per_batch": [-0.503264758784162, -0.5212149750156778, -0.5118989327924157, -0.509032675307594, -0.4859917833594548, -0.5041102201213458, -0.4784889810965285, -0.4992627654942472]}}, "s2_concept": {"alone": 0.24394271260453448, "alone_minus_off": {"mean": -0.47724122009222963, "se": 0.0030459897360259366, "per_batch": [-0.47044852344580024, -0.4716034084441754, -0.4687537634085354, -0.4938985943670891, -0.47575242540274654, -0.47682052365593, -0.47438017583151226, -0.4862723461820483]}}, "s3_rules": {"alone": 0.5034592174177095, "alone_minus_off": {"mean": -0.42615391876379405, "se": 0.003854800114533744, "per_batch": [-0.4117213926773634, -0.4236871716024687, -0.442290362963648, -0.410341145912864, -0.4260619510412451, -0.4339105083003576, -0.4286801719223179, -0.4325386456900876]}}, "s4_arith": {"alone": 0.7531657780810914, "alone_minus_off": {"mean": -0.480709089013143, "se": 0.003514095864453405, "per_batch": [-0.4824103037742731, -0.4703906082881222, -0.490546115055989, -0.47029558787352066, -0.4725516132918761, -0.49690809727670615, -0.47641963209676885, -0.48615075444788836]}}, "s5_causal": {"alone": 0.468494725289589, "alone_minus_off": {"mean": -0.19194295732194622, "se": 0.002210700040436461, "per_batch": [-0.1875164678540393, -0.1992422879855283, -0.19983549914401677, -0.1947394812163009, -0.19148918109764468, -0.18138197568475095, -0.1878665385398705, -0.19347222705341832]}}, "s6_tryfail": {"alone": 0.5038253093623601, "alone_minus_off": {"mean": -0.16899595177780455, "se": 0.002363709528038538, "per_batch": [-0.1689674510278819, -0.1804920514578614, -0.16629625949034343, -0.17642276995579653, -0.17014793994340416, -0.16627970616019783, -0.16280797837855587, -0.16055345780839525]}}, "web": {"alone": 0.9519326898962615, "alone_minus_off": {"mean": 0.001335928985155277, "se": 0.0004171900864352844, "per_batch": [0.0008697162654132118, -0.0001570201602377752, 0.0007944738556607822, 0.00027680607656377365, 0.003323134771296754, 0.002440089850441529, 0.002121320007067329, 0.0010189112150366109]}}}, "s2_concept": {"s1_perspective": {"alone": 0.5237967463236832, "alone_minus_off": {"mean": -0.03003655499487594, "se": 0.0008929364404090443, "per_batch": [-0.02682404806374683, -0.03162864138957566, -0.02918588580804582, -0.032230279697956554, -0.03334249150108626, -0.031257674811664526, -0.026360984776298668, -0.02946243391063319]}}, "s2_concept": {"alone": 0.36033454325401637, "alone_minus_off": {"mean": -0.36084938944274775, "se": 0.0021975474547336524, "per_batch": [-0.36082329913716604, -0.3578469673813343, -0.3548674109507988, -0.37462086527661115, -0.36102178861409306, -0.36006399573360676, -0.3551193225386502, -0.36243146590972153]}}, "s3_rules": {"alone": 0.6157804256378956, "alone_minus_off": {"mean": -0.31383271054360795, "se": 0.0028684777591963, "per_batch": [-0.3051689393960323, -0.3130032220951231, -0.3208357849677054, -0.3010766122187518, -0.30994988510736754, -0.3192230176592393, -0.3165332949910704, -0.32487092791357386]}}, "s4_arith": {"alone": 0.8761517594682684, "alone_minus_off": {"mean": -0.35772310762596604, "se": 0.004197781721024297, "per_batch": [-0.3465006699646368, -0.3508452096994077, -0.37445335417467074, -0.3532194301945688, -0.34704009354906795, -0.37528652412677577, -0.36393756097898977, -0.35050201831961103]}}, "s5_causal": {"alone": 0.6477866226906935, "alone_minus_off": {"mean": -0.012651059920841742, "se": 0.0009352908521067744, "per_batch": [-0.007499647456174596, -0.011507402164893321, -0.015175620125146061, -0.01262455846923527, -0.013140119460859356, -0.011161158093043633, -0.015682152027599816, -0.014417821569781886]}}, "s6_tryfail": {"alone": 0.6666359373486308, "alone_minus_off": {"mean": -0.0061853237915339004, "se": 0.0012459523146970725, "per_batch": [-0.007500163404127069, -0.013446420562046035, -0.006764636602049623, -0.006561095134753447, -0.006009288799460166, -0.003128321423531699, -0.0018538439893129866, -0.004218820416990177]}}, "web": {"alone": 0.9514616616622953, "alone_minus_off": {"mean": 0.0008649007511890849, "se": 0.00018835522783446044, "per_batch": [5.219673453682461e-05, 0.0011502199689704673, 0.0014158471732280686, 0.00112304671013419, 0.0007925820465013445, 0.0004966858956918507, 0.00034379331907474775, 0.001544834161375186]}}}, "s3_rules": {"s1_perspective": {"alone": 0.5535631380718851, "alone_minus_off": {"mean": -0.00027016324667410163, "se": 0.00023913782920347164, "per_batch": [0.0002715606057124287, -0.0004794876306054574, -0.0007019041938338688, 0.0007255088126647546, -0.0014021745524844809, -0.0003248322318171182, 0.00029976576045387393, -0.000549742543482945]}}, "s2_concept": {"alone": 0.7227973879358336, "alone_minus_off": {"mean": 0.0016134552390694584, "se": 0.0002752200095664509, "per_batch": [0.000995564570182128, 0.0010287142261358184, 0.0016847850435147693, 0.0009734218038836007, 0.0022790280979085864, 0.0019874315133706633, 0.0009180433903056562, 0.003040653267254445]}}, "s3_rules": {"alone": 0.8701872076558641, "alone_minus_off": {"mean": -0.05942592852563948, "se": 0.010330913371419828, "per_batch": [-0.06883390622470331, -0.052931358430022946, -0.047376318807154205, -0.08934343927937138, -0.029561753891506393, -0.06865332444129746, -0.10310746880189825, -0.015599858329161886]}}, "s4_arith": {"alone": 1.2291967885059467, "alone_minus_off": {"mean": -0.0046780785882879206, "se": 0.0004241556239844928, "per_batch": [-0.0034991590144548557, -0.00447249482701273, -0.00568497251559541, -0.003813715282882768, -0.004839591795279485, -0.0037889497811585127, -0.00709909986231505, -0.004226645627604553]}}, "s5_causal": {"alone": 0.666408248442262, "alone_minus_off": {"mean": 0.005970565830726776, "se": 0.0006340753987558168, "per_batch": [0.006426389723465098, 0.007101077662799882, 0.001980165246371679, 0.007604900838502515, 0.006885497409943375, 0.006719362169209986, 0.004959076741966384, 0.00608805685355529]}}, "s6_tryfail": {"alone": 0.6717302999433321, "alone_minus_off": {"mean": -0.0010909611968325655, "se": 0.0003520287905236149, "per_batch": [-0.0010005950627198779, -0.0029078826607884523, -0.00020315450633168197, -0.0009048437218519556, -0.0018541019632892786, -0.0005386496625022774, 0.00022035277141796428, -0.001538814768594965]}}, "web": {"alone": 0.9509403070051203, "alone_minus_off": {"mean": 0.0003435460940141577, "se": 0.0001181112928609609, "per_batch": [0.00013586629418149698, 0.00014833503636912937, 0.001022264876728518, 0.0005037371843772398, 0.0003148142424045375, 0.0005318563477932914, 4.4629497898962533e-05, 4.6865272360085974e-05]}}}, "s4_arith": {"s1_perspective": {"alone": 0.5525393683469615, "alone_minus_off": {"mean": -0.0012939329715976888, "se": 0.00033731606119881077, "per_batch": [-0.0005440671160045207, -0.002287799213102515, 5.077787766727404e-05, -0.002374908425764488, -0.002254219600521523, -0.0003551871696944042, -0.00143721601759772, -0.0011488441077636136]}}, "s2_concept": {"alone": 0.7197303675784188, "alone_minus_off": {"mean": -0.001453565118345393, "se": 0.00024222923191314095, "per_batch": [-0.0015742001990100007, -0.001012590852617512, -0.0006298004674595292, -0.0029371197114349545, -0.0012482930756249067, -0.0015436732784818163, -0.001588259780718082, -0.0010945835814163418]}}, "s3_rules": {"alone": 0.9300652463238748, "alone_minus_off": {"mean": 0.000452110142371287, "se": 0.0003056447991096855, "per_batch": [0.002266817329697246, -0.0006020252693452743, 0.00024163562446222198, 0.0009758295609955114, 0.00042324930377346526, 0.00028738300959174623, 0.00022779102106773585, -0.0002037994412723565]}}, "s4_arith": {"alone": 1.1550061816504822, "alone_minus_off": {"mean": -0.07886868544375247, "se": 0.023194415891436188, "per_batch": [-0.07496078816169094, 0.045132117196045485, -0.145121196572499, -0.12688088860464064, -0.013288755466867475, -0.09220961614732559, -0.08542928612838474, -0.13819106966465688]}}, "s5_causal": {"alone": 0.6593072245240404, "alone_minus_off": {"mean": -0.0011304580874947762, "se": 0.000321404023591384, "per_batch": [-0.0015519714413860797, -0.0018272296740919902, -0.0005069188634181376, -0.0016303525345167103, -0.00010731717413836606, -9.652526279668727e-05, -0.0007122661485483572, -0.002611083601061881]}}, "s6_tryfail": {"alone": 0.6594088393984112, "alone_minus_off": {"mean": -0.013412421741753555, "se": 0.0007618931787584773, "per_batch": [-0.013913052489499478, -0.015548650494867533, -0.01419347020173134, -0.016397126902899184, -0.013722925668970709, -0.01069121549956098, -0.01025343366178988, -0.012579499014709339]}}, "web": {"alone": 0.9506567613584259, "alone_minus_off": {"mean": 6.000044731975518e-05, "se": 9.631581610844272e-05, "per_batch": [-0.0005266108769419464, -4.480148054974986e-05, 0.00030853687564791343, 0.00030509722263061256, -7.223271336176396e-06, 0.0002469670866389606, 9.467644929994634e-05, 0.00010336157316848116]}}}, "s5_causal": {"s1_perspective": {"alone": 0.5516690125210654, "alone_minus_off": {"mean": -0.0021642887974937935, "se": 0.0003518279348647084, "per_batch": [-0.0022480712207532116, -0.002950921319166744, -0.0034509608765503907, -0.002130435087563054, -0.0009878683465559535, -0.0030761676846575714, -0.000640592378801208, -0.0018292934659022153]}}, "s2_concept": {"alone": 0.7201691168186011, "alone_minus_off": {"mean": -0.0010148158781630306, "se": 0.0002551944996123987, "per_batch": [-0.0015721364071996646, -0.0003884228169737103, -0.0008650297381771477, -0.0023907308296438057, -0.0003562620612622691, -0.00037191248249079933, -0.0007946458383116806, -0.001379386851245168]}}, "s3_rules": {"alone": 0.9277824453565631, "alone_minus_off": {"mean": -0.0018306908249405285, "se": 0.00022613039944255578, "per_batch": [-0.0009550196602411409, -0.00278887066639133, -0.002538291944084281, -0.002192434833199086, -0.0011808328808240454, -0.0015964289566340328, -0.0015663319927330388, -0.0018273156654172729]}}, "s4_arith": {"alone": 1.2316764666128335, "alone_minus_off": {"mean": -0.002198400481400936, "se": 0.00021403025089921716, "per_batch": [-0.003272657863268469, -0.002493060506907341, -0.0025671850294290977, -0.0021541686933821413, -0.001368895909542589, -0.0017796764711284219, -0.001641144445858389, -0.0023104149316910405]}}, "s5_causal": {"alone": 0.1417724646021276, "alone_minus_off": {"mean": -0.5186652180094076, "se": 0.007470902026256088, "per_batch": [-0.49118512734418557, -0.5035505832009803, -0.5335864502652528, -0.5333256923197978, -0.5513792916223135, -0.5048105495990326, -0.5011616904359175, -0.5303223592877806]}}, "s6_tryfail": {"alone": 0.6705546587887811, "alone_minus_off": {"mean": -0.0022666023513836925, "se": 0.0001809747475090292, "per_batch": [-0.0021789341951063967, -0.0022350435349504094, -0.002426288242709851, -0.0024120136826881744, -0.002950104401575171, -0.002410852799794805, -0.0023831205973432468, -0.0011364613569014859]}}, "web": {"alone": 0.9506011034730404, "alone_minus_off": {"mean": 4.342561934281286e-06, "se": 0.00010631213097956605, "per_batch": [-0.00025565221050771747, -0.00029426231562623695, 0.0002472250606152526, 0.0005403694890109279, 9.476244062556205e-05, -0.000294778263578821, -0.00012821306621824125, 0.0001252893611535244]}}}, "s6_tryfail": {"s1_perspective": {"alone": 0.5567579630367223, "alone_minus_off": {"mean": 0.0029246617181631768, "se": 0.0002543385910052468, "per_batch": [0.003838050827979944, 0.001479093793082864, 0.002770253544435386, 0.003165297693467295, 0.00271792782291036, 0.003174197795649425, 0.003590223828086603, 0.0026622484396935375]}}, "s2_concept": {"alone": 0.7224646821232816, "alone_minus_off": {"mean": 0.0012807494265175134, "se": 0.0003111885603827908, "per_batch": [0.00035630505692507697, 0.001014955614066837, 0.0019954717022985236, 1.2038785560442022e-05, 0.0021480633092765267, 0.0013731954758141596, 0.0024819246302638653, 0.0008640408379346765]}}, "s3_rules": {"alone": 0.9293230229473293, "alone_minus_off": {"mean": -0.00029011323417417467, "se": 0.0002934363752189996, "per_batch": [0.0003786198058745027, -0.001287204150382637, -0.0012136815671388579, 0.0005216233800668935, -0.0010353355581940171, 0.00047802577807321, 0.0004824973269955679, -0.0006454508886880594]}}, "s4_arith": {"alone": 1.2328923409587709, "alone_minus_off": {"mean": -0.0009825261354635961, "se": 0.0004543478028467116, "per_batch": [-5.899004924603268e-05, -0.001759210535675626, -0.002379895922639541, -0.0005530962051749633, -0.0010519318840023217, 0.0014321855250598592, -0.0010483202483342335, -0.00244094976369591]}}, "s5_causal": {"alone": 0.6578710404023479, "alone_minus_off": {"mean": -0.002566642209187356, "se": 0.0003369519063570487, "per_batch": [-0.0031445307883756213, -0.002387979107230098, -0.001532795375814744, -0.0035148954270086685, -0.002257960223177813, -0.003644914311060843, -0.0009811180275096643, -0.003068944413321395]}}, "s6_tryfail": {"alone": 0.14373820614727262, "alone_minus_off": {"mean": -0.5290830549928922, "se": 0.005677738773715752, "per_batch": [-0.5331285894529934, -0.5157443896232985, -0.5363273162742221, -0.546233312734522, -0.5415691432431227, -0.5325711076902215, -0.5310587997476909, -0.49603178117706603]}}, "web": {"alone": 0.9507048842539205, "alone_minus_off": {"mean": 0.00010812334281433411, "se": 0.00015651834464667334, "per_batch": [-0.00030096963900994034, -1.909007424583109e-05, 0.0009224289479027314, 0.00023716407553986407, 0.0005877507093237089, -0.00021110870393403758, -8.710921266197325e-05, -0.00026407936039984925]}}}}, "battery": {"s1_perspective": {"acc_off": {"in": 0.55, "transfer": 0.4625}, "acc_all_on": {"in": 1.0, "transfer": 0.475}, "acc_member_masked": {"in": 0.55, "transfer": 0.45}, "acc_member_alone": {"in": 0.925, "transfer": 0.4625}, "margin_all_on_minus_off": {"in": {"mean": 3.1756021139852235, "se": 0.16199088616819113, "per_batch": null}, "transfer": {"mean": -0.2191101521253586, "se": 0.12726630613605572, "per_batch": null}}}, "s2_concept": {"acc_off": {"in": 0.2976190476190476, "transfer": 0.6779661016949152}, "acc_all_on": {"in": 0.9642857142857143, "transfer": 0.7372881355932204}, "acc_member_masked": {"in": 0.9166666666666666, "transfer": 0.7372881355932204}, "acc_member_alone": {"in": 0.9047619047619048, "transfer": 0.7033898305084746}, "margin_all_on_minus_off": {"in": {"mean": 4.466500394578491, "se": 0.3465357498549737, "per_batch": null}, "transfer": {"mean": 0.11481928572816363, "se": 0.1925147665554178, "per_batch": null}}}, "s3_rules": {"acc_off": {"in": 0.155, "transfer": 0.07, "held-out": 0.05}, "acc_all_on": {"in": 0.89, "transfer": 0.33, "held-out": 0.4666666666666667}, "acc_member_masked": {"in": 0.825, "transfer": 0.17, "held-out": 0.43333333333333335}, "acc_member_alone": {"in": 0.335, "transfer": 0.1, "held-out": 0.14166666666666666}, "margin_all_on_minus_off": {"in": {"mean": 3.0284131610766054, "se": 0.13749497955835932, "per_batch": null}, "transfer": {"mean": 1.0794337111711503, "se": 0.15529813607299472, "per_batch": null}, "held-out": {"mean": 1.6932759422808885, "se": 0.210794614852458, "per_batch": null}}}, "s4_arith": {"acc_off": {"in": 0.34375, "transfer": 0.24166666666666667, "held-out": 0.35}, "acc_all_on": {"in": 0.61875, "transfer": 0.2833333333333333, "held-out": 0.325}, "acc_member_masked": {"in": 0.48125, "transfer": 0.2625, "held-out": 0.325}, "acc_member_alone": {"in": 0.4125, "transfer": 0.2375, "held-out": 0.4}, "margin_all_on_minus_off": {"in": {"mean": 1.601310035516508, "se": 0.22152825532701553, "per_batch": null}, "transfer": {"mean": -0.2971320405602455, "se": 0.07265048400038335, "per_batch": null}, "held-out": {"mean": -0.1357417047023773, "se": 0.10575029142048824, "per_batch": null}}}}, "probes_all_on": {"p0_grounding": {"acc": 0.3333333333333333, "n": 30, "families": {"count": 0.125, "permanence": 0.5714285714285714, "sequence": 0.25, "attribute": 0.42857142857142855}}, "p1_self_other": {"acc": 0.5, "n": 30, "families": {"belief": 0.125, "reality": 0.6666666666666666, "perspective": 0.8, "reference": 0.8571428571428571, "second_order": 0.0}}, "p2_difference": {"acc": 0.5666666666666667, "n": 30, "families": {"category": 1.0, "antonym": 0.5, "odd_one_out": 0.5, "property": 0.5, "analogy": 0.3333333333333333}}, "p3_logic": {"acc": 0.23333333333333334, "n": 30, "families": {"d1": 0.25, "d2": 0.25, "d3": 0.5, "d4": 0.5, "d5": 0.0, "d6": 0.0, "negation": 0.25, "contrapositive": 0.0}}, "p4_arithmetic": {"acc": 0.4, "n": 30, "families": {"add2": 0.4, "add3": 0.0, "sub2": 0.75, "sub3": 0.4, "mul1": 0.5, "mul2": 0.3333333333333333, "word": 0.4}}, "p5_causal": {"acc": 0.5333333333333333, "n": 30, "families": {"effect": 0.7142857142857143, "cause": 0.42857142857142855, "counterfactual": 0.5, "prevention": 0.4, "chain": 0.6}}, "p6_learning": {"acc": 0.36666666666666664, "n": 30, "families": {"error_detect": 0.25, "correction": 0.25, "strategy": 0.7142857142857143, "check": 0.2857142857142857}}, "p7_composite": {"acc": 0.36666666666666664, "n": 30, "families": {"logic_arith": 0.25, "causal_category": 0.375, "belief_causal": 0.42857142857142855, "three_step": 0.42857142857142855}}, "p8_articulation": {"acc": 0.5333333333333333, "n": 30, "families": {"definition": 0.5, "register": 0.5714285714285714, "grammar": 0.875, "paraphrase": 0.14285714285714285}}}, "probes_mean_off": 0.41111111111111115, "probes_mean_all_on": 0.4259259259259259, "gates": {"s1_perspective": {"mean": 0.21163099743248942, "max": 0.554569513244894}, "s2_concept": {"mean": 0.27282737975382454, "max": 0.48724048599867004}, "s3_rules": {"mean": 0.12374002504886239, "max": 0.17456351278495152}, "s4_arith": {"mean": 0.10761241921992425, "max": 0.16724209369972712}, "s5_causal": {"mean": 0.18640250725757604, "max": 0.3602895342887907}, "s6_tryfail": {"mean": 0.2293551327826076, "max": 0.580794738809484}}, "chain_proof": "compiled chain: 18 cache entries (limit 256), 0 limit warnings", "curve": [{"step": 1, "train_bpb": {"s6_tryfail": 0.49155728705173696}, "quiet_kl": {"s6_tryfail": 0.0001717384730000049}}, {"step": 50, "train_bpb": {"s6_tryfail": 0.48510362259967144}, "quiet_kl": {"s6_tryfail": 0.0001995968290430028}}, {"step": 100, "train_bpb": {"s6_tryfail": 0.45611882148675426}, "quiet_kl": {"s6_tryfail": 0.0004217393323779106}}, {"step": 150, "train_bpb": {"s6_tryfail": 0.41484230852794907}, "quiet_kl": {"s6_tryfail": 0.000516712284879759}}, {"step": 200, "train_bpb": {"s6_tryfail": 0.3715938519533967}, "quiet_kl": {"s6_tryfail": 0.0006918617122573779}}, {"step": 250, "train_bpb": {"s6_tryfail": 0.32760316813268914}, "quiet_kl": {"s6_tryfail": 0.0007880888570798561}}, {"step": 300, "train_bpb": {"s6_tryfail": 0.28631622829571873}, "quiet_kl": {"s6_tryfail": 0.0008126485056709498}}, {"step": 350, "train_bpb": {"s6_tryfail": 0.2533118035213901}, "quiet_kl": {"s6_tryfail": 0.0008763719414127991}}, {"step": 400, "train_bpb": {"s6_tryfail": 0.22075288733537943}, "quiet_kl": {"s6_tryfail": 0.0007161688123596832}}, {"step": 450, "train_bpb": {"s6_tryfail": 0.1976540170259379}, "quiet_kl": {"s6_tryfail": 0.0006395705335307866}}, {"step": 500, "train_bpb": {"s6_tryfail": 0.17285999821341896}, "quiet_kl": {"s6_tryfail": 0.000667454925714992}}, {"step": 550, "train_bpb": {"s6_tryfail": 0.1564732371348439}, "quiet_kl": {"s6_tryfail": 0.0007572354574222117}}, {"step": 600, "train_bpb": {"s6_tryfail": 0.13781563003297514}, "quiet_kl": {"s6_tryfail": 0.0007101643219357357}}, {"step": 650, "train_bpb": {"s6_tryfail": 0.12557150320684107}, "quiet_kl": {"s6_tryfail": 0.0006701899075414985}}, {"step": 700, "train_bpb": {"s6_tryfail": 0.11602510634612184}, "quiet_kl": {"s6_tryfail": 0.0007344852236565203}}, {"step": 750, "train_bpb": {"s6_tryfail": 0.1070316603652374}, "quiet_kl": {"s6_tryfail": 0.0007432269875425845}}, {"step": 800, "train_bpb": {"s6_tryfail": 0.10001562747835419}, "quiet_kl": {"s6_tryfail": 0.0008044227393111214}}, {"step": 850, "train_bpb": {"s6_tryfail": 0.09334157929799357}, "quiet_kl": {"s6_tryfail": 0.0007597845961572603}}, {"step": 900, "train_bpb": {"s6_tryfail": 0.08909518909466772}, "quiet_kl": {"s6_tryfail": 0.0007287623034790158}}, {"step": 950, "train_bpb": {"s6_tryfail": 0.08759682701031712}, "quiet_kl": {"s6_tryfail": 0.0005959654896287248}}, {"step": 1000, "train_bpb": {"s6_tryfail": 0.08547882229131266}, "quiet_kl": {"s6_tryfail": 0.0005987927364185453}}, {"step": 1050, "train_bpb": {"s6_tryfail": 0.08520182832370292}, "quiet_kl": {"s6_tryfail": 0.0006423041893867775}}, {"step": 1100, "train_bpb": {"s6_tryfail": 0.08115986519083598}, "quiet_kl": {"s6_tryfail": 0.0005896395310992375}}, {"step": 1150, "train_bpb": {"s6_tryfail": 0.07941949954663946}, "quiet_kl": {"s6_tryfail": 0.0005292985116830095}}, {"step": 1200, "train_bpb": {"s6_tryfail": 0.0786209565814593}, "quiet_kl": {"s6_tryfail": 0.0005952842661645264}}, {"step": 1250, "train_bpb": {"s6_tryfail": 0.07424929615004981}, "quiet_kl": {"s6_tryfail": 0.0005008310498669743}}, {"step": 1300, "train_bpb": {"s6_tryfail": 0.07499035541902727}, "quiet_kl": {"s6_tryfail": 0.0005419513088418171}}, {"step": 1350, "train_bpb": {"s6_tryfail": 0.0780668795377288}, "quiet_kl": {"s6_tryfail": 0.0006149668578291312}}, {"step": 1400, "train_bpb": {"s6_tryfail": 0.07482796026350427}, "quiet_kl": {"s6_tryfail": 0.0005366446240805089}}, {"step": 1450, "train_bpb": {"s6_tryfail": 0.07181429773849791}, "quiet_kl": {"s6_tryfail": 0.0005191713041858748}}, {"step": 1500, "train_bpb": {"s6_tryfail": 0.07446593291382841}, "quiet_kl": {"s6_tryfail": 0.0005060935512301513}}, {"step": 1550, "train_bpb": {"s6_tryfail": 0.07283287111511484}, "quiet_kl": {"s6_tryfail": 0.0005083268188172951}}, {"step": 1600, "train_bpb": {"s6_tryfail": 0.07208941688023496}, "quiet_kl": {"s6_tryfail": 0.000502754524932243}}, {"step": 1650, "train_bpb": {"s6_tryfail": 0.07338609018039585}, "quiet_kl": {"s6_tryfail": 0.0004483193124178797}}, {"step": 1700, "train_bpb": {"s6_tryfail": 0.0684141532207329}, "quiet_kl": {"s6_tryfail": 0.0004800604720367119}}, {"step": 1750, "train_bpb": {"s6_tryfail": 0.07025520965501922}, "quiet_kl": {"s6_tryfail": 0.0004987858620006591}}, {"step": 1800, "train_bpb": {"s6_tryfail": 0.06809557481554691}, "quiet_kl": {"s6_tryfail": 0.0005053052725270391}}, {"step": 1850, "train_bpb": {"s6_tryfail": 0.07023307323058088}, "quiet_kl": {"s6_tryfail": 0.0004826495741144754}}, {"step": 1900, "train_bpb": {"s6_tryfail": 0.06761255036172642}, "quiet_kl": {"s6_tryfail": 0.000460979706258513}}, {"step": 1950, "train_bpb": {"s6_tryfail": 0.06822432178040691}, "quiet_kl": {"s6_tryfail": 0.0004944313853047788}}, {"step": 2000, "train_bpb": {"s6_tryfail": 0.06882643443843664}, "quiet_kl": {"s6_tryfail": 0.0004895336634945125}}, {"step": 2050, "train_bpb": {"s6_tryfail": 0.06769086470409076}, "quiet_kl": {"s6_tryfail": 0.00045109153434168546}}, {"step": 2100, "train_bpb": {"s6_tryfail": 0.06456515945914608}, "quiet_kl": {"s6_tryfail": 0.0005105206291773356}}, {"step": 2150, "train_bpb": {"s6_tryfail": 0.06581880495733339}, "quiet_kl": {"s6_tryfail": 0.0004580349934985861}}, {"step": 2200, "train_bpb": {"s6_tryfail": 0.06629729068150846}, "quiet_kl": {"s6_tryfail": 0.0005081880319630727}}, {"step": 2250, "train_bpb": {"s6_tryfail": 0.06377733461400144}, "quiet_kl": {"s6_tryfail": 0.0004161071893759072}}, {"step": 2300, "train_bpb": {"s6_tryfail": 0.06286386225383281}, "quiet_kl": {"s6_tryfail": 0.00047481127781793473}}, {"step": 2350, "train_bpb": {"s6_tryfail": 0.0646472039977185}, "quiet_kl": {"s6_tryfail": 0.0004391746738110669}}, {"step": 2400, "train_bpb": {"s6_tryfail": 0.0642197278978327}, "quiet_kl": {"s6_tryfail": 0.00045204543333966286}}, {"step": 2450, "train_bpb": {"s6_tryfail": 0.06074159709460985}, "quiet_kl": {"s6_tryfail": 0.00043931029678788034}}, {"step": 2500, "train_bpb": {"s6_tryfail": 0.06406820451809485}, "quiet_kl": {"s6_tryfail": 0.0004298066342016682}}, {"step": 2550, "train_bpb": {"s6_tryfail": 0.06329016973160599}, "quiet_kl": {"s6_tryfail": 0.0003949411190114915}}, {"step": 2600, "train_bpb": {"s6_tryfail": 0.06448494954550174}, "quiet_kl": {"s6_tryfail": 0.0003970019350526854}}, {"step": 2650, "train_bpb": {"s6_tryfail": 0.06380762435970513}, "quiet_kl": {"s6_tryfail": 0.0004128340134047903}}, {"step": 2700, "train_bpb": {"s6_tryfail": 0.06144096959633383}, "quiet_kl": {"s6_tryfail": 0.0004097355276462622}}, {"step": 2750, "train_bpb": {"s6_tryfail": 0.0629059472146676}, "quiet_kl": {"s6_tryfail": 0.00043815117183839905}}, {"step": 2800, "train_bpb": {"s6_tryfail": 0.062356816714443396}, "quiet_kl": {"s6_tryfail": 0.00042133788228966297}}, {"step": 2850, "train_bpb": {"s6_tryfail": 0.06387039609246524}, "quiet_kl": {"s6_tryfail": 0.0004090366180753335}}, {"step": 2900, "train_bpb": {"s6_tryfail": 0.06128162423065642}, "quiet_kl": {"s6_tryfail": 0.00042521887284237893}}, {"step": 2950, "train_bpb": {"s6_tryfail": 0.05986025877909709}, "quiet_kl": {"s6_tryfail": 0.00039715677732601764}}, {"step": 3000, "train_bpb": {"s6_tryfail": 0.06375431941196176}, "quiet_kl": {"s6_tryfail": 0.00040114474220899867}}, {"step": 3050, "train_bpb": {"s6_tryfail": 0.06001530328863305}, "quiet_kl": {"s6_tryfail": 0.0004056774140917696}}, {"step": 3100, "train_bpb": {"s6_tryfail": 0.06091127883258242}, "quiet_kl": {"s6_tryfail": 0.00040533632651204246}}, {"step": 3150, "train_bpb": {"s6_tryfail": 0.06040031036892683}, "quiet_kl": {"s6_tryfail": 0.00041882034478476273}}, {"step": 3200, "train_bpb": {"s6_tryfail": 0.06195103316252298}, "quiet_kl": {"s6_tryfail": 0.0003920030963490717}}, {"step": 3250, "train_bpb": {"s6_tryfail": 0.05953841797051342}, "quiet_kl": {"s6_tryfail": 0.0004072490497492254}}, {"step": 3300, "train_bpb": {"s6_tryfail": 0.0594973914554055}, "quiet_kl": {"s6_tryfail": 0.00044139614416053517}}, {"step": 3350, "train_bpb": {"s6_tryfail": 0.060310055163623914}, "quiet_kl": {"s6_tryfail": 0.0004448577418224886}}, {"step": 3400, "train_bpb": {"s6_tryfail": 0.06219102087072247}, "quiet_kl": {"s6_tryfail": 0.00045055443508317693}}, {"step": 3450, "train_bpb": {"s6_tryfail": 0.057973513433782535}, "quiet_kl": {"s6_tryfail": 0.0004168304131599143}}, {"step": 3500, "train_bpb": {"s6_tryfail": 0.06029894621301432}, "quiet_kl": {"s6_tryfail": 0.0004025866521988064}}, {"step": 3550, "train_bpb": {"s6_tryfail": 0.06101604394541574}, "quiet_kl": {"s6_tryfail": 0.0003981472211307846}}, {"step": 3600, "train_bpb": {"s6_tryfail": 0.0593074138298263}, "quiet_kl": {"s6_tryfail": 0.0003823853383073583}}, {"step": 3650, "train_bpb": {"s6_tryfail": 0.059133218084714355}, "quiet_kl": {"s6_tryfail": 0.0004304540890734643}}, {"step": 3700, "train_bpb": {"s6_tryfail": 0.05705731939441942}, "quiet_kl": {"s6_tryfail": 0.0004119642172008753}}, {"step": 3750, "train_bpb": {"s6_tryfail": 0.05784513332941467}, "quiet_kl": {"s6_tryfail": 0.0004145273711765185}}, {"step": 3800, "train_bpb": {"s6_tryfail": 0.05721199129279338}, "quiet_kl": {"s6_tryfail": 0.0004099449410568923}}, {"step": 3850, "train_bpb": {"s6_tryfail": 0.05927681365543779}, "quiet_kl": {"s6_tryfail": 0.0003988850995665416}}, {"step": 3900, "train_bpb": {"s6_tryfail": 0.0613913874280467}, "quiet_kl": {"s6_tryfail": 0.000396496316534467}}, {"step": 3950, "train_bpb": {"s6_tryfail": 0.05902321540148274}, "quiet_kl": {"s6_tryfail": 0.00036665005318354816}}, {"step": 4000, "train_bpb": {"s6_tryfail": 0.06069666888434422}, "quiet_kl": {"s6_tryfail": 0.00042347596638137475}}], "steps_done": {"s1_perspective": 3200, "s2_concept": 2400, "s3_rules": 800, "s4_arith": 800, "s5_causal": 4000, "s6_tryfail": 4000, "s7_mixed": 0, "s8_register": 0}, "stage_seconds": 2952.410352230072, "inert_at_birth": true}, {"close": 7, "attached": ["s1_perspective", "s2_concept", "s3_rules", "s4_arith", "s5_causal", "s6_tryfail", "s7_mixed"], "rows": {"s1_perspective": {"off": 0.5538333013185592, "all_on": 0.04823001515736244, "all_on_minus_off": {"mean": -0.5056032861611968, "se": 0.0047920862064853905, "per_batch": [-0.5076209933391915, -0.5249806587688494, -0.5160747360488588, -0.5110465061578721, -0.49242225767087794, -0.5081362748590078, -0.4815119362741176, -0.503032926170799]}}, "s2_concept": {"off": 0.7211839326967642, "all_on": 0.12115864825477166, "all_on_minus_off": {"mean": -0.6000252844419924, "se": 0.003772737406354231, "per_batch": [-0.59946057806427, -0.5975090801760141, -0.5873481086868944, -0.6225469916703095, -0.5980904782761713, -0.598507826425237, -0.5909210590074849, -0.6058181532295587]}}, "s3_rules": {"off": 0.9296131361815035, "all_on": 0.22174432066730942, "all_on_minus_off": {"mean": -0.7078688155141941, "se": 0.005518172474742527, "per_batch": [-0.6874448814495353, -0.7059788927894048, -0.7230923919266418, -0.6923164405200439, -0.6943369142002114, -0.7312169598425575, -0.7138556552032588, -0.7147083881818992]}}, "s4_arith": {"off": 1.2338748670942346, "all_on": 0.3264866914822855, "all_on_minus_off": {"mean": -0.9073881756119491, "se": 0.006258134187956579, "per_batch": [-0.9239890885199762, -0.8980481498705783, -0.922389413390807, -0.881079933077658, -0.8856563914171172, -0.9279706588701011, -0.9113652394791756, -0.9086065302701795]}}, "s5_causal": {"off": 0.6604376826115352, "all_on": 0.042644493625049976, "all_on_minus_off": {"mean": -0.6177931889864852, "se": 0.006963389516303299, "per_batch": [-0.5834391338260361, -0.6275946770986751, -0.6286100841672004, -0.62923814331032, -0.643703700814632, -0.6156325098092378, -0.5951296464567487, -0.6189976164090317]}}, "s6_tryfail": {"off": 0.6728212611401647, "all_on": 0.054212520346179585, "all_on_minus_off": {"mean": -0.6186087407939851, "se": 0.0035519632278359077, "per_batch": [-0.6278146805303335, -0.6248261487552063, -0.6207270820077949, -0.6275969074986785, -0.6233529561168311, -0.6176783133092029, -0.6060920248521421, -0.6007818132816922]}}, "s7_mixed": {"off": 0.7829759286650222, "all_on": 0.7195827258471108, "all_on_minus_off": {"mean": -0.06339320281791139, "se": 0.03903838944674557, "per_batch": [0.00044130748211390625, 0.001293481517139261, -0.034716159929190304, 0.0014897137217736445, -0.0001588259780718193, -0.24243706361526685, 0.006658351323816369, -0.23971642706560536]}}, "s8_register": {"off": 0.4906504977761471, "all_on": 0.37811960239612546, "all_on_minus_off": {"mean": -0.11253089538002163, "se": 0.002179814290544167, "per_batch": [-0.10146780770640357, -0.11533001782474173, -0.10767193135145986, -0.11254776849040865, -0.11233463899032686, -0.12247280128037585, -0.11617655942795113, -0.1122456379685054]}}, "web": {"off": 0.9505967609111062, "all_on": 0.9511670123857046, "all_on_minus_off": {"mean": 0.0005702514745983822, "se": 0.00032413473656832907, "per_batch": [-0.00020087573620775068, -0.0007618831433222617, 0.0006244690052829016, -0.00030337739612196213, 0.001665737964931857, 0.0016881817008693734, 0.0007719441283977613, 0.0010778152729571389]}}}, "member_in_group": {"s1_perspective": {"s1_perspective": {"masked": 0.49214595141889333, "all_on_minus_masked": {"mean": -0.44391593626153086, "se": 0.006110209923085749, "per_batch": [-0.4495981745217006, -0.4679567581832323, -0.46340019071223687, -0.4426834884312311, -0.4222853248919125, -0.44515618511738464, -0.41895535381024407, -0.441292014424305]}}, "s2_concept": {"masked": 0.35161480518972954, "all_on_minus_masked": {"mean": -0.23045615693495786, "se": 0.001389582139390106, "per_batch": [-0.2278770983845469, -0.2297601256799281, -0.22404744245971225, -0.23670378547261015, -0.22917717973591295, -0.2305435926459336, -0.23070352576232034, -0.23483650533869857]}}, "s3_rules": {"masked": 0.41703416786420267, "all_on_minus_masked": {"mean": -0.19528984719689324, "se": 0.0030430465816673907, "per_batch": [-0.1911247928561291, -0.20249757560105947, -0.19926514018026165, -0.18602355095539325, -0.19727652628616635, -0.2104635970107112, -0.18888884641226175, -0.1867787482731632]}}, "s4_arith": {"masked": 0.4610068425216251, "all_on_minus_masked": {"mean": -0.13452015103933967, "se": 0.0045753771536862255, "per_batch": [-0.14578259885205475, -0.1486169374317679, -0.10894707521845082, -0.13664554757267822, -0.14573500265342848, -0.12970927228506757, -0.12882044444757723, -0.13190432985369233]}}, "s5_causal": {"masked": 0.07341230071259938, "all_on_minus_masked": {"mean": -0.030767807087549403, "se": 0.001940839360805322, "per_batch": [-0.029438122550596378, -0.04257451482494402, -0.02906542271003206, -0.030363262912479053, -0.028615553716579825, -0.029790254341009043, -0.033015810455772385, -0.02327951518898247]}}, "s6_tryfail": {"masked": 0.07537674432617615, "all_on_minus_masked": {"mean": -0.021164223979996566, "se": 0.0019656207190994245, "per_batch": [-0.02959274301540879, -0.02053629785470848, -0.016652048187140576, -0.017648999367445253, -0.025169230997144888, -0.02217223982537822, -0.012301790029228557, -0.025240442563517776]}}, "s7_mixed": {"masked": 0.7238786966063865, "all_on_minus_masked": {"mean": -0.0042959707592756385, "se": 0.004688235578545134, "per_batch": [4.531742850222287e-05, 0.00030423730937623183, -0.0005805274379874215, 0.0013961551597042599, 0.00012365552597037865, -0.03652227873288877, 0.0050072318842079855, -0.004141557211089997]}}, "web": {"masked": 0.9511541674314682, "all_on_minus_masked": {"mean": 1.2844954236354567e-05, "se": 0.0004084970607729815, "per_batch": [-9.751416303904747e-05, -0.0011815208114275277, -0.0008667925603484949, -0.0012426606438091792, 0.001722578231041938, 0.0009087563271591437, 0.0012908157860508362, -0.000430902531736832]}}}, "s2_concept": {"s1_perspective": {"masked": 0.05230108568208067, "all_on_minus_masked": {"mean": -0.004071070524718228, "se": 0.0003063687899632255, "per_batch": [-0.00399180331785122, -0.004028091657183271, -0.005059116900190133, -0.0030723141983867303, -0.005527339667164129, -0.0037292073078150703, -0.0030534820981172606, -0.004107209051038008]}}, "s2_concept": {"masked": 0.24197061173215093, "all_on_minus_masked": {"mean": -0.12081196347737927, "se": 0.00132646793136312, "per_batch": [-0.12437578931217287, -0.12386146444585194, -0.11621068723523176, -0.12588156191397082, -0.12199913956206834, -0.11975056314227824, -0.11644975386884675, -0.11796674833861351]}}, "s3_rules": {"masked": 0.32263836737814544, "all_on_minus_masked": {"mean": -0.10089404671083599, "se": 0.0011653913831251598, "per_batch": [-0.10139157639640861, -0.10246640347080704, -0.10342368040334082, -0.10104834202094914, -0.09466604434959894, -0.10096729519673003, -0.10527814782976311, -0.09791088401909026]}}, "s4_arith": {"masked": 0.425437358423432, "all_on_minus_masked": {"mean": -0.09895066694114654, "se": 0.003002979611944084, "per_batch": [-0.10809793237764703, -0.0972569414870158, -0.08421390623256858, -0.104415826818337, -0.09092628160658717, -0.10880877966932573, -0.09672392425632925, -0.10116174308136183]}}, "s5_causal": {"masked": 0.04352168576131789, "all_on_minus_masked": {"mean": -0.0008771921362679151, "se": 0.00020703456012069699, "per_batch": [-0.00036522128248069746, -0.0008533510412920622, -0.001553933118497386, -0.0007830961284145954, -0.000795731478795178, -0.0017243088064661932, -0.0010261586714332277, 8.426343723601931e-05]}}, "s6_tryfail": {"masked": 0.05657776616855598, "all_on_minus_masked": {"mean": -0.0023652458223764024, "se": 0.0004964502865756417, "per_batch": [-0.002625347412167814, -0.005613132137655477, -0.0022611741489658096, -0.0011231971949536498, -0.001725845901408285, -0.001311055994274224, -0.0021642511762889477, -0.0020979626132970122]}}, "s7_mixed": {"masked": 0.719770297112937, "all_on_minus_masked": {"mean": -0.00018757126582615719, "se": 0.00040235207745951734, "per_batch": [-0.00013818805996823613, -0.0015896356419249358, 0.0009257826095945276, -0.0013577170372364167, -0.0007118361919211669, -0.0007412882208814309, 0.0004183907938865028, 0.0016939216218418984]}}, "web": {"masked": 0.9522010473251015, "all_on_minus_masked": {"mean": -0.0010340349393969522, "se": 0.00017528399288518331, "per_batch": [-0.0014137833814176215, -0.0006430431315760377, -0.0009164095551223994, -0.0008719520398744462, -0.002027589462347157, -0.0006321222332462684, -0.0011914958051775226, -0.0005758839064141652]}}}, "s3_rules": {"s1_perspective": {"masked": 0.04816853874955851, "all_on_minus_masked": {"mean": 6.147640780392472e-05, "se": 6.568223455813486e-05, "per_batch": [0.00026870139414178484, 7.363007240093755e-07, -0.000326251088686752, 0.00015516059782529068, 0.00023356856324521086, 6.154829117752808e-05, -5.293840971871966e-06, 0.00010364104497619792]}}, "s2_concept": {"masked": 0.12169112535244535, "all_on_minus_masked": {"mean": -0.0005324770976736955, "se": 9.779373598288399e-05, "per_batch": [-0.0007464906960700235, -0.0006431506207328352, -0.00038099531623951577, -0.0010611759514862484, -0.0004910642127917142, -0.0003194040293992839, -0.00019079325330091657, -0.00042674270136902615]}}, "s3_rules": {"masked": 0.314198523016451, "all_on_minus_masked": {"mean": -0.09245420234914156, "se": 0.0015010694995817113, "per_batch": [-0.08961671971159235, -0.09024927190146576, -0.09382094310109046, -0.09836749795713101, -0.08617113329288137, -0.09831745100572997, -0.09222530956421668, -0.09086529225902495]}}, "s4_arith": {"masked": 0.3268355152938979, "all_on_minus_masked": {"mean": -0.00034882381161248366, "se": 7.305144836396506e-05, "per_batch": [-0.0006097644886340903, -0.0003244022751899278, -0.0004224968796758821, 5.335761743019418e-05, -0.0002681209526950723, -0.00034198750124070365, -0.0005805489358187699, -0.0002966270770756174]}}, "s5_causal": {"masked": 0.04276538735149808, "all_on_minus_masked": {"mean": -0.00012089372644810605, "se": 9.158543502943797e-05, "per_batch": [-0.0001824547819630068, 2.0272454970451936e-05, 2.4037262686994743e-05, 1.4048832792358523e-05, -5.9801592379699064e-05, -0.00019495845812651097, -0.0007049622603444924, 0.00011666873077905565]}}, "s6_tryfail": {"masked": 0.05452802789364525, "all_on_minus_masked": {"mean": -0.0003155075474656627, "se": 9.632648826833291e-05, "per_batch": [-0.00037711495767942665, -0.0001510975076986637, -0.00017055304507750807, -0.00018423641473678393, -1.914919328200454e-05, -0.0006377761628933651, -0.0001786093573788658, -0.0008055237409786836]}}, "s7_mixed": {"masked": 0.7196093106028139, "all_on_minus_masked": {"mean": -2.6584755702920015e-05, "se": 0.00043437244823704317, "per_batch": [0.00045455014623030365, 0.00017860398292102175, 0.0014194588088961568, -0.0003159321296349882, -0.0004873988325452272, 0.0007415031991949705, 0.0004799605828954001, -0.0026834238035809976]}}, "web": {"masked": 0.9510226759459679, "all_on_minus_masked": {"mean": 0.00014433643973665866, "se": 0.000170180288972182, "per_batch": [0.0001737024773715845, -0.0002724205189666984, 0.00010250165991432247, -0.00035703598319114516, 0.0009343817421376688, 0.000766784648871921, -0.00027886986837399874, 8.564736012961482e-05]}}}, "s4_arith": {"s1_perspective": {"masked": 0.04817508148017083, "all_on_minus_masked": {"mean": 5.493367719160754e-05, "se": 5.330048502413832e-05, "per_batch": [0.00024256003121063408, 2.4593519073384806e-05, -3.742772439402353e-05, 0.0002849268823591211, -0.00013322743538244503, 0.00010605955100395636, 5.035329549792078e-05, -9.836870183568824e-05]}}, "s2_concept": {"masked": 0.12119183015747251, "all_on_minus_masked": {"mean": -3.318190270084828e-05, "se": 6.040791062361161e-05, "per_batch": [-0.00018072151930978442, 0.00015483813035492566, -0.00011052035101072333, -0.00024976180471551757, -1.4812005805567297e-05, 0.00018623571305305398, 0.00012586980260022307, -0.00017658318677339635]}}, "s3_rules": {"masked": 0.22168091012649055, "all_on_minus_masked": {"mean": 6.341054081889994e-05, "se": 7.997902077408083e-05, "per_batch": [0.0001163892589713944, -0.00010456545172457532, -6.518142467701327e-05, -6.685825552291136e-05, 0.0002016711559680029, -0.0002674760177543978, 0.00029854038406648686, 0.00039476467722421305]}}, "s4_arith": {"masked": 0.49155023576305157, "all_on_minus_masked": {"mean": -0.16506354428076614, "se": 0.00194527438542723, "per_batch": [-0.17117385945146357, -0.16206817350803804, -0.1664870527436444, -0.1608405828440107, -0.1548992701873041, -0.1706066176732553, -0.16611333444331955, -0.16831946339509357]}}, "s5_causal": {"masked": 0.04281735130482977, "all_on_minus_masked": {"mean": -0.00017285767977979604, "se": 4.8830753184044855e-05, "per_batch": [9.75195374969956e-06, -8.993617748557364e-05, -0.0002673658413686747, -0.00015107063540945737, -0.00023123336131396222, -0.00034069763135922276, -0.00031703389349206634, 4.724148440889142e-06]}}, "s6_tryfail": {"masked": 0.05800730007047948, "all_on_minus_masked": {"mean": -0.003794779724299896, "se": 0.0005036103301095958, "per_batch": [-0.004758668583590649, -0.0065604501998127035, -0.0023722964392545454, -0.0031507866573007304, -0.0038019452202143547, -0.00291783615670705, -0.0023727102725081897, -0.004423544265010945]}}, "s7_mixed": {"masked": 0.7193774215576434, "all_on_minus_masked": {"mean": 0.00020530428946746726, "se": 0.00023767780167092934, "per_batch": [0.0006743439740330981, 0.0004939341732779212, 5.3228630441992664e-05, -0.0004182618068984123, 0.0008433169285058106, -0.0010404520420572716, 0.0009012320861838674, 0.00013509237225273196]}}, "web": {"masked": 0.9513185935946087, "all_on_minus_masked": {"mean": -0.00015158120890419735, "se": 0.00018551673632480586, "per_batch": [-0.001049266152914008, -0.000442941317297163, 0.0002207397323824578, -0.0005690046003795635, 0.00019227660366460952, 0.00061415004623111, -0.00016475937952653563, -1.3844603394486121e-05]}}}, "s5_causal": {"s1_perspective": {"masked": 0.048184326219461934, "all_on_minus_masked": {"mean": 4.568893790050325e-05, "se": 6.685387011216933e-05, "per_batch": [0.00030808004673152367, -0.00029634223081013294, 0.00019487784125891972, -5.47657253840983e-06, 0.00019573238005539395, -5.254069983858661e-05, 6.035516153716364e-05, -3.917442319184561e-05]}}, "s2_concept": {"masked": 0.12100348362610298, "all_on_minus_masked": {"mean": 0.0001551646286686685, "se": 0.00010219309696323772, "per_batch": [2.240074027487493e-05, 0.0004617196729882478, -5.448625357648562e-05, 0.0003778136371985624, 0.00012806258139871907, 0.000349425750890503, 0.00035141430029111964, -0.00039503340011619315]}}, "s3_rules": {"masked": 0.2213956688383478, "all_on_minus_masked": {"mean": 0.0003486518289616339, "se": 0.0001025277746233953, "per_batch": [0.00028276097584981064, 0.0006059378746524069, 0.00015097926962617803, 6.331111334886841e-05, 0.0009309850847831203, 0.0001454113313044958, 0.00022308299600043435, 0.0003867459861277567]}}, "s4_arith": {"masked": 0.32639461358335087, "all_on_minus_masked": {"mean": 9.207789893461021e-05, "se": 9.049223993755447e-05, "per_batch": [0.00035570311764704377, 0.0004317409471597533, -0.00015001186721508297, 0.00028658758983152177, -9.050587001652177e-05, 0.00014145573033469416, -0.00028078317536489594, 4.243671910036939e-05]}}, "s5_causal": {"masked": 0.4702006856969913, "all_on_minus_masked": {"mean": -0.4275561920719413, "se": 0.00455600329715842, "per_batch": [-0.4042913847586438, -0.43220707926259766, -0.4262188368400369, -0.4353123120152511, -0.44186090616106993, -0.441639983697121, -0.4181398497010575, -0.42077918413975257]}}, "s6_tryfail": {"masked": 0.054607495970874005, "all_on_minus_masked": {"mean": -0.00039497562469441826, "se": 7.65733296252945e-05, "per_batch": [-0.0005966938071685149, -7.003993456387858e-05, -0.0006091518004403898, -0.00035825598512075063, -0.00038054923623884185, -0.0006030464163347543, -9.691222376119468e-05, -0.0004451555939270213]}}, "s7_mixed": {"masked": 0.7194638912098226, "all_on_minus_masked": {"mean": 0.00011883463728833837, "se": 0.0001439288230160655, "per_batch": [0.0003991717326525812, 0.00029907782985039155, 1.401658604527345e-05, 0.0003873909210684312, 0.00022116968900953715, -0.0008215611231716657, 0.0003809845673238277, 7.042689552833048e-05]}}, "web": {"masked": 0.9510530631305922, "all_on_minus_masked": {"mean": 0.00011394925511229448, "se": 0.00015371160870231824, "per_batch": [0.0003868749731159582, -0.00012872901417082527, -0.00023303649191908082, -0.0005288466514031809, 0.0006941219788823005, 3.491247812514864e-05, 0.0006658308328153506, 2.0465935452684825e-05]}}}, "s6_tryfail": {"s1_perspective": {"masked": 0.04807661973303379, "all_on_minus_masked": {"mean": 0.00015339542432864142, "se": 5.6940566374025465e-05, "per_batch": [8.540013506906641e-05, 0.0001850103366656669, 0.0002306233603491803, 0.00045905394189987847, 5.58137446628193e-05, -9.650913942323147e-05, 0.00020171683885963215, 0.00010605417654611926]}}, "s2_concept": {"masked": 0.1207410233211273, "all_on_minus_masked": {"mean": 0.000417624933644372, "se": 0.00011769370776477818, "per_batch": [-5.189576489786352e-05, 0.0010303588102347633, 0.0005468295873339718, 0.0003045812746780424, 0.00027093716860299155, 0.00037327759478204825, 0.0006848026689886683, 0.000182108129432354]}}, "s3_rules": {"masked": 0.22164392310763933, "all_on_minus_masked": {"mean": 0.00010039755967008318, "se": 8.976196790925413e-05, "per_batch": [-0.00032769144338773426, 0.00036782789453282394, -3.635283282613089e-05, 0.0003329799099017483, 0.00017675516942425307, -0.00014635723588424243, 9.70842064120514e-05, 0.0003389348091878963]}}, "s4_arith": {"masked": 0.3280204408243667, "all_on_minus_masked": {"mean": -0.001533749342081274, "se": 0.00022605757419774048, "per_batch": [-0.002113623783441154, -0.0015897431310816779, -0.0011313878687009837, -0.0006758703200594463, -0.001568524771531521, -0.002537389035167148, -0.0008118226055665589, -0.0018416332211017017]}}, "s5_causal": {"masked": 0.04258270113194436, "all_on_minus_masked": {"mean": 6.179249310560905e-05, "se": 8.03361176887931e-05, "per_batch": [-0.00014147991539494031, 2.642352196771669e-05, 2.416624967514769e-05, 0.00012117790090637043, 0.0002285756919123555, -8.418013313950257e-05, 0.0005102268422183184, -0.00019057021330059348]}}, "s6_tryfail": {"masked": 0.46879082566979635, "all_on_minus_masked": {"mean": -0.41457830532361684, "se": 0.0028625404566528026, "per_batch": [-0.4212061606300376, -0.40063545549624086, -0.42460486027855704, -0.41049452219068655, -0.4182792953841602, -0.4196223186536636, -0.41519381913852466, -0.4065900108170639]}}, "s7_mixed": {"masked": 0.7195978388225555, "all_on_minus_masked": {"mean": -1.5112975444590837e-05, "se": 0.00025422316493365845, "per_batch": [-0.0002751722413805169, -0.00023613217963469602, 0.0008993832726870155, -0.0006707323383650099, 0.0004167139630405492, -0.001058252246421587, 0.0009560085604836677, -0.00015272059396614912]}}, "web": {"masked": 0.9512506389496865, "all_on_minus_masked": {"mean": -8.362656398203105e-05, "se": 0.00018504924150159645, "per_batch": [-0.00020706711163875902, -0.0005203335101854334, 0.0009354136380428368, -0.000602111260670668, -0.00017705613906326967, -0.0003031194221456701, 0.00047553202963568353, -0.0002702707358309686]}}}, "s7_mixed": {"s1_perspective": {"masked": 0.048118676545145664, "all_on_minus_masked": {"mean": 0.00011133861221676936, "se": 0.00010091511854947971, "per_batch": [-2.847387763347392e-05, -6.942724637017811e-05, 0.0007419761514848716, 0.00020524517043124524, -0.000161964661450055, 0.00015712764939453117, 0.00012107041174957284, -7.484469987235898e-05]}}, "s2_concept": {"masked": 0.12151103130108834, "all_on_minus_masked": {"mean": -0.00035238304631667744, "se": 7.958744485601067e-05, "per_batch": [-0.0006138705744234418, -0.00037037538754873534, -0.00033208774990041434, -0.000538714155996392, 0.00010258765123966063, -0.000524504089468858, -0.00024837519459292023, -0.0002937248698423184]}}, "s3_rules": {"masked": 0.2224989079841146, "all_on_minus_masked": {"mean": -0.0007545873168051664, "se": 0.00028474767206220235, "per_batch": [0.0006465687759187322, -0.0007211662507304595, -0.0009068430201682465, -0.0014497922489421222, -0.001671198414096603, -0.0015366219897964828, -0.00020880843597878562, -0.00018883695064736417]}}, "s4_arith": {"masked": 0.3363804556765427, "all_on_minus_masked": {"mean": -0.009893764194257249, "se": 0.0005493844784905836, "per_batch": [-0.010207557289672098, -0.008811423627799186, -0.00990807100102592, -0.008115646315902336, -0.007928744170077162, -0.011740094691510605, -0.012137439108497794, -0.010301137349572886]}}, "s5_causal": {"masked": 0.04261427976669097, "all_on_minus_masked": {"mean": 3.0213858359000485e-05, "se": 9.476759315245345e-05, "per_batch": [0.00028889054501572003, -0.00019839542391485165, 7.626624397089748e-05, 3.2238685350023666e-05, -0.0003286319735096299, 0.0004718666493891549, -0.00020001044849560917, 9.948658906629854e-05]}}, "s6_tryfail": {"masked": 0.05458008825131443, "all_on_minus_masked": {"mean": -0.00036756790513484995, "se": 5.7556784355605655e-05, "per_batch": [-0.0002811432640401415, -5.13690680295914e-05, -0.0002875872389896533, -0.0004079965924249779, -0.0004023319138621792, -0.00041057633218791884, -0.0005668064471232651, -0.0005327323844210724]}}, "s7_mixed": {"masked": 0.7763963238974697, "all_on_minus_masked": {"mean": -0.05681359805035884, "se": 0.032952165005254254, "per_batch": [-0.000636937747470423, 0.00016587726675720837, -0.03688666697440435, -0.000519903553558243, 7.223271336176396e-06, -0.17760833235951923, -0.007363609179342034, -0.23166643512666985]}}, "web": {"masked": 0.9520687174241782, "all_on_minus_masked": {"mean": -0.0009017050384736991, "se": 0.000186768170323333, "per_batch": [-0.0012462722794772674, -0.0012433485744126616, -4.660729838379396e-05, -0.0012702638592727578, -0.00143880685711828, -0.0011498760036686706, -0.0004981477482240981, -0.00032031768723206344]}}}}, "member_alone": {"s1_perspective": {"s1_perspective": {"alone": 0.05217516482213097, "alone_minus_off": {"mean": -0.5016581364964282, "se": 0.004888586924071733, "per_batch": [-0.503264758784162, -0.5212149750156778, -0.5118989327924157, -0.509032675307594, -0.4859917833594548, -0.5041102201213458, -0.4784889810965285, -0.4992627654942472]}}, "s2_concept": {"alone": 0.24394271260453448, "alone_minus_off": {"mean": -0.47724122009222963, "se": 0.0030459897360259366, "per_batch": [-0.47044852344580024, -0.4716034084441754, -0.4687537634085354, -0.4938985943670891, -0.47575242540274654, -0.47682052365593, -0.47438017583151226, -0.4862723461820483]}}, "s3_rules": {"alone": 0.5034592174177095, "alone_minus_off": {"mean": -0.42615391876379405, "se": 0.003854800114533744, "per_batch": [-0.4117213926773634, -0.4236871716024687, -0.442290362963648, -0.410341145912864, -0.4260619510412451, -0.4339105083003576, -0.4286801719223179, -0.4325386456900876]}}, "s4_arith": {"alone": 0.7531657780810914, "alone_minus_off": {"mean": -0.480709089013143, "se": 0.003514095864453405, "per_batch": [-0.4824103037742731, -0.4703906082881222, -0.490546115055989, -0.47029558787352066, -0.4725516132918761, -0.49690809727670615, -0.47641963209676885, -0.48615075444788836]}}, "s5_causal": {"alone": 0.468494725289589, "alone_minus_off": {"mean": -0.19194295732194622, "se": 0.002210700040436461, "per_batch": [-0.1875164678540393, -0.1992422879855283, -0.19983549914401677, -0.1947394812163009, -0.19148918109764468, -0.18138197568475095, -0.1878665385398705, -0.19347222705341832]}}, "s6_tryfail": {"alone": 0.5038253093623601, "alone_minus_off": {"mean": -0.16899595177780455, "se": 0.002363709528038538, "per_batch": [-0.1689674510278819, -0.1804920514578614, -0.16629625949034343, -0.17642276995579653, -0.17014793994340416, -0.16627970616019783, -0.16280797837855587, -0.16055345780839525]}}, "s7_mixed": {"alone": 0.7741873679808721, "alone_minus_off": {"mean": -0.00878856068415014, "se": 0.009343584315554394, "per_batch": [0.001439838753023448, 0.0025405277185455155, 0.00048086349181242216, 0.002514214372963397, 0.0011621727632054046, -0.07048446672069653, 0.011682351516483447, -0.01964398736853823]}}, "web": {"alone": 0.9519326898962615, "alone_minus_off": {"mean": 0.001335928985155277, "se": 0.0004171900864352844, "per_batch": [0.0008697162654132118, -0.0001570201602377752, 0.0007944738556607822, 0.00027680607656377365, 0.003323134771296754, 0.002440089850441529, 0.002121320007067329, 0.0010189112150366109]}}}, "s2_concept": {"s1_perspective": {"alone": 0.5237967463236832, "alone_minus_off": {"mean": -0.03003655499487594, "se": 0.0008929364404090443, "per_batch": [-0.02682404806374683, -0.03162864138957566, -0.02918588580804582, -0.032230279697956554, -0.03334249150108626, -0.031257674811664526, -0.026360984776298668, -0.02946243391063319]}}, "s2_concept": {"alone": 0.36033454325401637, "alone_minus_off": {"mean": -0.36084938944274775, "se": 0.0021975474547336524, "per_batch": [-0.36082329913716604, -0.3578469673813343, -0.3548674109507988, -0.37462086527661115, -0.36102178861409306, -0.36006399573360676, -0.3551193225386502, -0.36243146590972153]}}, "s3_rules": {"alone": 0.6157804256378956, "alone_minus_off": {"mean": -0.31383271054360795, "se": 0.0028684777591963, "per_batch": [-0.3051689393960323, -0.3130032220951231, -0.3208357849677054, -0.3010766122187518, -0.30994988510736754, -0.3192230176592393, -0.3165332949910704, -0.32487092791357386]}}, "s4_arith": {"alone": 0.8761517594682684, "alone_minus_off": {"mean": -0.35772310762596604, "se": 0.004197781721024297, "per_batch": [-0.3465006699646368, -0.3508452096994077, -0.37445335417467074, -0.3532194301945688, -0.34704009354906795, -0.37528652412677577, -0.36393756097898977, -0.35050201831961103]}}, "s5_causal": {"alone": 0.6477866226906935, "alone_minus_off": {"mean": -0.012651059920841742, "se": 0.0009352908521067744, "per_batch": [-0.007499647456174596, -0.011507402164893321, -0.015175620125146061, -0.01262455846923527, -0.013140119460859356, -0.011161158093043633, -0.015682152027599816, -0.014417821569781886]}}, "s6_tryfail": {"alone": 0.6666359373486308, "alone_minus_off": {"mean": -0.0061853237915339004, "se": 0.0012459523146970725, "per_batch": [-0.007500163404127069, -0.013446420562046035, -0.006764636602049623, -0.006561095134753447, -0.006009288799460166, -0.003128321423531699, -0.0018538439893129866, -0.004218820416990177]}}, "s7_mixed": {"alone": 0.7831651579510919, "alone_minus_off": {"mean": 0.0001892292860696254, "se": 0.0011789333136323083, "per_batch": [0.0005264388942911591, 0.0002920265411650025, 0.00241738814052761, 0.0004164559890642572, 0.00026923883992568953, 0.0038801005861157645, 0.001148199172822828, -0.007436013875355307]}}, "web": {"alone": 0.9514616616622953, "alone_minus_off": {"mean": 0.0008649007511890849, "se": 0.00018835522783446044, "per_batch": [5.219673453682461e-05, 0.0011502199689704673, 0.0014158471732280686, 0.00112304671013419, 0.0007925820465013445, 0.0004966858956918507, 0.00034379331907474775, 0.001544834161375186]}}}, "s3_rules": {"s1_perspective": {"alone": 0.5535631380718851, "alone_minus_off": {"mean": -0.00027016324667410163, "se": 0.00023913782920347164, "per_batch": [0.0002715606057124287, -0.0004794876306054574, -0.0007019041938338688, 0.0007255088126647546, -0.0014021745524844809, -0.0003248322318171182, 0.00029976576045387393, -0.000549742543482945]}}, "s2_concept": {"alone": 0.7227973879358336, "alone_minus_off": {"mean": 0.0016134552390694584, "se": 0.0002752200095664509, "per_batch": [0.000995564570182128, 0.0010287142261358184, 0.0016847850435147693, 0.0009734218038836007, 0.0022790280979085864, 0.0019874315133706633, 0.0009180433903056562, 0.003040653267254445]}}, "s3_rules": {"alone": 0.8701872076558641, "alone_minus_off": {"mean": -0.05942592852563948, "se": 0.010330913371419828, "per_batch": [-0.06883390622470331, -0.052931358430022946, -0.047376318807154205, -0.08934343927937138, -0.029561753891506393, -0.06865332444129746, -0.10310746880189825, -0.015599858329161886]}}, "s4_arith": {"alone": 1.2291967885059467, "alone_minus_off": {"mean": -0.0046780785882879206, "se": 0.0004241556239844928, "per_batch": [-0.0034991590144548557, -0.00447249482701273, -0.00568497251559541, -0.003813715282882768, -0.004839591795279485, -0.0037889497811585127, -0.00709909986231505, -0.004226645627604553]}}, "s5_causal": {"alone": 0.666408248442262, "alone_minus_off": {"mean": 0.005970565830726776, "se": 0.0006340753987558168, "per_batch": [0.006426389723465098, 0.007101077662799882, 0.001980165246371679, 0.007604900838502515, 0.006885497409943375, 0.006719362169209986, 0.004959076741966384, 0.00608805685355529]}}, "s6_tryfail": {"alone": 0.6717302999433321, "alone_minus_off": {"mean": -0.0010909611968325655, "se": 0.0003520287905236149, "per_batch": [-0.0010005950627198779, -0.0029078826607884523, -0.00020315450633168197, -0.0009048437218519556, -0.0018541019632892786, -0.0005386496625022774, 0.00022035277141796428, -0.001538814768594965]}}, "s7_mixed": {"alone": 0.7822497050494648, "alone_minus_off": {"mean": -0.0007262236155574156, "se": 0.0011695408156214792, "per_batch": [-0.00012589130043161312, 4.1361827532560014e-05, 0.0011479841945092328, 0.00033837586557261545, -0.0006783855663283767, 0.0020663715501165747, 3.9599005361157236e-05, -0.008639204500791475]}}, "web": {"alone": 0.9509403070051203, "alone_minus_off": {"mean": 0.0003435460940141577, "se": 0.0001181112928609609, "per_batch": [0.00013586629418149698, 0.00014833503636912937, 0.001022264876728518, 0.0005037371843772398, 0.0003148142424045375, 0.0005318563477932914, 4.4629497898962533e-05, 4.6865272360085974e-05]}}}, "s4_arith": {"s1_perspective": {"alone": 0.5525393683469615, "alone_minus_off": {"mean": -0.0012939329715976888, "se": 0.00033731606119881077, "per_batch": [-0.0005440671160045207, -0.002287799213102515, 5.077787766727404e-05, -0.002374908425764488, -0.002254219600521523, -0.0003551871696944042, -0.00143721601759772, -0.0011488441077636136]}}, "s2_concept": {"alone": 0.7197303675784188, "alone_minus_off": {"mean": -0.001453565118345393, "se": 0.00024222923191314095, "per_batch": [-0.0015742001990100007, -0.001012590852617512, -0.0006298004674595292, -0.0029371197114349545, -0.0012482930756249067, -0.0015436732784818163, -0.001588259780718082, -0.0010945835814163418]}}, "s3_rules": {"alone": 0.9300652463238748, "alone_minus_off": {"mean": 0.000452110142371287, "se": 0.0003056447991096855, "per_batch": [0.002266817329697246, -0.0006020252693452743, 0.00024163562446222198, 0.0009758295609955114, 0.00042324930377346526, 0.00028738300959174623, 0.00022779102106773585, -0.0002037994412723565]}}, "s4_arith": {"alone": 1.1550061816504822, "alone_minus_off": {"mean": -0.07886868544375247, "se": 0.023194415891436188, "per_batch": [-0.07496078816169094, 0.045132117196045485, -0.145121196572499, -0.12688088860464064, -0.013288755466867475, -0.09220961614732559, -0.08542928612838474, -0.13819106966465688]}}, "s5_causal": {"alone": 0.6593072245240404, "alone_minus_off": {"mean": -0.0011304580874947762, "se": 0.000321404023591384, "per_batch": [-0.0015519714413860797, -0.0018272296740919902, -0.0005069188634181376, -0.0016303525345167103, -0.00010731717413836606, -9.652526279668727e-05, -0.0007122661485483572, -0.002611083601061881]}}, "s6_tryfail": {"alone": 0.6594088393984112, "alone_minus_off": {"mean": -0.013412421741753555, "se": 0.0007618931787584773, "per_batch": [-0.013913052489499478, -0.015548650494867533, -0.01419347020173134, -0.016397126902899184, -0.013722925668970709, -0.01069121549956098, -0.01025343366178988, -0.012579499014709339]}}, "s7_mixed": {"alone": 0.7826947101585724, "alone_minus_off": {"mean": -0.0002812185064498818, "se": 0.00017174758051827807, "per_batch": [-0.00035652003523867215, -0.00019511431740371066, -0.00020844297284572377, -0.00039581807096078503, 8.418550759736743e-05, -0.0010846515833290438, 0.0005579977107244005, -0.0006513842901428868]}}, "web": {"alone": 0.9506567613584259, "alone_minus_off": {"mean": 6.000044731975518e-05, "se": 9.631581610844272e-05, "per_batch": [-0.0005266108769419464, -4.480148054974986e-05, 0.00030853687564791343, 0.00030509722263061256, -7.223271336176396e-06, 0.0002469670866389606, 9.467644929994634e-05, 0.00010336157316848116]}}}, "s5_causal": {"s1_perspective": {"alone": 0.5516690125210654, "alone_minus_off": {"mean": -0.0021642887974937935, "se": 0.0003518279348647084, "per_batch": [-0.0022480712207532116, -0.002950921319166744, -0.0034509608765503907, -0.002130435087563054, -0.0009878683465559535, -0.0030761676846575714, -0.000640592378801208, -0.0018292934659022153]}}, "s2_concept": {"alone": 0.7201691168186011, "alone_minus_off": {"mean": -0.0010148158781630306, "se": 0.0002551944996123987, "per_batch": [-0.0015721364071996646, -0.0003884228169737103, -0.0008650297381771477, -0.0023907308296438057, -0.0003562620612622691, -0.00037191248249079933, -0.0007946458383116806, -0.001379386851245168]}}, "s3_rules": {"alone": 0.9277824453565631, "alone_minus_off": {"mean": -0.0018306908249405285, "se": 0.00022613039944255578, "per_batch": [-0.0009550196602411409, -0.00278887066639133, -0.002538291944084281, -0.002192434833199086, -0.0011808328808240454, -0.0015964289566340328, -0.0015663319927330388, -0.0018273156654172729]}}, "s4_arith": {"alone": 1.2316764666128335, "alone_minus_off": {"mean": -0.002198400481400936, "se": 0.00021403025089921716, "per_batch": [-0.003272657863268469, -0.002493060506907341, -0.0025671850294290977, -0.0021541686933821413, -0.001368895909542589, -0.0017796764711284219, -0.001641144445858389, -0.0023104149316910405]}}, "s5_causal": {"alone": 0.1417724646021276, "alone_minus_off": {"mean": -0.5186652180094076, "se": 0.007470902026256088, "per_batch": [-0.49118512734418557, -0.5035505832009803, -0.5335864502652528, -0.5333256923197978, -0.5513792916223135, -0.5048105495990326, -0.5011616904359175, -0.5303223592877806]}}, "s6_tryfail": {"alone": 0.6705546587887811, "alone_minus_off": {"mean": -0.0022666023513836925, "se": 0.0001809747475090292, "per_batch": [-0.0021789341951063967, -0.0022350435349504094, -0.002426288242709851, -0.0024120136826881744, -0.002950104401575171, -0.002410852799794805, -0.0023831205973432468, -0.0011364613569014859]}}, "s7_mixed": {"alone": 0.782840250476865, "alone_minus_off": {"mean": -0.0001356781881572261, "se": 0.00016600548853656571, "per_batch": [0.00014326154816868275, 0.0003542842607775487, 1.6510334482799927e-05, 2.966700727380367e-05, 0.0002087009468220158, -0.0009347687031020513, -0.00012124776885824584, -0.0007818331308223625]}}, "web": {"alone": 0.9506011034730404, "alone_minus_off": {"mean": 4.342561934281286e-06, "se": 0.00010631213097956605, "per_batch": [-0.00025565221050771747, -0.00029426231562623695, 0.0002472250606152526, 0.0005403694890109279, 9.476244062556205e-05, -0.000294778263578821, -0.00012821306621824125, 0.0001252893611535244]}}}, "s6_tryfail": {"s1_perspective": {"alone": 0.5567579630367223, "alone_minus_off": {"mean": 0.0029246617181631768, "se": 0.0002543385910052468, "per_batch": [0.003838050827979944, 0.001479093793082864, 0.002770253544435386, 0.003165297693467295, 0.00271792782291036, 0.003174197795649425, 0.003590223828086603, 0.0026622484396935375]}}, "s2_concept": {"alone": 0.7224646821232816, "alone_minus_off": {"mean": 0.0012807494265175134, "se": 0.0003111885603827908, "per_batch": [0.00035630505692507697, 0.001014955614066837, 0.0019954717022985236, 1.2038785560442022e-05, 0.0021480633092765267, 0.0013731954758141596, 0.0024819246302638653, 0.0008640408379346765]}}, "s3_rules": {"alone": 0.9293230229473293, "alone_minus_off": {"mean": -0.00029011323417417467, "se": 0.0002934363752189996, "per_batch": [0.0003786198058745027, -0.001287204150382637, -0.0012136815671388579, 0.0005216233800668935, -0.0010353355581940171, 0.00047802577807321, 0.0004824973269955679, -0.0006454508886880594]}}, "s4_arith": {"alone": 1.2328923409587709, "alone_minus_off": {"mean": -0.0009825261354635961, "se": 0.0004543478028467116, "per_batch": [-5.899004924603268e-05, -0.001759210535675626, -0.002379895922639541, -0.0005530962051749633, -0.0010519318840023217, 0.0014321855250598592, -0.0010483202483342335, -0.00244094976369591]}}, "s5_causal": {"alone": 0.6578710404023479, "alone_minus_off": {"mean": -0.002566642209187356, "se": 0.0003369519063570487, "per_batch": [-0.0031445307883756213, -0.002387979107230098, -0.001532795375814744, -0.0035148954270086685, -0.002257960223177813, -0.003644914311060843, -0.0009811180275096643, -0.003068944413321395]}}, "s6_tryfail": {"alone": 0.14373820614727262, "alone_minus_off": {"mean": -0.5290830549928922, "se": 0.005677738773715752, "per_batch": [-0.5331285894529934, -0.5157443896232985, -0.5363273162742221, -0.546233312734522, -0.5415691432431227, -0.5325711076902215, -0.5310587997476909, -0.49603178117706603]}}, "s7_mixed": {"alone": 0.78304436701115, "alone_minus_off": {"mean": 6.843834612772076e-05, "se": 0.0002105852935495399, "per_batch": [0.000313352389871957, -0.0006411513224164889, 0.0010871023361037624, -8.358356831927871e-05, 0.00017172467688653104, -6.608433359400756e-05, 0.0005001685443717929, -0.0007340219538825021]}}, "web": {"alone": 0.9507048842539205, "alone_minus_off": {"mean": 0.00010812334281433411, "se": 0.00015651834464667334, "per_batch": [-0.00030096963900994034, -1.909007424583109e-05, 0.0009224289479027314, 0.00023716407553986407, 0.0005877507093237089, -0.00021110870393403758, -8.710921266197325e-05, -0.00026407936039984925]}}}, "s7_mixed": {"s1_perspective": {"alone": 0.5301810647911811, "alone_minus_off": {"mean": -0.023652236527378127, "se": 0.0018263450525808275, "per_batch": [-0.0223814352223215, -0.01724728014177679, -0.01742059565818377, -0.030730719969421005, -0.02557897966282552, -0.023594385863194067, -0.03055925326651071, -0.02170524243479166]}}, "s2_concept": {"alone": 0.7189457719762669, "alone_minus_off": {"mean": -0.0022381607204972342, "se": 0.0005344855806794802, "per_batch": [-0.002248458181717594, -0.0005420033241941846, -0.001995342715310322, -0.004689192967099909, -0.0016715423793982609, -0.002954747933148427, -0.00017086476363226133, -0.003633133499476915]}}, "s3_rules": {"alone": 0.8766972239433113, "alone_minus_off": {"mean": -0.05291591223819228, "se": 0.0015800186622646332, "per_batch": [-0.05299146636649943, -0.05466778126445926, -0.05627255137966014, -0.045676614268677085, -0.05665005329830397, -0.05012391363733837, -0.05850703597099516, -0.048437881719604836]}}, "s4_arith": {"alone": 1.1925181882923392, "alone_minus_off": {"mean": -0.04135667880189531, "se": 0.0013366082522658334, "per_batch": [-0.03575338729654387, -0.041290712706373744, -0.04637022029960658, -0.041969614220654705, -0.037375785633458225, -0.04309360683536867, -0.045790466783548256, -0.039209636639608414]}}, "s5_causal": {"alone": 0.5628390330950863, "alone_minus_off": {"mean": -0.09759864951644884, "se": 0.0023637599508642563, "per_batch": [-0.0913786389740191, -0.09711692611138212, -0.10398582719551719, -0.10537798375858898, -0.10422965559877806, -0.0871903024818943, -0.09315156512610112, -0.09835829688530984]}}, "s6_tryfail": {"alone": 0.595420689472129, "alone_minus_off": {"mean": -0.07740057166803564, "se": 0.0030540581316401906, "per_batch": [-0.07864039997256345, -0.09383296038881961, -0.07079928096310084, -0.08369561501636213, -0.08048697769487745, -0.07407069195215044, -0.0684162033614204, -0.06926244399499082]}}, "s7_mixed": {"alone": 0.7253837275330026, "alone_minus_off": {"mean": -0.05759220113201966, "se": 0.033931745802587225, "per_batch": [-0.00029391835032455127, -1.6854299784485605e-05, -0.03701101043097821, -0.00046848074095029446, -0.0001168622112612816, -0.20071557736414453, 4.351161066828979e-05, -0.22215841726938224]}}, "web": {"alone": 0.9502741107091719, "alone_minus_off": {"mean": -0.00032265020193432414, "se": 0.00021183016570555945, "per_batch": [-0.0010698180796921974, -0.0006851788810374737, 0.0006554258824381654, 1.616636918111425e-05, -0.0004909244768878107, -0.0008346318046371648, -0.0005040811496789255, 0.0003318405248396994]}}}}, "battery": {"s1_perspective": {"acc_off": {"in": 0.55, "transfer": 0.4625}, "acc_all_on": {"in": 1.0, "transfer": 0.425}, "acc_member_masked": {"in": 0.55, "transfer": 0.4375}, "acc_member_alone": {"in": 0.925, "transfer": 0.4625}, "margin_all_on_minus_off": {"in": {"mean": 2.9345000850444194, "se": 0.14126824992022288, "per_batch": null}, "transfer": {"mean": -0.1433504432439804, "se": 0.09440425493672008, "per_batch": null}}}, "s2_concept": {"acc_off": {"in": 0.2976190476190476, "transfer": 0.6779661016949152}, "acc_all_on": {"in": 0.9523809523809523, "transfer": 0.7372881355932204}, "acc_member_masked": {"in": 0.9047619047619048, "transfer": 0.7372881355932204}, "acc_member_alone": {"in": 0.9047619047619048, "transfer": 0.7033898305084746}, "margin_all_on_minus_off": {"in": {"mean": 4.3878605065956, "se": 0.3587289496273999, "per_batch": null}, "transfer": {"mean": 0.08992467239751654, "se": 0.20150904225794508, "per_batch": null}}}, "s3_rules": {"acc_off": {"in": 0.155, "transfer": 0.07, "held-out": 0.05}, "acc_all_on": {"in": 0.925, "transfer": 0.35, "held-out": 0.48333333333333334}, "acc_member_masked": {"in": 0.85, "transfer": 0.24, "held-out": 0.475}, "acc_member_alone": {"in": 0.335, "transfer": 0.1, "held-out": 0.14166666666666666}, "margin_all_on_minus_off": {"in": {"mean": 3.990097544824821, "se": 0.1911417047430129, "per_batch": null}, "transfer": {"mean": 1.0009039276838303, "se": 0.17322923213916994, "per_batch": null}, "held-out": {"mean": 1.915777622532672, "se": 0.3280263754669791, "per_batch": null}}}, "s4_arith": {"acc_off": {"in": 0.34375, "transfer": 0.24166666666666667, "held-out": 0.35}, "acc_all_on": {"in": 0.625, "transfer": 0.25833333333333336, "held-out": 0.2875}, "acc_member_masked": {"in": 0.48125, "transfer": 0.2708333333333333, "held-out": 0.2875}, "acc_member_alone": {"in": 0.4125, "transfer": 0.2375, "held-out": 0.4}, "margin_all_on_minus_off": {"in": {"mean": 1.8118917116647935, "se": 0.24097546440870174, "per_batch": null}, "transfer": {"mean": -0.32058004463712375, "se": 0.07234312653476457, "per_batch": null}, "held-out": {"mean": -0.13230020999908448, "se": 0.11122603395085565, "per_batch": null}}}}, "probes_all_on": {"p0_grounding": {"acc": 0.3333333333333333, "n": 30, "families": {"count": 0.125, "permanence": 0.5714285714285714, "sequence": 0.375, "attribute": 0.2857142857142857}}, "p1_self_other": {"acc": 0.5333333333333333, "n": 30, "families": {"belief": 0.25, "reality": 0.6666666666666666, "perspective": 0.8, "reference": 0.8571428571428571, "second_order": 0.0}}, "p2_difference": {"acc": 0.5666666666666667, "n": 30, "families": {"category": 1.0, "antonym": 0.5, "odd_one_out": 0.5, "property": 0.5, "analogy": 0.3333333333333333}}, "p3_logic": {"acc": 0.23333333333333334, "n": 30, "families": {"d1": 0.25, "d2": 0.25, "d3": 0.5, "d4": 0.5, "d5": 0.0, "d6": 0.0, "negation": 0.25, "contrapositive": 0.0}}, "p4_arithmetic": {"acc": 0.36666666666666664, "n": 30, "families": {"add2": 0.4, "add3": 0.0, "sub2": 0.75, "sub3": 0.2, "mul1": 0.5, "mul2": 0.3333333333333333, "word": 0.4}}, "p5_causal": {"acc": 0.5333333333333333, "n": 30, "families": {"effect": 0.7142857142857143, "cause": 0.42857142857142855, "counterfactual": 0.5, "prevention": 0.4, "chain": 0.6}}, "p6_learning": {"acc": 0.36666666666666664, "n": 30, "families": {"error_detect": 0.25, "correction": 0.25, "strategy": 0.7142857142857143, "check": 0.2857142857142857}}, "p7_composite": {"acc": 0.26666666666666666, "n": 30, "families": {"logic_arith": 0.25, "causal_category": 0.25, "belief_causal": 0.14285714285714285, "three_step": 0.42857142857142855}}, "p8_articulation": {"acc": 0.5333333333333333, "n": 30, "families": {"definition": 0.5, "register": 0.5714285714285714, "grammar": 0.875, "paraphrase": 0.14285714285714285}}}, "probes_mean_off": 0.41111111111111115, "probes_mean_all_on": 0.4148148148148148, "gates": {"s1_perspective": {"mean": 0.21163099743248942, "max": 0.554569513244894}, "s2_concept": {"mean": 0.27282737975382454, "max": 0.48724048599867004}, "s3_rules": {"mean": 0.12374002504886239, "max": 0.17456351278495152}, "s4_arith": {"mean": 0.10761241921992425, "max": 0.16724209369972712}, "s5_causal": {"mean": 0.18640250725757604, "max": 0.3602895342887907}, "s6_tryfail": {"mean": 0.2293551327826076, "max": 0.580794738809484}, "s7_mixed": {"mean": 0.2532132549279672, "max": 0.46651751205856823}}, "chain_proof": "compiled chain: 23 cache entries (limit 256), 0 limit warnings", "curve": [{"step": 1, "train_bpb": {"s7_mixed": 0.18391037862641058}, "quiet_kl": {"s7_mixed": 0.00010507344268262386}}, {"step": 50, "train_bpb": {"s7_mixed": 0.43951175708177337}, "quiet_kl": {"s7_mixed": 0.00014182399143464864}}, {"step": 100, "train_bpb": {"s7_mixed": 0.50807439834969}, "quiet_kl": {"s7_mixed": 0.00015605382839567028}}, {"step": 150, "train_bpb": {"s7_mixed": 0.4528918703777141}, "quiet_kl": {"s7_mixed": 0.00020163679917459376}}, {"step": 200, "train_bpb": {"s7_mixed": 0.46620069380143137}, "quiet_kl": {"s7_mixed": 0.00021775517554488034}}, {"step": 250, "train_bpb": {"s7_mixed": 0.49306301208958303}, "quiet_kl": {"s7_mixed": 0.0003050484345294535}}, {"step": 300, "train_bpb": {"s7_mixed": 0.5268911334204167}, "quiet_kl": {"s7_mixed": 0.0003348659080802463}}, {"step": 350, "train_bpb": {"s7_mixed": 0.37687946882239237}, "quiet_kl": {"s7_mixed": 0.00037770215625641866}}, {"step": 400, "train_bpb": {"s7_mixed": 0.4406005360874894}, "quiet_kl": {"s7_mixed": 0.00035777518700342626}}, {"step": 450, "train_bpb": {"s7_mixed": 0.4244067072426907}, "quiet_kl": {"s7_mixed": 0.00034811638994142413}}, {"step": 500, "train_bpb": {"s7_mixed": 0.4225100709601468}, "quiet_kl": {"s7_mixed": 0.00045300420752028006}}, {"step": 550, "train_bpb": {"s7_mixed": 0.5462027794721336}, "quiet_kl": {"s7_mixed": 0.00045439474226441237}}, {"step": 600, "train_bpb": {"s7_mixed": 0.3887113156132034}, "quiet_kl": {"s7_mixed": 0.0004033842362696305}}, {"step": 650, "train_bpb": {"s7_mixed": 0.4558821571283043}, "quiet_kl": {"s7_mixed": 0.0004077054397203028}}, {"step": 700, "train_bpb": {"s7_mixed": 0.529396996100876}, "quiet_kl": {"s7_mixed": 0.0005606229681870901}}, {"step": 750, "train_bpb": {"s7_mixed": 0.4394158539124715}, "quiet_kl": {"s7_mixed": 0.0005720105880754999}}, {"step": 800, "train_bpb": {"s7_mixed": 0.4267374783096607}, "quiet_kl": {"s7_mixed": 0.0007040595621219837}}, {"step": 850, "train_bpb": {"s7_mixed": 0.46898347450841094}, "quiet_kl": {"s7_mixed": 0.0006275923288194462}}, {"step": 900, "train_bpb": {"s7_mixed": 0.5083611837195702}, "quiet_kl": {"s7_mixed": 0.0007855595098226331}}, {"step": 950, "train_bpb": {"s7_mixed": 0.482274248621643}, "quiet_kl": {"s7_mixed": 0.0007818817064980977}}, {"step": 1000, "train_bpb": {"s7_mixed": 0.4055768883230769}, "quiet_kl": {"s7_mixed": 0.0006589336018078029}}, {"step": 1050, "train_bpb": {"s7_mixed": 0.4017134195341911}, "quiet_kl": {"s7_mixed": 0.0006686708654160611}}, {"step": 1100, "train_bpb": {"s7_mixed": 0.3963190596473921}, "quiet_kl": {"s7_mixed": 0.0006770050773047842}}, {"step": 1150, "train_bpb": {"s7_mixed": 0.47629623884434097}, "quiet_kl": {"s7_mixed": 0.0006732076904154382}}, {"step": 1200, "train_bpb": {"s7_mixed": 0.42714465986555444}, "quiet_kl": {"s7_mixed": 0.0005595282648573629}}, {"step": 1250, "train_bpb": {"s7_mixed": 0.4475169099726988}, "quiet_kl": {"s7_mixed": 0.0005549186043208465}}, {"step": 1300, "train_bpb": {"s7_mixed": 0.4104264790284437}, "quiet_kl": {"s7_mixed": 0.0007088875194313004}}, {"step": 1350, "train_bpb": {"s7_mixed": 0.5276948854176241}, "quiet_kl": {"s7_mixed": 0.0008498149606748484}}, {"step": 1400, "train_bpb": {"s7_mixed": 0.4393258917763546}, "quiet_kl": {"s7_mixed": 0.000784389371983707}}, {"step": 1450, "train_bpb": {"s7_mixed": 0.3758816147869154}, "quiet_kl": {"s7_mixed": 0.0005825610674219206}}, {"step": 1500, "train_bpb": {"s7_mixed": 0.43074227099343043}, "quiet_kl": {"s7_mixed": 0.0006582151740440168}}, {"step": 1550, "train_bpb": {"s7_mixed": 0.37648010393172515}, "quiet_kl": {"s7_mixed": 0.0005620559109956958}}, {"step": 1600, "train_bpb": {"s7_mixed": 0.4728459082304615}, "quiet_kl": {"s7_mixed": 0.0005436321225715801}}, {"step": 1650, "train_bpb": {"s7_mixed": 0.44505064344913814}, "quiet_kl": {"s7_mixed": 0.000595029141404666}}, {"step": 1700, "train_bpb": {"s7_mixed": 0.396250932159906}, "quiet_kl": {"s7_mixed": 0.0006003971752943471}}, {"step": 1750, "train_bpb": {"s7_mixed": 0.4505914023313588}, "quiet_kl": {"s7_mixed": 0.0006262613250873983}}, {"step": 1800, "train_bpb": {"s7_mixed": 0.44400996412125876}, "quiet_kl": {"s7_mixed": 0.000600151822436601}}, {"step": 1850, "train_bpb": {"s7_mixed": 0.4506725095644839}, "quiet_kl": {"s7_mixed": 0.0006069937266875059}}, {"step": 1900, "train_bpb": {"s7_mixed": 0.4530119874143876}, "quiet_kl": {"s7_mixed": 0.0006449732993496582}}, {"step": 1950, "train_bpb": {"s7_mixed": 0.45101998954665823}, "quiet_kl": {"s7_mixed": 0.0006712710618739947}}, {"step": 2000, "train_bpb": {"s7_mixed": 0.4289319741936967}, "quiet_kl": {"s7_mixed": 0.0006858829472912476}}, {"step": 2050, "train_bpb": {"s7_mixed": 0.38098950467702025}, "quiet_kl": {"s7_mixed": 0.0006425891746766866}}, {"step": 2100, "train_bpb": {"s7_mixed": 0.35036785083780786}, "quiet_kl": {"s7_mixed": 0.0006072497041895986}}, {"step": 2150, "train_bpb": {"s7_mixed": 0.472408118008536}, "quiet_kl": {"s7_mixed": 0.0005673468514578417}}, {"step": 2200, "train_bpb": {"s7_mixed": 0.4324069957655702}, "quiet_kl": {"s7_mixed": 0.0006037085317075252}}, {"step": 2250, "train_bpb": {"s7_mixed": 0.35037158866574586}, "quiet_kl": {"s7_mixed": 0.0005888624180806801}}, {"step": 2300, "train_bpb": {"s7_mixed": 0.4314261021216754}, "quiet_kl": {"s7_mixed": 0.0006379408313659951}}, {"step": 2350, "train_bpb": {"s7_mixed": 0.4041860200712938}, "quiet_kl": {"s7_mixed": 0.0006755296676419676}}, {"step": 2400, "train_bpb": {"s7_mixed": 0.4465602274014575}, "quiet_kl": {"s7_mixed": 0.0006209077264065855}}, {"step": 2450, "train_bpb": {"s7_mixed": 0.4689742616127825}, "quiet_kl": {"s7_mixed": 0.0007118177850497886}}, {"step": 2500, "train_bpb": {"s7_mixed": 0.4237538064417793}, "quiet_kl": {"s7_mixed": 0.0006543430540477857}}, {"step": 2550, "train_bpb": {"s7_mixed": 0.3320753638173786}, "quiet_kl": {"s7_mixed": 0.0005500493821455166}}, {"step": 2600, "train_bpb": {"s7_mixed": 0.40198912873765436}, "quiet_kl": {"s7_mixed": 0.0006813018309185281}}, {"step": 2650, "train_bpb": {"s7_mixed": 0.3682999901341251}, "quiet_kl": {"s7_mixed": 0.0006372535979608073}}, {"step": 2700, "train_bpb": {"s7_mixed": 0.44538568080899005}, "quiet_kl": {"s7_mixed": 0.0006543421547394245}}, {"step": 2750, "train_bpb": {"s7_mixed": 0.39675399990181454}, "quiet_kl": {"s7_mixed": 0.0006275615433696658}}, {"step": 2800, "train_bpb": {"s7_mixed": 0.4422792964175107}, "quiet_kl": {"s7_mixed": 0.000630935745430179}}, {"step": 2850, "train_bpb": {"s7_mixed": 0.3864777311898118}, "quiet_kl": {"s7_mixed": 0.0006663269607815891}}, {"step": 2900, "train_bpb": {"s7_mixed": 0.37489234155708007}, "quiet_kl": {"s7_mixed": 0.0008668358420254662}}, {"step": 2950, "train_bpb": {"s7_mixed": 0.3731075154954283}, "quiet_kl": {"s7_mixed": 0.0007113764173118397}}, {"step": 3000, "train_bpb": {"s7_mixed": 0.4007390855773231}, "quiet_kl": {"s7_mixed": 0.0006348444864852354}}, {"step": 3050, "train_bpb": {"s7_mixed": 0.3594466596012793}, "quiet_kl": {"s7_mixed": 0.0006329703889787197}}, {"step": 3100, "train_bpb": {"s7_mixed": 0.4813486541180729}, "quiet_kl": {"s7_mixed": 0.0006572600704384967}}, {"step": 3150, "train_bpb": {"s7_mixed": 0.47415633820788655}, "quiet_kl": {"s7_mixed": 0.0006009200244443491}}, {"step": 3200, "train_bpb": {"s7_mixed": 0.44332235152447785}, "quiet_kl": {"s7_mixed": 0.0005902046879054979}}, {"step": 3250, "train_bpb": {"s7_mixed": 0.43420497054581547}, "quiet_kl": {"s7_mixed": 0.000607842771569267}}, {"step": 3300, "train_bpb": {"s7_mixed": 0.36617496739432936}, "quiet_kl": {"s7_mixed": 0.0006804684142116457}}, {"step": 3350, "train_bpb": {"s7_mixed": 0.38782746651222166}, "quiet_kl": {"s7_mixed": 0.0006279479002114385}}, {"step": 3400, "train_bpb": {"s7_mixed": 0.4314135925873529}, "quiet_kl": {"s7_mixed": 0.0007009542034938931}}, {"step": 3450, "train_bpb": {"s7_mixed": 0.3233689771424902}, "quiet_kl": {"s7_mixed": 0.0007189983641728759}}, {"step": 3500, "train_bpb": {"s7_mixed": 0.3901798100617689}, "quiet_kl": {"s7_mixed": 0.0006233361706836149}}, {"step": 3550, "train_bpb": {"s7_mixed": 0.4181547686048295}, "quiet_kl": {"s7_mixed": 0.0005844940530369058}}, {"step": 3600, "train_bpb": {"s7_mixed": 0.3987737536733546}, "quiet_kl": {"s7_mixed": 0.0006440422084415332}}, {"step": 3650, "train_bpb": {"s7_mixed": 0.4449815107230575}, "quiet_kl": {"s7_mixed": 0.0007193569594528527}}, {"step": 3700, "train_bpb": {"s7_mixed": 0.45657340952291237}, "quiet_kl": {"s7_mixed": 0.0006878400593996048}}, {"step": 3750, "train_bpb": {"s7_mixed": 0.519433772843458}, "quiet_kl": {"s7_mixed": 0.0007054416893515736}}, {"step": 3800, "train_bpb": {"s7_mixed": 0.3606202371252538}, "quiet_kl": {"s7_mixed": 0.0005959226679988206}}, {"step": 3850, "train_bpb": {"s7_mixed": 0.3988663512286177}, "quiet_kl": {"s7_mixed": 0.0006451727595413103}}, {"step": 3900, "train_bpb": {"s7_mixed": 0.4599975742663331}, "quiet_kl": {"s7_mixed": 0.0009181026584701613}}, {"step": 3950, "train_bpb": {"s7_mixed": 0.3999072041514852}, "quiet_kl": {"s7_mixed": 0.0006448366813128814}}, {"step": 4000, "train_bpb": {"s7_mixed": 0.42408230593490326}, "quiet_kl": {"s7_mixed": 0.0006477077031740918}}], "steps_done": {"s1_perspective": 3200, "s2_concept": 2400, "s3_rules": 800, "s4_arith": 800, "s5_causal": 4000, "s6_tryfail": 4000, "s7_mixed": 4000, "s8_register": 0}, "stage_seconds": 3180.8521752357483, "inert_at_birth": true}, {"close": 8, "attached": ["s1_perspective", "s2_concept", "s3_rules", "s4_arith", "s5_causal", "s6_tryfail", "s7_mixed", "s8_register"], "rows": {"s1_perspective": {"off": 0.5538333013185592, "all_on": 0.048373615296375044, "all_on_minus_off": {"mean": -0.5054596860221842, "se": 0.00476469862979187, "per_batch": [-0.5075277787424239, -0.524819973228365, -0.5155556977830127, -0.5109784708960824, -0.49236133281680977, -0.5080338860627083, -0.4814757876706894, -0.5029245609773819]}}, "s2_concept": {"off": 0.7211839326967642, "all_on": 0.12122929684668586, "all_on_minus_off": {"mean": -0.5999546358500782, "se": 0.0038003577783819304, "per_batch": [-0.5989522403439822, -0.5977073224278805, -0.5871718157208443, -0.6226180097562001, -0.5983347151382279, -0.5978286776858955, -0.5909489954393344, -0.6060753102882613]}}, "s3_rules": {"off": 0.9296131361815035, "all_on": 0.22223428311624405, "all_on_minus_off": {"mean": -0.7073788530652595, "se": 0.005481734405791783, "per_batch": [-0.6871795337170853, -0.7053607441465413, -0.7227630666480709, -0.6918663834205673, -0.6940680193255875, -0.7305813979562941, -0.7132290364148401, -0.7139826428930895]}}, "s4_arith": {"off": 1.2338748670942346, "all_on": 0.3266083503971684, "all_on_minus_off": {"mean": -0.9072665166970661, "se": 0.006218215976645246, "per_batch": [-0.9231981618064895, -0.8984708617285679, -0.9223980125233502, -0.8811832301573326, -0.8856287237081597, -0.9283192462055686, -0.9108328886812611, -0.9081010087657996]}}, "s5_causal": {"off": 0.6604376826115352, "all_on": 0.042796246816605044, "all_on_minus_off": {"mean": -0.6176414357949302, "se": 0.006923647014033164, "per_batch": [-0.5835290135717144, -0.6275026717549214, -0.6277719267182206, -0.6293146110964599, -0.6434476885154492, -0.6158446719069084, -0.5949645646097518, -0.6187563381860158]}}, "s6_tryfail": {"off": 0.6728212611401647, "all_on": 0.05430685618921398, "all_on_minus_off": {"mean": -0.6185144049509508, "se": 0.003618081455233805, "per_batch": [-0.6277507137331282, -0.6248157008091663, -0.6212136800461221, -0.6274168524121408, -0.6235811878434416, -0.6173387228161585, -0.605550182761683, -0.6004481991857654]}}, "s7_mixed": {"off": 0.7829759286650222, "all_on": 0.7197570947572544, "all_on_minus_off": {"mean": -0.06321883390776786, "se": 0.03903765001186607, "per_batch": [3.5944374030316695e-05, 0.0017838900460744878, -0.03476732476782196, 0.0026619474700545487, -0.00018857897667101664, -0.2425915900270671, 0.006483831928853323, -0.2391687913095955]}}, "s8_register": {"off": 0.4906504977761471, "all_on": 0.020783710349540914, "all_on_minus_off": {"mean": -0.46986678742660626, "se": 0.008584880672924525, "per_batch": [-0.424971662995257, -0.4888615100568119, -0.45916697615353813, -0.4885827611137393, -0.4627237856336223, -0.49746558441393574, -0.483911404628594, -0.45325061441735126]}}, "web": {"off": 0.9505967609111062, "all_on": 0.9516240885271194, "all_on_minus_off": {"mean": 0.0010273276160133182, "se": 0.0001970294852778097, "per_batch": [0.0007651508136885532, 0.0003732023523722594, 0.0012779170872361245, 0.0003506726251093495, 0.0016574827976905127, 0.0013410347201027184, 0.0006872426728478098, 0.0017659178590592184]}}}, "member_in_group": {"s1_perspective": {"s1_perspective": {"masked": 0.488875502457797, "all_on_minus_masked": {"mean": -0.4405018871614219, "se": 0.006383191856100887, "per_batch": [-0.4470547230980905, -0.4635954823867494, -0.4617328242865797, -0.4396146138858616, -0.4171928326085393, -0.4423971373175445, -0.4139922892205608, -0.4384351944874494]}}, "s2_concept": {"masked": 0.3500910765242899, "all_on_minus_masked": {"mean": -0.22886177967760402, "se": 0.0013236892365547828, "per_batch": [-0.22627925057104323, -0.22900442316312125, -0.22289901755966016, -0.2347777840123446, -0.22804699574576798, -0.22776279441521718, -0.22890285665887106, -0.2332211152948069]}}, "s3_rules": {"masked": 0.4165784783329103, "all_on_minus_masked": {"mean": -0.19434419521666627, "se": 0.0031431616344076564, "per_batch": [-0.18974929711235972, -0.2000638061130372, -0.19835257873695222, -0.18455957013775498, -0.1966997394708352, -0.21098281263366603, -0.18826128171926335, -0.1860844758094614]}}, "s4_arith": {"masked": 0.45663014203114294, "all_on_minus_masked": {"mean": -0.13002179163397448, "se": 0.0041644760582301155, "per_batch": [-0.14069782429013905, -0.1442819642319514, -0.10665063387714241, -0.13080495225588318, -0.1382069135663676, -0.12730420240190793, -0.12508405686408863, -0.12714378558431572]}}, "s5_causal": {"masked": 0.07104123771141899, "all_on_minus_masked": {"mean": -0.028244990894813943, "se": 0.0016682071789153874, "per_batch": [-0.028242039645652847, -0.03632557357596442, -0.028463445810807356, -0.02913863357472122, -0.029634857267038815, -0.024472679763319155, -0.02985978907653602, -0.01982290844447171]}}, "s6_tryfail": {"masked": 0.07494261779154293, "all_on_minus_masked": {"mean": -0.020635761602328953, "se": 0.0020503766135546134, "per_batch": [-0.03092224414898806, -0.020367088424090837, -0.01620042174597485, -0.01880393273594092, -0.024250085842982126, -0.01911778495039202, -0.01162730631929166, -0.02379722865097115]}}, "s7_mixed": {"masked": 0.7240511844562861, "all_on_minus_masked": {"mean": -0.004294089699031799, "se": 0.004597930532360021, "per_batch": [0.0003286588457989126, 0.0010644866175153478, -0.0010564894242501843, 0.0018389244943505245, -0.00031696402554015624, -0.03580786280120468, 0.004450352060714202, -0.004853823359638354]}}, "s8_register": {"masked": 0.042163245859502034, "all_on_minus_masked": {"mean": -0.021379535509961124, "se": 0.0017002995822370094, "per_batch": [-0.029738064324544487, -0.022452168461946642, -0.02476085147459885, -0.023107185885589007, -0.020254291990185105, -0.020090416708973127, -0.0154123824599624, -0.015220922773889374]}}, "web": {"masked": 0.9511940029129744, "all_on_minus_masked": {"mean": 0.000430085614145162, "se": 0.00026380924781764276, "per_batch": [0.000383521311424051, 0.00011832406379364002, -0.0003021735175658957, -0.0005186996750021766, 0.0015886037460197677, 0.0005635011555519265, 0.001404754292247401, 0.0002028535366925821]}}}, "s2_concept": {"s1_perspective": {"masked": 0.052558273643915854, "all_on_minus_masked": {"mean": -0.004184658347540812, "se": 0.0003078561510877705, "per_batch": [-0.004100958556570708, -0.004250739322098698, -0.005116827828470283, -0.0032685679008525523, -0.005624085282732297, -0.003926315549077367, -0.0029810720276461714, -0.00420870031287842]}}, "s2_concept": {"masked": 0.2402348822027759, "all_on_minus_masked": {"mean": -0.11900558535609002, "se": 0.0011530373762330902, "per_batch": [-0.1225278787243147, -0.1214530086542535, -0.11499925219146981, -0.12318742144145789, -0.12033267067087045, -0.11748599233595787, -0.11542543594922354, -0.11663302288117242]}}, "s3_rules": {"masked": 0.3228259117716824, "all_on_minus_masked": {"mean": -0.10059162865543836, "se": 0.0011905375020761421, "per_batch": [-0.10040074134912524, -0.1032851268802395, -0.103470911138834, -0.10010516766578614, -0.09463798967967696, -0.1003108374183867, -0.10494445849142656, -0.09757779662003174]}}, "s4_arith": {"masked": 0.4222925427887909, "all_on_minus_masked": {"mean": -0.09568419239162246, "se": 0.0029541603318599197, "per_batch": [-0.10575189553939607, -0.09572429659602055, -0.08212474698121258, -0.10196733131583458, -0.08748252250354166, -0.10445579128683125, -0.09133529934600165, -0.09663165556414138]}}, "s5_causal": {"masked": 0.04368663089554353, "all_on_minus_masked": {"mean": -0.0008903840789384939, "se": 0.00017618365743653227, "per_batch": [-6.346159816837671e-05, -0.0008099899154432738, -0.0016361085788627813, -0.00112446825423268, -0.0008541518355101377, -0.0013532669861453725, -0.0008610042692554568, -0.0004206211938898724]}}, "s6_tryfail": {"masked": 0.05656437973769234, "all_on_minus_masked": {"mean": -0.002257523548478361, "se": 0.0005706087923100705, "per_batch": [-0.003172854181189566, -0.005777800151398782, -0.001976284887811576, -0.0007715087973127199, -0.001647647539844102, -0.0008875702143559192, -0.0015741303310580906, -0.0022523922848561306]}}, "s7_mixed": {"masked": 0.7201275588233574, "all_on_minus_masked": {"mean": -0.00037046406610310306, "se": 0.0003768025602000594, "per_batch": [-0.00040003164590696194, -0.001629621608250531, 0.0006745159566838854, -0.0007244769167595866, -0.001114189603614646, -0.001266265262640176, -5.460449164895742e-05, 0.001550961043312149]}}, "s8_register": {"masked": 0.02130314699707438, "all_on_minus_masked": {"mean": -0.0005194366475334668, "se": 7.488068529863647e-05, "per_batch": [-0.00024942321387162736, -0.0003920935716780066, -0.0008815145439853046, -0.0006048011768193425, -0.0006605920800363471, -0.0003619724227167387, -0.0006460702949541187, -0.00035902587620624885]}}, "web": {"masked": 0.9523509732009912, "all_on_minus_masked": {"mean": -0.00072688467387165, "se": 0.0003504474129747058, "per_batch": [-0.0012860862631520753, 0.0004939341732779212, -0.0003022595088914004, 0.00021454835695144947, -0.0024405198070684975, -0.0014394087963963687, -0.001019685136965598, -3.560040872863102e-05]}}}, "s3_rules": {"s1_perspective": {"masked": 0.048330860812454884, "all_on_minus_masked": {"mean": 4.275448392015643e-05, "se": 7.038878754519719e-05, "per_batch": [0.00017965737665756598, 4.574201067166633e-05, -0.0003538811764394259, -2.4948233290786326e-05, 0.00014630886576369462, 0.00015877760795125068, -8.800137266336272e-05, 0.0002783807927106488]}}, "s2_concept": {"masked": 0.1217418427674619, "all_on_minus_masked": {"mean": -0.0005125459207760454, "se": 9.49006175210238e-05, "per_batch": [-0.0006866622314011112, -0.0006640465128126549, -0.0001627815790416487, -0.000960856321454831, -0.0004916769009854216, -0.0001609650122919165, -0.000521150427777034, -0.0004522283804437449]}}, "s3_rules": {"masked": 0.31371950296367757, "all_on_minus_masked": {"mean": -0.09148521984743348, "se": 0.0015010901211691878, "per_batch": [-0.08856321848590082, -0.08969024229483619, -0.09240202173797823, -0.09745719378611423, -0.08502276213740773, -0.09731924220229088, -0.09139471935187474, -0.09003235878306504]}}, "s4_arith": {"masked": 0.3269873141683446, "all_on_minus_masked": {"mean": -0.00037896377117621616, "se": 0.00010077575084447828, "per_batch": [6.920151914091566e-05, -0.0005631356924189479, -0.0004638157115456343, -0.00032311240530852325, -4.02654381332912e-05, -0.0007889919086644936, -0.0003437503234120509, -0.0005778402090677037]}}, "s5_causal": {"masked": 0.042946159928157326, "all_on_minus_masked": {"mean": -0.00014991311155228455, "se": 0.00011390409355475711, "per_batch": [-8.24764300043912e-05, 0.00017325102291292083, 2.4856867507513924e-05, -0.0001312550093553605, -0.00020973821718503483, -0.0003813446559990802, -0.000789593847942513, 0.0001969953776476685]}}, "s6_tryfail": {"masked": 0.054598344612787855, "all_on_minus_masked": {"mean": -0.0002914884235738725, "se": 7.050274213793236e-05, "per_batch": [-0.0003105469228798369, -0.0005272235651356261, -0.0003800977817803239, -3.826613981697258e-05, -0.00012880963103840265, -0.0003880036092621733, -4.033530608516661e-05, -0.000518624432592478]}}, "s7_mixed": {"masked": 0.7198911821058911, "all_on_minus_masked": {"mean": -0.00013408734863678412, "se": 0.0003520697900518573, "per_batch": [-0.000182473592565513, 8.882903917051266e-05, 0.0008888923309844365, -9.957795484960563e-05, -0.00033777392629452674, 0.0008923319840016819, -1.3500638092744932e-05, -0.0023094260314485138]}}, "s8_register": {"masked": 0.020768950408795676, "all_on_minus_masked": {"mean": 1.4759940745239127e-05, "se": 6.152329481645823e-05, "per_batch": [0.00017121947784960387, 9.076921845068558e-05, -0.00020381422103148833, -0.00011794382090147063, 2.3894839554249037e-05, 0.00025103017676562575, -0.00021040868080050498, 0.00011333253607521271]}}, "web": {"masked": 0.9510348867141791, "all_on_minus_masked": {"mean": 0.0005892018129402959, "se": 0.00010415958377666816, "per_batch": [0.00014025185177857225, 0.0009393692390126107, 0.0006312623199919987, 0.0008789173372344417, 0.000838759388257948, 0.0006007353994637032, 0.0004234212864242526, 0.0002608976813588404]}}}, "s4_arith": {"s1_perspective": {"masked": 0.04835886543273825, "all_on_minus_masked": {"mean": 1.474986363679022e-05, "se": 6.25378520879157e-05, "per_batch": [-1.359737833384056e-06, -3.069890317901336e-05, -0.00027948793102557795, 0.00033029268098202363, -4.5817253081420484e-05, 0.0001581971665045867, 4.536311139398397e-05, -5.84902246668767e-05]}}, "s2_concept": {"masked": 0.1214020816353812, "all_on_minus_masked": {"mean": -0.00017278478869534736, "se": 6.00177881118449e-05, "per_batch": [-0.0003696337123668819, -0.0001384460339445559, -0.0002502562548367393, -0.00020786253139906674, -0.000267744740646364, 0.00020526129380475655, -0.00021278553478000506, -0.0001408107953939225]}}, "s3_rules": {"masked": 0.22211632989004138, "all_on_minus_masked": {"mean": 0.00011795322620268556, "se": 7.009488593297508e-05, "per_batch": [0.00010733867196974223, -5.7399209725472744e-05, 2.730224582447871e-05, 0.00017881896123458918, -5.555039622870406e-05, 8.345458133118822e-05, 9.394552303379489e-05, 0.000565715432181868]}}, "s4_arith": {"masked": 0.4909305930209086, "all_on_minus_masked": {"mean": -0.16432224262374015, "se": 0.0018882491762791302, "per_batch": [-0.16982837468026973, -0.16114881875001963, -0.16601018784846455, -0.16117953915102995, -0.1541931309206928, -0.17033168190801973, -0.16480585783796642, -0.16708034989345838]}}, "s5_causal": {"masked": 0.04303224161166116, "all_on_minus_masked": {"mean": -0.00023599479505611535, "se": 6.824347388217175e-05, "per_batch": [-6.667283672745494e-05, -0.00011403255920879735, -0.0003849267321490851, -0.00027006919366192456, -0.00013061813610137918, -0.0005548966485510176, -0.00037775989262016363, 1.1017638570899602e-05]}}, "s6_tryfail": {"masked": 0.05824583799180451, "all_on_minus_masked": {"mean": -0.00393898180259052, "se": 0.0005470665786617034, "per_batch": [-0.005015089341569252, -0.007007336369164036, -0.0026410891991795957, -0.003239540454061614, -0.00412116114358925, -0.002820128513185617, -0.0023348525914870105, -0.004332656808487788]}}, "s7_mixed": {"masked": 0.7198070584045589, "all_on_minus_masked": {"mean": -4.9963647304585024e-05, "se": 0.0001930215771214022, "per_batch": [0.00029907782985039155, 0.0006916282304447741, 0.0002137744350224624, 0.00026700306546445507, -0.0001558162816817088, -0.001034432649277106, -0.0005589006196414226, -0.00012204318861852581]}}, "s8_register": {"masked": 0.020842862640232884, "all_on_minus_masked": {"mean": -5.915229069197118e-05, "se": 6.296298682679545e-05, "per_batch": [1.2722685320470245e-05, -0.0003539913528251351, -0.00022805571311638323, -1.2503676163507643e-05, -2.906775522473412e-05, 0.0002272105796211274, -0.00013387237032317506, 4.43392771755681e-05]}}, "web": {"masked": 0.9514533312526441, "all_on_minus_masked": {"mean": 0.00017075727447540823, "se": 0.00013522556768225008, "per_batch": [0.00010318959051769383, 0.0006172457339466142, 0.0007245629080849803, 0.0003351941865316066, -0.00043193442764177803, -0.00016020183927878406, 7.859607144433678e-05, 9.940597219859626e-05]}}}, "s5_causal": {"s1_perspective": {"masked": 0.04828712918491879, "all_on_minus_masked": {"mean": 8.648611145625068e-05, "se": 7.868220478531187e-05, "per_batch": [0.0004717699091480454, -0.00015607425565800082, 0.00031349750023370454, 0.00016342651398238567, 9.72293167737226e-05, 1.2001164355519844e-05, -0.0001435437072052903, -6.641754998008148e-05]}}, "s2_concept": {"masked": 0.12103151948542254, "all_on_minus_masked": {"mean": 0.00019777736126330887, "se": 9.420967631587925e-05, "per_batch": [8.126180253267834e-05, 0.00039718318325197843, 0.0004810354744632789, 0.0002236096928686926, 7.52101630054347e-05, 0.0005646405406139476, -4.704800392665853e-05, -0.00019367396270288106]}}, "s3_rules": {"masked": 0.2218094268495767, "all_on_minus_masked": {"mean": 0.0004248562666673665, "se": 0.0001097632665110703, "per_batch": [0.00026564870208894975, 0.0007817471394968856, 0.0001875040851031795, 0.0004239587322081573, 0.0009555786038564773, 6.204274129875675e-05, 0.00021248456514100234, 0.0005098855641455236]}}, "s4_arith": {"masked": 0.3265086407681018, "all_on_minus_masked": {"mean": 9.970962906662162e-05, "se": 0.00011613822830719095, "per_batch": [0.0006064323247736425, 0.00033704300002834753, -0.0004748870946949535, -3.370859956913774e-05, 0.00034645905016311707, 6.784715576541034e-05, -4.68867701914899e-05, -4.6220337419633495e-06]}}, "s5_causal": {"masked": 0.4673439141302562, "all_on_minus_masked": {"mean": -0.4245476673136512, "se": 0.004496257388605948, "per_batch": [-0.4021316884353738, -0.42759330606235896, -0.42055966172207077, -0.43151684839119114, -0.43663747596135394, -0.44238292725101697, -0.41768355823048714, -0.4178758724553566]}}, "s6_tryfail": {"masked": 0.054627000550180646, "all_on_minus_masked": {"mean": -0.00032014436096665985, "se": 0.00015537827999503574, "per_batch": [-0.0011603884432027298, 0.0001649851067558189, -0.0005011305723250301, -0.0004576082127410522, -0.0004296986531807309, -0.0003269766404950733, 0.00020693275019283836, -5.72702227373198e-05]}}, "s7_mixed": {"masked": 0.7197000744468066, "all_on_minus_masked": {"mean": 5.702031044778344e-05, "se": 0.0002128433349668969, "per_batch": [-0.0004629772961226575, 0.00041026461363324884, 0.0008173475482254888, 0.0007329040666518294, 0.00019279255161730458, -0.0007768026382846127, -0.0005677147304981589, 0.00011034836835982498]}}, "s8_register": {"masked": 0.0208461823756596, "all_on_minus_masked": {"mean": -6.247202611868495e-05, "se": 3.315162192184047e-05, "per_batch": [-0.00018080751063520584, -2.6559227028166016e-05, -6.768995287357443e-05, -2.8060044379833032e-05, -0.00018053878774323615, -7.482454565545943e-05, -5.456955767298502e-05, 0.00011327341703898028]}}, "web": {"masked": 0.9513387048158439, "all_on_minus_masked": {"mean": 0.000285383711275497, "se": 0.00010219697921883184, "per_batch": [0.0007118361919211669, -0.00017284256411709276, 0.0001700908417033853, 1.1006889655273966e-05, 0.00041207043146740396, 0.0002327785179427888, 0.00034654504148856624, 0.0005715843401424836]}}}, "s6_tryfail": {"s1_perspective": {"masked": 0.04833200758739638, "all_on_minus_masked": {"mean": 4.160770897866043e-05, "se": 6.7072907342847e-05, "per_batch": [-0.00018754708076589022, 0.00015612262577855557, 9.270939773077524e-06, 0.0003603681470516554, 0.00014334753949415274, -0.0001973662152385952, 0.00011263788739945824, -6.397217166313057e-05]}}, "s2_concept": {"masked": 0.12089030694930511, "all_on_minus_masked": {"mean": 0.00033898989738073704, "se": 6.683925809745319e-05, "per_batch": [0.0002552867473744891, 0.0006707538361963167, 0.0004286560083598817, 0.0005032104875088983, 0.00010349056015669655, 0.00032628333543377463, 0.00028382511850202907, 0.00014041308551381027]}}, "s3_rules": {"masked": 0.22220542496487505, "all_on_minus_masked": {"mean": 2.885815136898992e-05, "se": 9.447378444553171e-05, "per_batch": [-0.000389863171674637, -0.00015379548553406952, -0.0001000939028021619, 0.00035329536053491095, 0.00031021370649378377, -0.00015813804746836468, 0.00012724566380711844, 0.00024200108759533934]}}, "s4_arith": {"masked": 0.3281873634231737, "all_on_minus_masked": {"mean": -0.0015790130260052507, "se": 0.00024199306777055948, "per_batch": [-0.0013890823731875224, -0.0019512936588580998, -0.0013771940724466858, -0.0009639197624233264, -0.001802399678873634, -0.002857040289627022, -0.0005858159045014633, -0.0017053584681242517]}}, "s5_causal": {"masked": 0.04275379296642028, "all_on_minus_masked": {"mean": 4.245385018476072e-05, "se": 9.023587994328433e-05, "per_batch": [-4.751558175868781e-05, -4.984809646099708e-06, 0.00020825217959241638, -0.00015233363300173192, 0.00021456985478271462, -0.0003454943349809475, 0.0004855876402532974, -1.845051376287571e-05]}}, "s6_tryfail": {"masked": 0.46634543660392513, "all_on_minus_masked": {"mean": -0.4120385804147112, "se": 0.0026571131546053164, "per_batch": [-0.4188988521349778, -0.40254953640901847, -0.42113340659426474, -0.4065976210493646, -0.4166627982019536, -0.41726674752720544, -0.410962863195733, -0.4022368182051715]}}, "s7_mixed": {"masked": 0.719866112947299, "all_on_minus_masked": {"mean": -0.0001090181900446005, "se": 0.00010749507493611866, "per_batch": [-0.0003455131455833982, 2.7603215463356534e-05, -2.5539423653131443e-05, -0.00010103980738196405, -0.00011729216788836094, -0.0006981635711775791, 0.00032577813639689257, 6.202124346738058e-05]}}, "s8_register": {"masked": 0.021277932392216925, "all_on_minus_masked": {"mean": -0.0004942220426760128, "se": 0.00011408404962565606, "per_batch": [-0.0002909986761032504, -0.0003546577855972309, -0.0005645975449512439, -0.00081833644846796, -0.0006723903586084296, -0.0005042746301611133, -0.0008729799049362365, 0.00012445900741736224]}}, "web": {"masked": 0.9513693177276976, "all_on_minus_masked": {"mean": 0.0002547707994218912, "se": 0.00019314210648614568, "per_batch": [0.0003916044960146081, 0.00032109160916082846, 0.0013596088463959655, 0.00036675300296495905, -0.00021730007936504592, -0.0003608196015102427, -0.0002077550422422414, 0.0003849831639562984]}}}, "s7_mixed": {"s1_perspective": {"masked": 0.04831214056808919, "all_on_minus_masked": {"mean": 6.147472828584976e-05, "se": 6.797061297093839e-05, "per_batch": [-3.841662463647394e-05, -3.683115957384131e-05, 0.0004612305973248562, 0.00020218710392058692, -0.00011031612161280929, -1.8434390389357458e-05, 0.00011486560017391623, -8.248717892007928e-05]}}, "s2_concept": {"masked": 0.12155814111128015, "all_on_minus_masked": {"mean": -0.00032884426459428573, "se": 0.00010185190311924741, "per_batch": [-0.0007795973563611003, -0.00048520605374668957, -0.0001893636475156224, -0.00023419737481242264, 0.0001972426027082863, -0.0005429922244366142, -0.0002841260881410457, -0.00031251397444907736]}}, "s3_rules": {"masked": 0.22293477651489502, "all_on_minus_masked": {"mean": -0.0007004933986509837, "se": 0.00030495135094782286, "per_batch": [0.0008038469101327717, -0.001114619560241864, -0.0008634818943194511, -0.0014082799365901233, -0.0016945880546139769, -0.0012548714120204918, -0.0002189339145483582, 0.00014698067299362405]}}, "s4_arith": {"masked": 0.33654175927967844, "all_on_minus_masked": {"mean": -0.009933408882510063, "se": 0.0005167137984719889, "per_batch": [-0.009507501909335081, -0.008937336426062148, -0.009753222121755334, -0.00938962929999998, -0.0076815191094619095, -0.012156056230453682, -0.011753466342615204, -0.010288539620397164]}}, "s5_causal": {"masked": 0.042754618953409484, "all_on_minus_masked": {"mean": 4.162786319555911e-05, "se": 7.1851274965418e-05, "per_batch": [0.00012250807922163653, 2.3190785577283102e-05, 0.000342906533531237, 9.197847146456828e-05, -0.00024075421337656916, 0.00012395918283828605, -0.0002647565420876158, 0.00013399060839564686]}}, "s6_tryfail": {"masked": 0.054629107337653714, "all_on_minus_masked": {"mean": -0.00032225114843972835, "se": 6.760704936653524e-05, "per_batch": [-0.00036204497789756734, -0.00028780221730323463, -0.0004814116865120427, -0.0001996181130733271, -0.0005977042052423345, -0.00014899072022558218, -0.000469931844566944, -3.050542269679435e-05]}}, "s7_mixed": {"masked": 0.7766799017909111, "all_on_minus_masked": {"mean": -0.056922807033656736, "se": 0.032939035727690615, "per_batch": [-0.0008640408379347875, -5.7012248761090234e-05, -0.037091154346280186, -8.048788060377454e-05, 0.0001338025023712719, -0.17738914047099463, -0.008149354915471474, -0.23188506807157921]}}, "s8_register": {"masked": 0.02093909667459546, "all_on_minus_masked": {"mean": -0.0001553863250545475, "se": 7.312767669065217e-05, "per_batch": [-0.00014958594143130513, 0.00011938148837352738, -0.00012939410332844792, -0.00030041741346691075, -7.482320204100362e-05, -0.00019007038872152077, 3.504012149883884e-05, -0.000553221161319558]}}, "web": {"masked": 0.9523267343961352, "all_on_minus_masked": {"mean": -0.0007026458690156806, "se": 0.00022885747306965263, "per_batch": [-0.0009652526279675389, -0.0002135164610461704, 0.0002614996206369291, -0.0011134156816857699, -0.0013346713620208117, -0.0013679500049628146, -0.000968606289659335, 8.074585458006656e-05]}}}, "s8_register": {"s1_perspective": {"masked": 0.04823001515736244, "all_on_minus_masked": {"mean": 0.00014360013901260679, "se": 5.522928877483401e-05, "per_batch": [9.321459676764343e-05, 0.00016068554048425526, 0.0005190382658461223, 6.803526178975755e-05, 6.09248540681534e-05, 0.00010238879629960451, 3.614860342823073e-05, 0.00010836519341708706]}}, "s2_concept": {"masked": 0.12115864825477166, "all_on_minus_masked": {"mean": 7.064859191420123e-05, "se": 0.0001254780195877063, "per_batch": [0.0005083377202877576, -0.00019824225186641786, 0.00017629296605006783, -7.101808589066172e-05, -0.00024423686205654604, 0.0006791487393415507, -2.7936431849534538e-05, -0.0002571570587026062]}}, "s3_rules": {"masked": 0.22174432066730942, "all_on_minus_masked": {"mean": 0.0004899624489346187, "se": 6.527671591224411e-05, "per_batch": [0.0002653477324499609, 0.0006181486428636362, 0.0003293252785709355, 0.00045005709947659733, 0.0002688948746239761, 0.0006355618862635137, 0.0006266187884186314, 0.0007257452888096982]}}, "s4_arith": {"masked": 0.3264866914822855, "all_on_minus_masked": {"mean": 0.00012165891488298552, "se": 0.00015571353483787197, "per_batch": [0.0007909267134866838, -0.0004227118579894773, -8.599132543141152e-06, -0.00010329707967449142, 2.7667708957512804e-05, -0.00034858733546755394, 0.0005323507979144715, 0.00050552150437988]}}, "s5_causal": {"masked": 0.042644493625049976, "all_on_minus_masked": {"mean": 0.00015175319155507055, "se": 0.00011469591664432343, "per_batch": [-8.987974567824936e-05, 9.200534375376768e-05, 0.000838157448979894, -7.646778613987909e-05, 0.0002560122991828104, -0.00021216209767063732, 0.00016508184699693534, 0.00024127822301592272]}}, "s6_tryfail": {"masked": 0.054212520346179585, "all_on_minus_masked": {"mean": 9.433584303439931e-05, "se": 0.00011760307223549397, "per_batch": [6.396679720528653e-05, 1.0447946039909839e-05, -0.000486598038327124, 0.00018005508653768515, -0.00022823172661062818, 0.0003395904930442936, 0.0005418420904589605, 0.00033361409592681107]}}, "s7_mixed": {"masked": 0.7195827258471108, "all_on_minus_masked": {"mean": 0.00017436891014353112, "se": 0.00018387976636143535, "per_batch": [-0.00040536310808358955, 0.0004904085289352267, -5.116483863165655e-05, 0.0011722337482809042, -2.9752998599197333e-05, -0.00015452641180024873, -0.00017451939496304636, 0.0005476357560098566]}}, "s8_register": {"masked": 0.37811960239612546, "all_on_minus_masked": {"mean": -0.35733589204658456, "se": 0.006715568453991392, "per_batch": [-0.32350385528885345, -0.37353149223207016, -0.35149504480207827, -0.3760349926233306, -0.35038914664329546, -0.3749927831335599, -0.36773484520064287, -0.34100497644884586]}}, "web": {"masked": 0.9511670123857046, "all_on_minus_masked": {"mean": 0.000457076141414936, "se": 0.00018927927641685608, "per_batch": [0.0009660265498963039, 0.001135085495694521, 0.0006534480819532229, 0.0006540500212313116, -8.255167241344452e-06, -0.00034714698076665496, -8.470145554995145e-05, 0.0006881025861020795]}}}}, "member_alone": {"s1_perspective": {"s1_perspective": {"alone": 0.05217516482213097, "alone_minus_off": {"mean": -0.5016581364964282, "se": 0.004888586924071733, "per_batch": [-0.503264758784162, -0.5212149750156778, -0.5118989327924157, -0.509032675307594, -0.4859917833594548, -0.5041102201213458, -0.4784889810965285, -0.4992627654942472]}}, "s2_concept": {"alone": 0.24394271260453448, "alone_minus_off": {"mean": -0.47724122009222963, "se": 0.0030459897360259366, "per_batch": [-0.47044852344580024, -0.4716034084441754, -0.4687537634085354, -0.4938985943670891, -0.47575242540274654, -0.47682052365593, -0.47438017583151226, -0.4862723461820483]}}, "s3_rules": {"alone": 0.5034592174177095, "alone_minus_off": {"mean": -0.42615391876379405, "se": 0.003854800114533744, "per_batch": [-0.4117213926773634, -0.4236871716024687, -0.442290362963648, -0.410341145912864, -0.4260619510412451, -0.4339105083003576, -0.4286801719223179, -0.4325386456900876]}}, "s4_arith": {"alone": 0.7531657780810914, "alone_minus_off": {"mean": -0.480709089013143, "se": 0.003514095864453405, "per_batch": [-0.4824103037742731, -0.4703906082881222, -0.490546115055989, -0.47029558787352066, -0.4725516132918761, -0.49690809727670615, -0.47641963209676885, -0.48615075444788836]}}, "s5_causal": {"alone": 0.468494725289589, "alone_minus_off": {"mean": -0.19194295732194622, "se": 0.002210700040436461, "per_batch": [-0.1875164678540393, -0.1992422879855283, -0.19983549914401677, -0.1947394812163009, -0.19148918109764468, -0.18138197568475095, -0.1878665385398705, -0.19347222705341832]}}, "s6_tryfail": {"alone": 0.5038253093623601, "alone_minus_off": {"mean": -0.16899595177780455, "se": 0.002363709528038538, "per_batch": [-0.1689674510278819, -0.1804920514578614, -0.16629625949034343, -0.17642276995579653, -0.17014793994340416, -0.16627970616019783, -0.16280797837855587, -0.16055345780839525]}}, "s7_mixed": {"alone": 0.7741873679808721, "alone_minus_off": {"mean": -0.00878856068415014, "se": 0.009343584315554394, "per_batch": [0.001439838753023448, 0.0025405277185455155, 0.00048086349181242216, 0.002514214372963397, 0.0011621727632054046, -0.07048446672069653, 0.011682351516483447, -0.01964398736853823]}}, "s8_register": {"alone": 0.38014544247244586, "alone_minus_off": {"mean": -0.11050505530370128, "se": 0.0022538323811418055, "per_batch": [-0.09835154656626355, -0.11003101737334497, -0.10848214161967451, -0.11394736330313016, -0.10724494142503016, -0.11986889795498751, -0.11520842609058163, -0.11090610809659768]}}, "web": {"alone": 0.9519326898962615, "alone_minus_off": {"mean": 0.001335928985155277, "se": 0.0004171900864352844, "per_batch": [0.0008697162654132118, -0.0001570201602377752, 0.0007944738556607822, 0.00027680607656377365, 0.003323134771296754, 0.002440089850441529, 0.002121320007067329, 0.0010189112150366109]}}}, "s2_concept": {"s1_perspective": {"alone": 0.5237967463236832, "alone_minus_off": {"mean": -0.03003655499487594, "se": 0.0008929364404090443, "per_batch": [-0.02682404806374683, -0.03162864138957566, -0.02918588580804582, -0.032230279697956554, -0.03334249150108626, -0.031257674811664526, -0.026360984776298668, -0.02946243391063319]}}, "s2_concept": {"alone": 0.36033454325401637, "alone_minus_off": {"mean": -0.36084938944274775, "se": 0.0021975474547336524, "per_batch": [-0.36082329913716604, -0.3578469673813343, -0.3548674109507988, -0.37462086527661115, -0.36102178861409306, -0.36006399573360676, -0.3551193225386502, -0.36243146590972153]}}, "s3_rules": {"alone": 0.6157804256378956, "alone_minus_off": {"mean": -0.31383271054360795, "se": 0.0028684777591963, "per_batch": [-0.3051689393960323, -0.3130032220951231, -0.3208357849677054, -0.3010766122187518, -0.30994988510736754, -0.3192230176592393, -0.3165332949910704, -0.32487092791357386]}}, "s4_arith": {"alone": 0.8761517594682684, "alone_minus_off": {"mean": -0.35772310762596604, "se": 0.004197781721024297, "per_batch": [-0.3465006699646368, -0.3508452096994077, -0.37445335417467074, -0.3532194301945688, -0.34704009354906795, -0.37528652412677577, -0.36393756097898977, -0.35050201831961103]}}, "s5_causal": {"alone": 0.6477866226906935, "alone_minus_off": {"mean": -0.012651059920841742, "se": 0.0009352908521067744, "per_batch": [-0.007499647456174596, -0.011507402164893321, -0.015175620125146061, -0.01262455846923527, -0.013140119460859356, -0.011161158093043633, -0.015682152027599816, -0.014417821569781886]}}, "s6_tryfail": {"alone": 0.6666359373486308, "alone_minus_off": {"mean": -0.0061853237915339004, "se": 0.0012459523146970725, "per_batch": [-0.007500163404127069, -0.013446420562046035, -0.006764636602049623, -0.006561095134753447, -0.006009288799460166, -0.003128321423531699, -0.0018538439893129866, -0.004218820416990177]}}, "s7_mixed": {"alone": 0.7831651579510919, "alone_minus_off": {"mean": 0.0001892292860696254, "se": 0.0011789333136323083, "per_batch": [0.0005264388942911591, 0.0002920265411650025, 0.00241738814052761, 0.0004164559890642572, 0.00026923883992568953, 0.0038801005861157645, 0.001148199172822828, -0.007436013875355307]}}, "s8_register": {"alone": 0.4713110486866243, "alone_minus_off": {"mean": -0.019339449089522814, "se": 0.0010556350189690068, "per_batch": [-0.016995626527901764, -0.021460855087915565, -0.016846775543579995, -0.018466981101695812, -0.025184064500781755, -0.02110523796159397, -0.01751075756289855, -0.017145294429815106]}}, "web": {"alone": 0.9514616616622953, "alone_minus_off": {"mean": 0.0008649007511890849, "se": 0.00018835522783446044, "per_batch": [5.219673453682461e-05, 0.0011502199689704673, 0.0014158471732280686, 0.00112304671013419, 0.0007925820465013445, 0.0004966858956918507, 0.00034379331907474775, 0.001544834161375186]}}}, "s3_rules": {"s1_perspective": {"alone": 0.5535631380718851, "alone_minus_off": {"mean": -0.00027016324667410163, "se": 0.00023913782920347164, "per_batch": [0.0002715606057124287, -0.0004794876306054574, -0.0007019041938338688, 0.0007255088126647546, -0.0014021745524844809, -0.0003248322318171182, 0.00029976576045387393, -0.000549742543482945]}}, "s2_concept": {"alone": 0.7227973879358336, "alone_minus_off": {"mean": 0.0016134552390694584, "se": 0.0002752200095664509, "per_batch": [0.000995564570182128, 0.0010287142261358184, 0.0016847850435147693, 0.0009734218038836007, 0.0022790280979085864, 0.0019874315133706633, 0.0009180433903056562, 0.003040653267254445]}}, "s3_rules": {"alone": 0.8701872076558641, "alone_minus_off": {"mean": -0.05942592852563948, "se": 0.010330913371419828, "per_batch": [-0.06883390622470331, -0.052931358430022946, -0.047376318807154205, -0.08934343927937138, -0.029561753891506393, -0.06865332444129746, -0.10310746880189825, -0.015599858329161886]}}, "s4_arith": {"alone": 1.2291967885059467, "alone_minus_off": {"mean": -0.0046780785882879206, "se": 0.0004241556239844928, "per_batch": [-0.0034991590144548557, -0.00447249482701273, -0.00568497251559541, -0.003813715282882768, -0.004839591795279485, -0.0037889497811585127, -0.00709909986231505, -0.004226645627604553]}}, "s5_causal": {"alone": 0.666408248442262, "alone_minus_off": {"mean": 0.005970565830726776, "se": 0.0006340753987558168, "per_batch": [0.006426389723465098, 0.007101077662799882, 0.001980165246371679, 0.007604900838502515, 0.006885497409943375, 0.006719362169209986, 0.004959076741966384, 0.00608805685355529]}}, "s6_tryfail": {"alone": 0.6717302999433321, "alone_minus_off": {"mean": -0.0010909611968325655, "se": 0.0003520287905236149, "per_batch": [-0.0010005950627198779, -0.0029078826607884523, -0.00020315450633168197, -0.0009048437218519556, -0.0018541019632892786, -0.0005386496625022774, 0.00022035277141796428, -0.001538814768594965]}}, "s7_mixed": {"alone": 0.7822497050494648, "alone_minus_off": {"mean": -0.0007262236155574156, "se": 0.0011695408156214792, "per_batch": [-0.00012589130043161312, 4.1361827532560014e-05, 0.0011479841945092328, 0.00033837586557261545, -0.0006783855663283767, 0.0020663715501165747, 3.9599005361157236e-05, -0.008639204500791475]}}, "s8_register": {"alone": 0.4862129476740065, "alone_minus_off": {"mean": -0.004437550102140565, "se": 0.00032575148500688426, "per_batch": [-0.003361744876415329, -0.004211855119630181, -0.00402490997814231, -0.005412294022652642, -0.005111754340269781, -0.0030379875361660202, -0.00491732795346933, -0.005422526990378929]}}, "web": {"alone": 0.9509403070051203, "alone_minus_off": {"mean": 0.0003435460940141577, "se": 0.0001181112928609609, "per_batch": [0.00013586629418149698, 0.00014833503636912937, 0.001022264876728518, 0.0005037371843772398, 0.0003148142424045375, 0.0005318563477932914, 4.4629497898962533e-05, 4.6865272360085974e-05]}}}, "s4_arith": {"s1_perspective": {"alone": 0.5525393683469615, "alone_minus_off": {"mean": -0.0012939329715976888, "se": 0.00033731606119881077, "per_batch": [-0.0005440671160045207, -0.002287799213102515, 5.077787766727404e-05, -0.002374908425764488, -0.002254219600521523, -0.0003551871696944042, -0.00143721601759772, -0.0011488441077636136]}}, "s2_concept": {"alone": 0.7197303675784188, "alone_minus_off": {"mean": -0.001453565118345393, "se": 0.00024222923191314095, "per_batch": [-0.0015742001990100007, -0.001012590852617512, -0.0006298004674595292, -0.0029371197114349545, -0.0012482930756249067, -0.0015436732784818163, -0.001588259780718082, -0.0010945835814163418]}}, "s3_rules": {"alone": 0.9300652463238748, "alone_minus_off": {"mean": 0.000452110142371287, "se": 0.0003056447991096855, "per_batch": [0.002266817329697246, -0.0006020252693452743, 0.00024163562446222198, 0.0009758295609955114, 0.00042324930377346526, 0.00028738300959174623, 0.00022779102106773585, -0.0002037994412723565]}}, "s4_arith": {"alone": 1.1550061816504822, "alone_minus_off": {"mean": -0.07886868544375247, "se": 0.023194415891436188, "per_batch": [-0.07496078816169094, 0.045132117196045485, -0.145121196572499, -0.12688088860464064, -0.013288755466867475, -0.09220961614732559, -0.08542928612838474, -0.13819106966465688]}}, "s5_causal": {"alone": 0.6593072245240404, "alone_minus_off": {"mean": -0.0011304580874947762, "se": 0.000321404023591384, "per_batch": [-0.0015519714413860797, -0.0018272296740919902, -0.0005069188634181376, -0.0016303525345167103, -0.00010731717413836606, -9.652526279668727e-05, -0.0007122661485483572, -0.002611083601061881]}}, "s6_tryfail": {"alone": 0.6594088393984112, "alone_minus_off": {"mean": -0.013412421741753555, "se": 0.0007618931787584773, "per_batch": [-0.013913052489499478, -0.015548650494867533, -0.01419347020173134, -0.016397126902899184, -0.013722925668970709, -0.01069121549956098, -0.01025343366178988, -0.012579499014709339]}}, "s7_mixed": {"alone": 0.7826947101585724, "alone_minus_off": {"mean": -0.0002812185064498818, "se": 0.00017174758051827807, "per_batch": [-0.00035652003523867215, -0.00019511431740371066, -0.00020844297284572377, -0.00039581807096078503, 8.418550759736743e-05, -0.0010846515833290438, 0.0005579977107244005, -0.0006513842901428868]}}, "s8_register": {"alone": 0.4892565568930727, "alone_minus_off": {"mean": -0.0013939408830744224, "se": 0.0002616046534423677, "per_batch": [-0.0021693461623207844, -0.0008443488244109787, -0.0007489414488447976, -0.001347957021800017, -0.0016850860131537582, -0.0027125103694082275, -0.0006939070005687054, -0.0009494302240881103]}}, "web": {"alone": 0.9506567613584259, "alone_minus_off": {"mean": 6.000044731975518e-05, "se": 9.631581610844272e-05, "per_batch": [-0.0005266108769419464, -4.480148054974986e-05, 0.00030853687564791343, 0.00030509722263061256, -7.223271336176396e-06, 0.0002469670866389606, 9.467644929994634e-05, 0.00010336157316848116]}}}, "s5_causal": {"s1_perspective": {"alone": 0.5516690125210654, "alone_minus_off": {"mean": -0.0021642887974937935, "se": 0.0003518279348647084, "per_batch": [-0.0022480712207532116, -0.002950921319166744, -0.0034509608765503907, -0.002130435087563054, -0.0009878683465559535, -0.0030761676846575714, -0.000640592378801208, -0.0018292934659022153]}}, "s2_concept": {"alone": 0.7201691168186011, "alone_minus_off": {"mean": -0.0010148158781630306, "se": 0.0002551944996123987, "per_batch": [-0.0015721364071996646, -0.0003884228169737103, -0.0008650297381771477, -0.0023907308296438057, -0.0003562620612622691, -0.00037191248249079933, -0.0007946458383116806, -0.001379386851245168]}}, "s3_rules": {"alone": 0.9277824453565631, "alone_minus_off": {"mean": -0.0018306908249405285, "se": 0.00022613039944255578, "per_batch": [-0.0009550196602411409, -0.00278887066639133, -0.002538291944084281, -0.002192434833199086, -0.0011808328808240454, -0.0015964289566340328, -0.0015663319927330388, -0.0018273156654172729]}}, "s4_arith": {"alone": 1.2316764666128335, "alone_minus_off": {"mean": -0.002198400481400936, "se": 0.00021403025089921716, "per_batch": [-0.003272657863268469, -0.002493060506907341, -0.0025671850294290977, -0.0021541686933821413, -0.001368895909542589, -0.0017796764711284219, -0.001641144445858389, -0.0023104149316910405]}}, "s5_causal": {"alone": 0.1417724646021276, "alone_minus_off": {"mean": -0.5186652180094076, "se": 0.007470902026256088, "per_batch": [-0.49118512734418557, -0.5035505832009803, -0.5335864502652528, -0.5333256923197978, -0.5513792916223135, -0.5048105495990326, -0.5011616904359175, -0.5303223592877806]}}, "s6_tryfail": {"alone": 0.6705546587887811, "alone_minus_off": {"mean": -0.0022666023513836925, "se": 0.0001809747475090292, "per_batch": [-0.0021789341951063967, -0.0022350435349504094, -0.002426288242709851, -0.0024120136826881744, -0.002950104401575171, -0.002410852799794805, -0.0023831205973432468, -0.0011364613569014859]}}, "s7_mixed": {"alone": 0.782840250476865, "alone_minus_off": {"mean": -0.0001356781881572261, "se": 0.00016600548853656571, "per_batch": [0.00014326154816868275, 0.0003542842607775487, 1.6510334482799927e-05, 2.966700727380367e-05, 0.0002087009468220158, -0.0009347687031020513, -0.00012124776885824584, -0.0007818331308223625]}}, "s8_register": {"alone": 0.4897797603637444, "alone_minus_off": {"mean": -0.0008707374124027126, "se": 0.00011358742512405733, "per_batch": [-0.0001393489428616057, -0.0008538938615338942, -0.0010459124912221562, -0.0009752706173802306, -0.0011606249193476526, -0.0007594753862101289, -0.0009554496168683313, -0.001075923463797701]}}, "web": {"alone": 0.9506011034730404, "alone_minus_off": {"mean": 4.342561934281286e-06, "se": 0.00010631213097956605, "per_batch": [-0.00025565221050771747, -0.00029426231562623695, 0.0002472250606152526, 0.0005403694890109279, 9.476244062556205e-05, -0.000294778263578821, -0.00012821306621824125, 0.0001252893611535244]}}}, "s6_tryfail": {"s1_perspective": {"alone": 0.5567579630367223, "alone_minus_off": {"mean": 0.0029246617181631768, "se": 0.0002543385910052468, "per_batch": [0.003838050827979944, 0.001479093793082864, 0.002770253544435386, 0.003165297693467295, 0.00271792782291036, 0.003174197795649425, 0.003590223828086603, 0.0026622484396935375]}}, "s2_concept": {"alone": 0.7224646821232816, "alone_minus_off": {"mean": 0.0012807494265175134, "se": 0.0003111885603827908, "per_batch": [0.00035630505692507697, 0.001014955614066837, 0.0019954717022985236, 1.2038785560442022e-05, 0.0021480633092765267, 0.0013731954758141596, 0.0024819246302638653, 0.0008640408379346765]}}, "s3_rules": {"alone": 0.9293230229473293, "alone_minus_off": {"mean": -0.00029011323417417467, "se": 0.0002934363752189996, "per_batch": [0.0003786198058745027, -0.001287204150382637, -0.0012136815671388579, 0.0005216233800668935, -0.0010353355581940171, 0.00047802577807321, 0.0004824973269955679, -0.0006454508886880594]}}, "s4_arith": {"alone": 1.2328923409587709, "alone_minus_off": {"mean": -0.0009825261354635961, "se": 0.0004543478028467116, "per_batch": [-5.899004924603268e-05, -0.001759210535675626, -0.002379895922639541, -0.0005530962051749633, -0.0010519318840023217, 0.0014321855250598592, -0.0010483202483342335, -0.00244094976369591]}}, "s5_causal": {"alone": 0.6578710404023479, "alone_minus_off": {"mean": -0.002566642209187356, "se": 0.0003369519063570487, "per_batch": [-0.0031445307883756213, -0.002387979107230098, -0.001532795375814744, -0.0035148954270086685, -0.002257960223177813, -0.003644914311060843, -0.0009811180275096643, -0.003068944413321395]}}, "s6_tryfail": {"alone": 0.14373820614727262, "alone_minus_off": {"mean": -0.5290830549928922, "se": 0.005677738773715752, "per_batch": [-0.5331285894529934, -0.5157443896232985, -0.5363273162742221, -0.546233312734522, -0.5415691432431227, -0.5325711076902215, -0.5310587997476909, -0.49603178117706603]}}, "s7_mixed": {"alone": 0.78304436701115, "alone_minus_off": {"mean": 6.843834612772076e-05, "se": 0.0002105852935495399, "per_batch": [0.000313352389871957, -0.0006411513224164889, 0.0010871023361037624, -8.358356831927871e-05, 0.00017172467688653104, -6.608433359400756e-05, 0.0005001685443717929, -0.0007340219538825021]}}, "s8_register": {"alone": 0.4905744223254296, "alone_minus_off": {"mean": -7.607545071758315e-05, "se": 0.00021777094369922266, "per_batch": [2.4077571120773023e-06, 0.00037960870611697395, 0.00033695700870295386, 0.0005512903873406971, -0.0009772054222024762, 0.0004076848738703287, -0.0009394122346753075, -0.0003699346820059124]}}, "web": {"alone": 0.9507048842539205, "alone_minus_off": {"mean": 0.00010812334281433411, "se": 0.00015651834464667334, "per_batch": [-0.00030096963900994034, -1.909007424583109e-05, 0.0009224289479027314, 0.00023716407553986407, 0.0005877507093237089, -0.00021110870393403758, -8.710921266197325e-05, -0.00026407936039984925]}}}, "s7_mixed": {"s1_perspective": {"alone": 0.5301810647911811, "alone_minus_off": {"mean": -0.023652236527378127, "se": 0.0018263450525808275, "per_batch": [-0.0223814352223215, -0.01724728014177679, -0.01742059565818377, -0.030730719969421005, -0.02557897966282552, -0.023594385863194067, -0.03055925326651071, -0.02170524243479166]}}, "s2_concept": {"alone": 0.7189457719762669, "alone_minus_off": {"mean": -0.0022381607204972342, "se": 0.0005344855806794802, "per_batch": [-0.002248458181717594, -0.0005420033241941846, -0.001995342715310322, -0.004689192967099909, -0.0016715423793982609, -0.002954747933148427, -0.00017086476363226133, -0.003633133499476915]}}, "s3_rules": {"alone": 0.8766972239433113, "alone_minus_off": {"mean": -0.05291591223819228, "se": 0.0015800186622646332, "per_batch": [-0.05299146636649943, -0.05466778126445926, -0.05627255137966014, -0.045676614268677085, -0.05665005329830397, -0.05012391363733837, -0.05850703597099516, -0.048437881719604836]}}, "s4_arith": {"alone": 1.1925181882923392, "alone_minus_off": {"mean": -0.04135667880189531, "se": 0.0013366082522658334, "per_batch": [-0.03575338729654387, -0.041290712706373744, -0.04637022029960658, -0.041969614220654705, -0.037375785633458225, -0.04309360683536867, -0.045790466783548256, -0.039209636639608414]}}, "s5_causal": {"alone": 0.5628390330950863, "alone_minus_off": {"mean": -0.09759864951644884, "se": 0.0023637599508642563, "per_batch": [-0.0913786389740191, -0.09711692611138212, -0.10398582719551719, -0.10537798375858898, -0.10422965559877806, -0.0871903024818943, -0.09315156512610112, -0.09835829688530984]}}, "s6_tryfail": {"alone": 0.595420689472129, "alone_minus_off": {"mean": -0.07740057166803564, "se": 0.0030540581316401906, "per_batch": [-0.07864039997256345, -0.09383296038881961, -0.07079928096310084, -0.08369561501636213, -0.08048697769487745, -0.07407069195215044, -0.0684162033614204, -0.06926244399499082]}}, "s7_mixed": {"alone": 0.7253837275330026, "alone_minus_off": {"mean": -0.05759220113201966, "se": 0.033931745802587225, "per_batch": [-0.00029391835032455127, -1.6854299784485605e-05, -0.03701101043097821, -0.00046848074095029446, -0.0001168622112612816, -0.20071557736414453, 4.351161066828979e-05, -0.22215841726938224]}}, "s8_register": {"alone": 0.4845792522267137, "alone_minus_off": {"mean": -0.006071245549433418, "se": 0.00026164170598460857, "per_batch": [-0.00563071198924836, -0.005967024063010529, -0.005102338290135067, -0.006354457979741768, -0.006670433105039453, -0.006858926090385076, -0.006892462707303315, -0.0050936101706037795]}}, "web": {"alone": 0.9502741107091719, "alone_minus_off": {"mean": -0.00032265020193432414, "se": 0.00021183016570555945, "per_batch": [-0.0010698180796921974, -0.0006851788810374737, 0.0006554258824381654, 1.616636918111425e-05, -0.0004909244768878107, -0.0008346318046371648, -0.0005040811496789255, 0.0003318405248396994]}}}, "s8_register": {"s1_perspective": {"alone": 0.5493115657088212, "alone_minus_off": {"mean": -0.004521735609737898, "se": 0.0006483562522522641, "per_batch": [-0.0018310992837362594, -0.005235194887926564, -0.002640664617010291, -0.00491633905322697, -0.007160282690359399, -0.0031806041493940285, -0.00630875359027494, -0.004900946605974732]}}, "s2_concept": {"alone": 0.7162121722342967, "alone_minus_off": {"mean": -0.0049717604624675005, "se": 0.0006073309574707387, "per_batch": [-0.004969223718367277, -0.005426912547976004, -0.004289376299506542, -0.007920660985486605, -0.0033564564099012317, -0.006193654201185006, -0.0023122207495250846, -0.005305578787792253]}}, "s3_rules": {"alone": 0.9233612228573387, "alone_minus_off": {"mean": -0.006251913324164804, "se": 0.00034853917656300936, "per_batch": [-0.00586907994334418, -0.005130844414515612, -0.007794253737102519, -0.006117250908539318, -0.006330337412958187, -0.007652540032791477, -0.0053623330625769405, -0.005758667081490199]}}, "s4_arith": {"alone": 1.2275061668012182, "alone_minus_off": {"mean": -0.006368700293016422, "se": 0.00038143246571641337, "per_batch": [-0.005995487191728488, -0.005599927094743995, -0.007759685224278945, -0.004728491002822244, -0.007192658424384435, -0.0060641942607482235, -0.00771024021215605, -0.005898918933268993]}}, "s5_causal": {"alone": 0.6465631434869132, "alone_minus_off": {"mean": -0.013874539124621901, "se": 0.001017778063521396, "per_batch": [-0.013983049428400673, -0.016636311774586687, -0.013160327422335638, -0.0121587434593734, -0.011673494409964014, -0.009312129617954912, -0.01624109564290388, -0.017831161241456006]}}, "s6_tryfail": {"alone": 0.6541641424690243, "alone_minus_off": {"mean": -0.018657118671140338, "se": 0.001357677767058051, "per_batch": [-0.019474154500811136, -0.011719671751720728, -0.02272548651537254, -0.016797932470735022, -0.017247495120090384, -0.020239090336186094, -0.023913370684881974, -0.017139747989324827]}}, "s7_mixed": {"alone": 0.7831211948859651, "alone_minus_off": {"mean": 0.00014526622094283154, "se": 0.00019995371292810766, "per_batch": [0.0003052692052813999, 0.0002072390942897684, -0.0009182153729565545, 0.0005899004924594387, 0.0007511342276432798, 0.0003830483591341638, 0.00034461023666637614, -0.0005008564749752198]}}, "s8_register": {"alone": 0.05160327413789158, "alone_minus_off": {"mean": -0.43904722363825555, "se": 0.009771668843442608, "per_batch": [-0.38445327466017315, -0.4513535261966476, -0.4194941700809394, -0.4559032466083556, -0.43940374165807244, -0.46988133541217036, -0.4617996798548737, -0.4300888146348121]}}, "web": {"alone": 0.950643873408527, "alone_minus_off": {"mean": 4.711249742075929e-05, "se": 0.0001226631879616403, "per_batch": [0.00044466114380581345, 0.00013964991250059455, 0.00040054759385954597, 6.156978900895282e-05, 0.00019554427403112307, -0.0005403694890109279, -0.0003813715282883212, 5.6668283459293534e-05]}}}}, "battery": {"s1_perspective": {"acc_off": {"in": 0.55, "transfer": 0.4625}, "acc_all_on": {"in": 1.0, "transfer": 0.425}, "acc_member_masked": {"in": 0.5625, "transfer": 0.4375}, "acc_member_alone": {"in": 0.925, "transfer": 0.4625}, "margin_all_on_minus_off": {"in": {"mean": 2.914699345713598, "se": 0.14566151517398032, "per_batch": null}, "transfer": {"mean": -0.15880678445100785, "se": 0.09845007783196043, "per_batch": null}}}, "s2_concept": {"acc_off": {"in": 0.2976190476190476, "transfer": 0.6779661016949152}, "acc_all_on": {"in": 0.9523809523809523, "transfer": 0.7203389830508474}, "acc_member_masked": {"in": 0.8809523809523809, "transfer": 0.7372881355932204}, "acc_member_alone": {"in": 0.9047619047619048, "transfer": 0.7033898305084746}, "margin_all_on_minus_off": {"in": {"mean": 4.338690999185755, "se": 0.36369140060469507, "per_batch": null}, "transfer": {"mean": 0.07150405342296018, "se": 0.20424059512301354, "per_batch": null}}}, "s3_rules": {"acc_off": {"in": 0.155, "transfer": 0.07, "held-out": 0.05}, "acc_all_on": {"in": 0.92, "transfer": 0.37, "held-out": 0.4583333333333333}, "acc_member_masked": {"in": 0.845, "transfer": 0.24, "held-out": 0.475}, "acc_member_alone": {"in": 0.335, "transfer": 0.1, "held-out": 0.14166666666666666}, "margin_all_on_minus_off": {"in": {"mean": 3.9818199194761474, "se": 0.1906673520994028, "per_batch": null}, "transfer": {"mean": 0.9188200873136521, "se": 0.19524691605246175, "per_batch": null}, "held-out": {"mean": 1.8548445386208186, "se": 0.34438878769214903, "per_batch": null}}}, "s4_arith": {"acc_off": {"in": 0.34375, "transfer": 0.24166666666666667, "held-out": 0.35}, "acc_all_on": {"in": 0.65625, "transfer": 0.2708333333333333, "held-out": 0.325}, "acc_member_masked": {"in": 0.5125, "transfer": 0.25416666666666665, "held-out": 0.275}, "acc_member_alone": {"in": 0.4125, "transfer": 0.2375, "held-out": 0.4}, "margin_all_on_minus_off": {"in": {"mean": 1.8067259269664646, "se": 0.23766444790041827, "per_batch": null}, "transfer": {"mean": -0.3123103785018126, "se": 0.07182336375020273, "per_batch": null}, "held-out": {"mean": -0.1395750105381012, "se": 0.11209902149995847, "per_batch": null}}}}, "probes_all_on": {"p0_grounding": {"acc": 0.3333333333333333, "n": 30, "families": {"count": 0.125, "permanence": 0.5714285714285714, "sequence": 0.375, "attribute": 0.2857142857142857}}, "p1_self_other": {"acc": 0.5333333333333333, "n": 30, "families": {"belief": 0.25, "reality": 0.6666666666666666, "perspective": 0.8, "reference": 0.8571428571428571, "second_order": 0.0}}, "p2_difference": {"acc": 0.6, "n": 30, "families": {"category": 1.0, "antonym": 0.6666666666666666, "odd_one_out": 0.5, "property": 0.5, "analogy": 0.3333333333333333}}, "p3_logic": {"acc": 0.3, "n": 30, "families": {"d1": 0.5, "d2": 0.25, "d3": 0.5, "d4": 0.5, "d5": 0.0, "d6": 0.0, "negation": 0.25, "contrapositive": 0.25}}, "p4_arithmetic": {"acc": 0.36666666666666664, "n": 30, "families": {"add2": 0.4, "add3": 0.0, "sub2": 0.75, "sub3": 0.2, "mul1": 0.5, "mul2": 0.3333333333333333, "word": 0.4}}, "p5_causal": {"acc": 0.5333333333333333, "n": 30, "families": {"effect": 0.7142857142857143, "cause": 0.42857142857142855, "counterfactual": 0.5, "prevention": 0.4, "chain": 0.6}}, "p6_learning": {"acc": 0.36666666666666664, "n": 30, "families": {"error_detect": 0.25, "correction": 0.25, "strategy": 0.7142857142857143, "check": 0.2857142857142857}}, "p7_composite": {"acc": 0.3, "n": 30, "families": {"logic_arith": 0.25, "causal_category": 0.25, "belief_causal": 0.2857142857142857, "three_step": 0.42857142857142855}}, "p8_articulation": {"acc": 0.5333333333333333, "n": 30, "families": {"definition": 0.5, "register": 0.5714285714285714, "grammar": 0.875, "paraphrase": 0.14285714285714285}}}, "probes_mean_off": 0.41111111111111115, "probes_mean_all_on": 0.4296296296296296, "gates": {"s1_perspective": {"mean": 0.21163099743248942, "max": 0.554569513244894}, "s2_concept": {"mean": 0.27282737975382454, "max": 0.48724048599867004}, "s3_rules": {"mean": 0.12374002504886239, "max": 0.17456351278495152}, "s4_arith": {"mean": 0.10761241921992425, "max": 0.16724209369972712}, "s5_causal": {"mean": 0.18640250725757604, "max": 0.3602895342887907}, "s6_tryfail": {"mean": 0.2293551327826076, "max": 0.580794738809484}, "s7_mixed": {"mean": 0.2532132549279672, "max": 0.46651751205856823}, "s8_register": {"mean": 0.20503783160560274, "max": 0.4696574997215004}}, "chain_proof": "compiled chain: 28 cache entries (limit 256), 0 limit warnings", "curve": [{"step": 1, "train_bpb": {"s8_register": 0.3816426589373634}, "quiet_kl": {"s8_register": 0.00015617026656400412}}, {"step": 50, "train_bpb": {"s8_register": 0.37602695630401234}, "quiet_kl": {"s8_register": 0.00015097479306859895}}, {"step": 100, "train_bpb": {"s8_register": 0.35652581815523754}, "quiet_kl": {"s8_register": 0.00026360212155850603}}, {"step": 150, "train_bpb": {"s8_register": 0.33412798223737317}, "quiet_kl": {"s8_register": 0.0004432446527061984}}, {"step": 200, "train_bpb": {"s8_register": 0.31082399681938017}, "quiet_kl": {"s8_register": 0.0005530692404136061}}, {"step": 250, "train_bpb": {"s8_register": 0.2773647854226762}, "quiet_kl": {"s8_register": 0.0006871957902330905}}, {"step": 300, "train_bpb": {"s8_register": 0.24347875522228024}, "quiet_kl": {"s8_register": 0.0007819080783519894}}, {"step": 350, "train_bpb": {"s8_register": 0.20645377765026882}, "quiet_kl": {"s8_register": 0.000842336752684787}}, {"step": 400, "train_bpb": {"s8_register": 0.1762712599047141}, "quiet_kl": {"s8_register": 0.0007836403581313789}}, {"step": 450, "train_bpb": {"s8_register": 0.14462417359081167}, "quiet_kl": {"s8_register": 0.0007973512221360579}}, {"step": 500, "train_bpb": {"s8_register": 0.11943245822682792}, "quiet_kl": {"s8_register": 0.0006721140583977103}}, {"step": 550, "train_bpb": {"s8_register": 0.09592503130645459}, "quiet_kl": {"s8_register": 0.0007301794056547806}}, {"step": 600, "train_bpb": {"s8_register": 0.08025145315144609}, "quiet_kl": {"s8_register": 0.0007406585075659677}}, {"step": 650, "train_bpb": {"s8_register": 0.06685850188806422}, "quiet_kl": {"s8_register": 0.000720252874889411}}, {"step": 700, "train_bpb": {"s8_register": 0.05742184640443642}, "quiet_kl": {"s8_register": 0.0006950935971690342}}, {"step": 750, "train_bpb": {"s8_register": 0.048940330968905404}, "quiet_kl": {"s8_register": 0.0006782737706089393}}, {"step": 800, "train_bpb": {"s8_register": 0.043671913332398365}, "quiet_kl": {"s8_register": 0.0007164402474882081}}, {"step": 850, "train_bpb": {"s8_register": 0.039501252840917295}, "quiet_kl": {"s8_register": 0.0006420536822406575}}, {"step": 900, "train_bpb": {"s8_register": 0.03662991880445415}, "quiet_kl": {"s8_register": 0.0006049824302317574}}, {"step": 950, "train_bpb": {"s8_register": 0.0336673145963251}, "quiet_kl": {"s8_register": 0.0006159160780953244}}, {"step": 1000, "train_bpb": {"s8_register": 0.032422012084020305}, "quiet_kl": {"s8_register": 0.0005596209713257849}}, {"step": 1050, "train_bpb": {"s8_register": 0.03020115502172758}, "quiet_kl": {"s8_register": 0.0005417256319196895}}, {"step": 1100, "train_bpb": {"s8_register": 0.03357618910506422}, "quiet_kl": {"s8_register": 0.0009038055426208303}}, {"step": 1150, "train_bpb": {"s8_register": 0.028849812252722474}, "quiet_kl": {"s8_register": 0.0005869707418605686}}, {"step": 1200, "train_bpb": {"s8_register": 0.028077739970377578}, "quiet_kl": {"s8_register": 0.000514500854187645}}, {"step": 1250, "train_bpb": {"s8_register": 0.027101416523578847}, "quiet_kl": {"s8_register": 0.00046720630140043793}}, {"step": 1300, "train_bpb": {"s8_register": 0.027416523405212577}, "quiet_kl": {"s8_register": 0.00044800467789173125}}, {"step": 1350, "train_bpb": {"s8_register": 0.026068202237021682}, "quiet_kl": {"s8_register": 0.00044591556768864394}}, {"step": 1400, "train_bpb": {"s8_register": 0.025762688574525367}, "quiet_kl": {"s8_register": 0.00044529465609230103}}, {"step": 1450, "train_bpb": {"s8_register": 0.02576324241240573}, "quiet_kl": {"s8_register": 0.00045845248270779846}}, {"step": 1500, "train_bpb": {"s8_register": 0.025446774099984394}, "quiet_kl": {"s8_register": 0.00044546768127474935}}, {"step": 1550, "train_bpb": {"s8_register": 0.02461874435997404}, "quiet_kl": {"s8_register": 0.00039712456811685114}}, {"step": 1600, "train_bpb": {"s8_register": 0.025056742412184207}, "quiet_kl": {"s8_register": 0.0004089923790888861}}, {"step": 1650, "train_bpb": {"s8_register": 0.025022185158849953}, "quiet_kl": {"s8_register": 0.0004289007332408801}}, {"step": 1700, "train_bpb": {"s8_register": 0.024962973440090422}, "quiet_kl": {"s8_register": 0.0003988529060734436}}, {"step": 1750, "train_bpb": {"s8_register": 0.024660447384962488}, "quiet_kl": {"s8_register": 0.0004088730551302433}}, {"step": 1800, "train_bpb": {"s8_register": 0.02448054350879612}, "quiet_kl": {"s8_register": 0.0004316028399625793}}, {"step": 1850, "train_bpb": {"s8_register": 0.024597016609309817}, "quiet_kl": {"s8_register": 0.00040664577740244566}}, {"step": 1900, "train_bpb": {"s8_register": 0.024282962583966842}, "quiet_kl": {"s8_register": 0.00036117719049798323}}, {"step": 1950, "train_bpb": {"s8_register": 0.02427203373143945}, "quiet_kl": {"s8_register": 0.00035612454754300414}}, {"step": 2000, "train_bpb": {"s8_register": 0.02404518197749598}, "quiet_kl": {"s8_register": 0.00040391481772530827}}, {"step": 2050, "train_bpb": {"s8_register": 0.024355109601598975}, "quiet_kl": {"s8_register": 0.00035723241278901697}}, {"step": 2100, "train_bpb": {"s8_register": 0.023997323720305438}, "quiet_kl": {"s8_register": 0.0003881071141222492}}, {"step": 2150, "train_bpb": {"s8_register": 0.023803482370113153}, "quiet_kl": {"s8_register": 0.00037466409616172315}}, {"step": 2200, "train_bpb": {"s8_register": 0.02347716466502148}, "quiet_kl": {"s8_register": 0.0003761590947397053}}, {"step": 2250, "train_bpb": {"s8_register": 0.023292328165194637}, "quiet_kl": {"s8_register": 0.0003605752787552774}}, {"step": 2300, "train_bpb": {"s8_register": 0.023202005187978356}, "quiet_kl": {"s8_register": 0.0003708549024304375}}, {"step": 2350, "train_bpb": {"s8_register": 0.023302323205672423}, "quiet_kl": {"s8_register": 0.0003598161722766235}}, {"step": 2400, "train_bpb": {"s8_register": 0.02359036219471576}, "quiet_kl": {"s8_register": 0.0003521157937939279}}, {"step": 2450, "train_bpb": {"s8_register": 0.023822357869129684}, "quiet_kl": {"s8_register": 0.00036448297440074386}}, {"step": 2500, "train_bpb": {"s8_register": 0.0230092801664239}, "quiet_kl": {"s8_register": 0.00036471968865953386}}, {"step": 2550, "train_bpb": {"s8_register": 0.023025160910043127}, "quiet_kl": {"s8_register": 0.00032843652385054154}}, {"step": 2600, "train_bpb": {"s8_register": 0.023455692604188973}, "quiet_kl": {"s8_register": 0.0003160517013748176}}, {"step": 2650, "train_bpb": {"s8_register": 0.023293207319008014}, "quiet_kl": {"s8_register": 0.0003819568126345985}}, {"step": 2700, "train_bpb": {"s8_register": 0.02265107539988636}, "quiet_kl": {"s8_register": 0.0003505235171178356}}, {"step": 2750, "train_bpb": {"s8_register": 0.023261377656771873}, "quiet_kl": {"s8_register": 0.0003481186632416211}}, {"step": 2800, "train_bpb": {"s8_register": 0.02232718608982488}, "quiet_kl": {"s8_register": 0.00031486601626966147}}, {"step": 2850, "train_bpb": {"s8_register": 0.022731435475882734}, "quiet_kl": {"s8_register": 0.00035111790872178973}}, {"step": 2900, "train_bpb": {"s8_register": 0.022832046562762782}, "quiet_kl": {"s8_register": 0.00037463288521394133}}, {"step": 2950, "train_bpb": {"s8_register": 0.024496762359558027}, "quiet_kl": {"s8_register": 0.0007187775388592854}}, {"step": 3000, "train_bpb": {"s8_register": 0.023655349897059943}, "quiet_kl": {"s8_register": 0.0004241743826423772}}, {"step": 3050, "train_bpb": {"s8_register": 0.023158773156606877}, "quiet_kl": {"s8_register": 0.00039935260952915995}}, {"step": 3100, "train_bpb": {"s8_register": 0.022681625672433824}, "quiet_kl": {"s8_register": 0.0003484048042446375}}, {"step": 3150, "train_bpb": {"s8_register": 0.022140507628935}, "quiet_kl": {"s8_register": 0.0003099649288924411}}, {"step": 3200, "train_bpb": {"s8_register": 0.02306839971323148}, "quiet_kl": {"s8_register": 0.00034806542011210694}}, {"step": 3250, "train_bpb": {"s8_register": 0.02269433207314704}, "quiet_kl": {"s8_register": 0.0003293591510737315}}, {"step": 3300, "train_bpb": {"s8_register": 0.022599522176058907}, "quiet_kl": {"s8_register": 0.00031632880360120907}}, {"step": 3350, "train_bpb": {"s8_register": 0.022824596623667162}, "quiet_kl": {"s8_register": 0.0003373481606831774}}, {"step": 3400, "train_bpb": {"s8_register": 0.02209795036074526}, "quiet_kl": {"s8_register": 0.00032100793527206406}}, {"step": 3450, "train_bpb": {"s8_register": 0.022739335310844096}, "quiet_kl": {"s8_register": 0.0003408048985875212}}, {"step": 3500, "train_bpb": {"s8_register": 0.022371650350017504}, "quiet_kl": {"s8_register": 0.00034481779584893957}}, {"step": 3550, "train_bpb": {"s8_register": 0.02235635399807145}, "quiet_kl": {"s8_register": 0.0003516048367600888}}, {"step": 3600, "train_bpb": {"s8_register": 0.022746635759394906}, "quiet_kl": {"s8_register": 0.00035183920786948877}}, {"step": 3650, "train_bpb": {"s8_register": 0.022704699456063945}, "quiet_kl": {"s8_register": 0.0003018930152757093}}, {"step": 3700, "train_bpb": {"s8_register": 0.022229632048308495}, "quiet_kl": {"s8_register": 0.0003285306057659909}}, {"step": 3750, "train_bpb": {"s8_register": 0.022059337084696432}, "quiet_kl": {"s8_register": 0.0003525336235179566}}, {"step": 3800, "train_bpb": {"s8_register": 0.022441210264879444}, "quiet_kl": {"s8_register": 0.0003436481370590627}}, {"step": 3850, "train_bpb": {"s8_register": 0.02298508462595353}, "quiet_kl": {"s8_register": 0.00041857310978230086}}, {"step": 3900, "train_bpb": {"s8_register": 0.02206442161740775}, "quiet_kl": {"s8_register": 0.0003460431046551093}}, {"step": 3950, "train_bpb": {"s8_register": 0.02239204440209657}, "quiet_kl": {"s8_register": 0.0003313261023140512}}, {"step": 4000, "train_bpb": {"s8_register": 0.022473403267337832}, "quiet_kl": {"s8_register": 0.0003418517005047761}}], "steps_done": {"s1_perspective": 3200, "s2_concept": 2400, "s3_rules": 800, "s4_arith": 800, "s5_causal": 4000, "s6_tryfail": 4000, "s7_mixed": 4000, "s8_register": 4000}, "stage_seconds": 3347.7979440689087, "inert_at_birth": true}], "seconds": 12471.75503730774, "log": ["[ 0.2s] THE JOB: gXB: the group ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail', 's7_mixed', 's8_register'], staged, EARLIER ARMS FROZEN (on, not training); 4000 steps per stage; each step = one task chunk + one abstention chunk of 2 x 4096 bytes per newest member; lambda 2.0; pure Adam lr 0.001; the compiled chain with recompute; trunk step 245,674; card 31.4 GB with 16.4 GB free; host RSS 0.6 GB; the first 4 members loaded from gRB:4 (the later members attach fresh); a member's partner rows come from every other member of the roster, attached or not", "[ 15.8s] mini-beatrix-3 @ 245,674: trunk loaded from /root/v3/runs/mission/mini-beatrix-3/checkpoints/step_00245674.safetensors and frozen (376.0M); amoe-lora 0.2.11, torch 2.8.0+cu128", "[ 79.5s] ARMS OFF (the bare trunk): s1_perspective rows 0.5538; s2_concept rows 0.7212; s3_rules rows 0.9296; s4_arith rows 1.2339; s5_causal rows 0.6604; s6_tryfail rows 0.6728; s7_mixed rows 0.7830; s8_register rows 0.4907; web rows 0.9506; s1_perspective battery in 0.550, transfer 0.463; s2_concept battery in 0.298, transfer 0.678; s3_rules battery in 0.155, transfer 0.070, held-out 0.050; s4_arith battery in 0.344, transfer 0.242, held-out 0.350; probes mean 0.411", "[ 84.2s] RESUMED from gRB at its close 4: ['s1_perspective', 's2_concept', 's3_rules', 's4_arith'] loaded (on; frozen while the later arms train); the next stage attaches over them", "[ 221.9s] CLOSE 4/8: the group ['s1_perspective', 's2_concept', 's3_rules', 's4_arith'] all on, against arms off:", "[ 221.9s] s1_perspective rows: off 0.5538 -> all on 0.0479 (-0.5060 +- 0.0048); with s1_perspective masked 0.5211 (its share -0.4733); s1_perspective alone 0.0522 (-0.5017); gate mean 0.212 / widest 0.555", "[ 221.9s] s2_concept rows: off 0.7212 -> all on 0.1212 (-0.6000 +- 0.0037); with s2_concept masked 0.2424 (its share -0.1212); s2_concept alone 0.3603 (-0.3608); gate mean 0.273 / widest 0.487", "[ 221.9s] s3_rules rows: off 0.9296 -> all on 0.2222 (-0.7074 +- 0.0054); with s3_rules masked 0.3181 (its share -0.0958); s3_rules alone 0.8702 (-0.0594); gate mean 0.124 / widest 0.175", "[ 221.9s] s4_arith rows: off 1.2339 -> all on 0.3379 (-0.8960 +- 0.0059); with s4_arith masked 0.5116 (its share -0.1737); s4_arith alone 1.1550 (-0.0789); gate mean 0.108 / widest 0.167", "[ 221.9s] web holdout: off 0.9506 -> all on 0.9517 (+0.00110 +- 0.00039; the quiet bar +.012); each member's share: s1_perspective +0.00004, s2_concept -0.00055, s3_rules +0.00038, s4_arith -0.00001; alone: s1_perspective +0.00134, s2_concept +0.00086, s3_rules +0.00034, s4_arith +0.00006", "[ 221.9s] a member's share on a PARTNER's rows (all on minus that member masked; near 0 = it stays out): s1_perspective on s2_concept -0.2380; s1_perspective on s3_rules -0.1951; s1_perspective on s4_arith -0.1591; s2_concept on s1_perspective -0.0041; s2_concept on s3_rules -0.1005; s2_concept on s4_arith -0.1138; s3_rules on s1_perspective +0.0001; s3_rules on s2_concept -0.0006; s3_rules on s4_arith -0.0003; s4_arith on s1_perspective +0.0000; s4_arith on s2_concept -0.0000; s4_arith on s3_rules +0.0000", "[ 221.9s] a member ALONE on a partner's rows (alone minus off): s1_perspective on s2_concept -0.4772; s1_perspective on s3_rules -0.4262; s1_perspective on s4_arith -0.4807; s2_concept on s1_perspective -0.0300; s2_concept on s3_rules -0.3138; s2_concept on s4_arith -0.3577; s3_rules on s1_perspective -0.0003; s3_rules on s2_concept +0.0016; s3_rules on s4_arith -0.0047; s4_arith on s1_perspective -0.0013; s4_arith on s2_concept -0.0015; s4_arith on s3_rules +0.0005", "[ 221.9s] the group on the rows of stages not yet attached: s5_causal -0.1748; s6_tryfail -0.1850; s7_mixed -0.0060; s8_register -0.1104", "[ 221.9s] s1_perspective battery accuracy, off -> all on: in 0.550 -> 1.000, transfer 0.463 -> 0.463; alone: in 0.925, transfer 0.463", "[ 221.9s] s2_concept battery accuracy, off -> all on: in 0.298 -> 0.952, transfer 0.678 -> 0.754; alone: in 0.905, transfer 0.703", "[ 221.9s] s3_rules battery accuracy, off -> all on: in 0.155 -> 0.885, transfer 0.070 -> 0.330, held-out 0.050 -> 0.483; alone: in 0.335, transfer 0.100, held-out 0.142", "[ 221.9s] s4_arith battery accuracy, off -> all on: in 0.344 -> 0.594, transfer 0.242 -> 0.271, held-out 0.350 -> 0.287; alone: in 0.412, transfer 0.237, held-out 0.400", "[ 221.9s] probes mean: off 0.411 -> all on 0.415", "[ 221.9s] compiled chain: 8 cache entries (limit 256), 0 limit warnings; card peak 3.0 GB; host RSS 4.6 GB", "[ 225.4s] STAGE 5/8: s5_causal attached over ['s1_perspective', 's2_concept', 's3_rules', 's4_arith'] (on, frozen); at birth the group reads exactly as before on its rows and on web text; 4000 steps of 1 task + 1 abstention chunks", "[ 235.3s] stage 5/8 s5_causal step 1/4000: train bpb s5 0.5003; quiet KL s5 0.00014; 9.95 s/step; the job: 3.9 min spent, about 2653.7 min of steps left; card peak 13.9 GB", "[ 268.5s] stage 5/8 s5_causal step 50/4000: train bpb s5 0.4748; quiet KL s5 0.00021; 0.86 s/step; the job: 4.5 min spent, about 229.4 min of steps left; card peak 14.0 GB", "[ 300.7s] stage 5/8 s5_causal step 100/4000: train bpb s5 0.4433; quiet KL s5 0.00058; 0.75 s/step; the job: 5.0 min spent, about 199.7 min of steps left; card peak 14.0 GB", "[ 332.8s] stage 5/8 s5_causal step 150/4000: train bpb s5 0.4122; quiet KL s5 0.00060; 0.72 s/step; the job: 5.5 min spent, about 189.2 min of steps left; card peak 14.0 GB", "[ 365.2s] stage 5/8 s5_causal step 200/4000: train bpb s5 0.3751; quiet KL s5 0.00061; 0.70 s/step; the job: 6.1 min spent, about 184.1 min of steps left; card peak 14.0 GB", "[ 397.3s] stage 5/8 s5_causal step 250/4000: train bpb s5 0.3346; quiet KL s5 0.00069; 0.69 s/step; the job: 6.6 min spent, about 180.5 min of steps left; card peak 14.0 GB", "[ 429.3s] stage 5/8 s5_causal step 300/4000: train bpb s5 0.2950; quiet KL s5 0.00084; 0.68 s/step; the job: 7.2 min spent, about 177.9 min of steps left; card peak 14.0 GB", "[ 461.5s] stage 5/8 s5_causal step 350/4000: train bpb s5 0.2538; quiet KL s5 0.00090; 0.67 s/step; the job: 7.7 min spent, about 176.0 min of steps left; card peak 14.0 GB", "[ 493.5s] stage 5/8 s5_causal step 400/4000: train bpb s5 0.2116; quiet KL s5 0.00085; 0.67 s/step; the job: 8.2 min spent, about 174.3 min of steps left; card peak 14.0 GB", "[ 525.5s] stage 5/8 s5_causal step 450/4000: train bpb s5 0.1766; quiet KL s5 0.00090; 0.67 s/step; the job: 8.8 min spent, about 172.9 min of steps left; card peak 14.0 GB", "[ 557.6s] stage 5/8 s5_causal step 500/4000: train bpb s5 0.1449; quiet KL s5 0.00088; 0.66 s/step; the job: 9.3 min spent, about 171.6 min of steps left; card peak 14.0 GB", "[ 589.6s] stage 5/8 s5_causal step 550/4000: train bpb s5 0.1175; quiet KL s5 0.00088; 0.66 s/step; the job: 9.8 min spent, about 170.5 min of steps left; card peak 14.0 GB", "[ 621.6s] stage 5/8 s5_causal step 600/4000: train bpb s5 0.0999; quiet KL s5 0.00083; 0.66 s/step; the job: 10.4 min spent, about 169.5 min of steps left; card peak 14.0 GB", "[ 653.8s] stage 5/8 s5_causal step 650/4000: train bpb s5 0.0838; quiet KL s5 0.00082; 0.66 s/step; the job: 10.9 min spent, about 168.6 min of steps left; card peak 14.0 GB", "[ 685.8s] stage 5/8 s5_causal step 700/4000: train bpb s5 0.0732; quiet KL s5 0.00075; 0.66 s/step; the job: 11.4 min spent, about 167.7 min of steps left; card peak 14.0 GB", "[ 717.9s] stage 5/8 s5_causal step 750/4000: train bpb s5 0.0662; quiet KL s5 0.00073; 0.66 s/step; the job: 12.0 min spent, about 166.9 min of steps left; card peak 14.0 GB", "[ 750.2s] stage 5/8 s5_causal step 800/4000: train bpb s5 0.0610; quiet KL s5 0.00070; 0.66 s/step; the job: 12.5 min spent, about 166.2 min of steps left; card peak 14.0 GB", "[ 782.4s] stage 5/8 s5_causal step 850/4000: train bpb s5 0.0570; quiet KL s5 0.00063; 0.66 s/step; the job: 13.0 min spent, about 165.5 min of steps left; card peak 14.0 GB", "[ 814.2s] stage 5/8 s5_causal step 900/4000: train bpb s5 0.0549; quiet KL s5 0.00067; 0.65 s/step; the job: 13.6 min spent, about 164.7 min of steps left; card peak 14.0 GB", "[ 846.1s] stage 5/8 s5_causal step 950/4000: train bpb s5 0.0533; quiet KL s5 0.00061; 0.65 s/step; the job: 14.1 min spent, about 163.9 min of steps left; card peak 14.0 GB", "[ 878.2s] stage 5/8 s5_causal step 1000/4000: train bpb s5 0.0513; quiet KL s5 0.00056; 0.65 s/step; the job: 14.6 min spent, about 163.2 min of steps left; card peak 14.0 GB", "[ 910.1s] stage 5/8 s5_causal step 1050/4000: train bpb s5 0.0503; quiet KL s5 0.00057; 0.65 s/step; the job: 15.2 min spent, about 162.5 min of steps left; card peak 14.0 GB", "[ 942.2s] stage 5/8 s5_causal step 1100/4000: train bpb s5 0.0491; quiet KL s5 0.00051; 0.65 s/step; the job: 15.7 min spent, about 161.8 min of steps left; card peak 14.0 GB", "[ 974.6s] stage 5/8 s5_causal step 1150/4000: train bpb s5 0.0489; quiet KL s5 0.00049; 0.65 s/step; the job: 16.2 min spent, about 161.2 min of steps left; card peak 14.0 GB", "[ 1006.7s] stage 5/8 s5_causal step 1200/4000: train bpb s5 0.0481; quiet KL s5 0.00047; 0.65 s/step; the job: 16.8 min spent, about 160.6 min of steps left; card peak 14.0 GB", "[ 1039.1s] stage 5/8 s5_causal step 1250/4000: train bpb s5 0.0479; quiet KL s5 0.00048; 0.65 s/step; the job: 17.3 min spent, about 160.0 min of steps left; card peak 14.0 GB", "[ 1071.2s] stage 5/8 s5_causal step 1300/4000: train bpb s5 0.0477; quiet KL s5 0.00050; 0.65 s/step; the job: 17.9 min spent, about 159.4 min of steps left; card peak 14.0 GB", "[ 1103.3s] stage 5/8 s5_causal step 1350/4000: train bpb s5 0.0472; quiet KL s5 0.00046; 0.65 s/step; the job: 18.4 min spent, about 158.8 min of steps left; card peak 14.0 GB", "[ 1135.4s] stage 5/8 s5_causal step 1400/4000: train bpb s5 0.0468; quiet KL s5 0.00044; 0.65 s/step; the job: 18.9 min spent, about 158.2 min of steps left; card peak 14.0 GB", "[ 1167.5s] stage 5/8 s5_causal step 1450/4000: train bpb s5 0.0464; quiet KL s5 0.00043; 0.65 s/step; the job: 19.5 min spent, about 157.6 min of steps left; card peak 14.0 GB", "[ 1199.4s] stage 5/8 s5_causal step 1500/4000: train bpb s5 0.0459; quiet KL s5 0.00041; 0.65 s/step; the job: 20.0 min spent, about 156.9 min of steps left; card peak 14.0 GB", "[ 1231.6s] stage 5/8 s5_causal step 1550/4000: train bpb s5 0.0457; quiet KL s5 0.00039; 0.65 s/step; the job: 20.5 min spent, about 156.3 min of steps left; card peak 14.0 GB", "[ 1264.0s] stage 5/8 s5_causal step 1600/4000: train bpb s5 0.0456; quiet KL s5 0.00038; 0.65 s/step; the job: 21.1 min spent, about 155.8 min of steps left; card peak 14.0 GB", "[ 1296.0s] stage 5/8 s5_causal step 1650/4000: train bpb s5 0.0453; quiet KL s5 0.00039; 0.65 s/step; the job: 21.6 min spent, about 155.2 min of steps left; card peak 14.0 GB", "[ 1328.1s] stage 5/8 s5_causal step 1700/4000: train bpb s5 0.0453; quiet KL s5 0.00041; 0.65 s/step; the job: 22.1 min spent, about 154.6 min of steps left; card peak 14.0 GB", "[ 1360.1s] stage 5/8 s5_causal step 1750/4000: train bpb s5 0.0453; quiet KL s5 0.00038; 0.65 s/step; the job: 22.7 min spent, about 154.0 min of steps left; card peak 14.0 GB", "[ 1392.1s] stage 5/8 s5_causal step 1800/4000: train bpb s5 0.0449; quiet KL s5 0.00036; 0.65 s/step; the job: 23.2 min spent, about 153.4 min of steps left; card peak 14.0 GB", "[ 1424.2s] stage 5/8 s5_causal step 1850/4000: train bpb s5 0.0452; quiet KL s5 0.00048; 0.65 s/step; the job: 23.7 min spent, about 152.8 min of steps left; card peak 14.0 GB", "[ 1456.5s] stage 5/8 s5_causal step 1900/4000: train bpb s5 0.0447; quiet KL s5 0.00040; 0.65 s/step; the job: 24.3 min spent, about 152.3 min of steps left; card peak 14.0 GB", "[ 1488.7s] stage 5/8 s5_causal step 1950/4000: train bpb s5 0.0444; quiet KL s5 0.00038; 0.65 s/step; the job: 24.8 min spent, about 151.7 min of steps left; card peak 14.0 GB", "[ 1520.8s] stage 5/8 s5_causal step 2000/4000: train bpb s5 0.0448; quiet KL s5 0.00034; 0.65 s/step; the job: 25.3 min spent, about 151.1 min of steps left; card peak 14.0 GB", "[ 1553.0s] stage 5/8 s5_causal step 2050/4000: train bpb s5 0.0442; quiet KL s5 0.00034; 0.65 s/step; the job: 25.9 min spent, about 150.6 min of steps left; card peak 14.0 GB", "[ 1585.1s] stage 5/8 s5_causal step 2100/4000: train bpb s5 0.0445; quiet KL s5 0.00037; 0.65 s/step; the job: 26.4 min spent, about 150.0 min of steps left; card peak 14.0 GB", "[ 1617.3s] stage 5/8 s5_causal step 2150/4000: train bpb s5 0.0441; quiet KL s5 0.00035; 0.65 s/step; the job: 27.0 min spent, about 149.4 min of steps left; card peak 14.0 GB", "[ 1650.3s] stage 5/8 s5_causal step 2200/4000: train bpb s5 0.0439; quiet KL s5 0.00035; 0.65 s/step; the job: 27.5 min spent, about 149.0 min of steps left; card peak 14.0 GB", "[ 1682.7s] stage 5/8 s5_causal step 2250/4000: train bpb s5 0.0449; quiet KL s5 0.00036; 0.65 s/step; the job: 28.0 min spent, about 148.4 min of steps left; card peak 14.0 GB", "[ 1714.6s] stage 5/8 s5_causal step 2300/4000: train bpb s5 0.0445; quiet KL s5 0.00036; 0.65 s/step; the job: 28.6 min spent, about 147.8 min of steps left; card peak 14.0 GB", "[ 1746.7s] stage 5/8 s5_causal step 2350/4000: train bpb s5 0.0440; quiet KL s5 0.00035; 0.65 s/step; the job: 29.1 min spent, about 147.3 min of steps left; card peak 14.0 GB", "[ 1778.8s] stage 5/8 s5_causal step 2400/4000: train bpb s5 0.0441; quiet KL s5 0.00037; 0.65 s/step; the job: 29.6 min spent, about 146.7 min of steps left; card peak 14.0 GB", "[ 1811.0s] stage 5/8 s5_causal step 2450/4000: train bpb s5 0.0438; quiet KL s5 0.00032; 0.65 s/step; the job: 30.2 min spent, about 146.2 min of steps left; card peak 14.0 GB", "[ 1843.4s] stage 5/8 s5_causal step 2500/4000: train bpb s5 0.0439; quiet KL s5 0.00033; 0.65 s/step; the job: 30.7 min spent, about 145.6 min of steps left; card peak 14.0 GB", "[ 1875.5s] stage 5/8 s5_causal step 2550/4000: train bpb s5 0.0437; quiet KL s5 0.00036; 0.65 s/step; the job: 31.3 min spent, about 145.1 min of steps left; card peak 14.0 GB", "[ 1907.7s] stage 5/8 s5_causal step 2600/4000: train bpb s5 0.0438; quiet KL s5 0.00033; 0.65 s/step; the job: 31.8 min spent, about 144.5 min of steps left; card peak 14.0 GB", "[ 1940.0s] stage 5/8 s5_causal step 2650/4000: train bpb s5 0.0440; quiet KL s5 0.00035; 0.65 s/step; the job: 32.3 min spent, about 144.0 min of steps left; card peak 14.0 GB", "[ 1972.2s] stage 5/8 s5_causal step 2700/4000: train bpb s5 0.0438; quiet KL s5 0.00037; 0.65 s/step; the job: 32.9 min spent, about 143.4 min of steps left; card peak 14.0 GB", "[ 2004.5s] stage 5/8 s5_causal step 2750/4000: train bpb s5 0.0435; quiet KL s5 0.00033; 0.65 s/step; the job: 33.4 min spent, about 142.9 min of steps left; card peak 14.0 GB", "[ 2036.9s] stage 5/8 s5_causal step 2800/4000: train bpb s5 0.0435; quiet KL s5 0.00033; 0.65 s/step; the job: 33.9 min spent, about 142.3 min of steps left; card peak 14.0 GB", "[ 2069.1s] stage 5/8 s5_causal step 2850/4000: train bpb s5 0.0436; quiet KL s5 0.00032; 0.65 s/step; the job: 34.5 min spent, about 141.8 min of steps left; card peak 14.0 GB", "[ 2101.3s] stage 5/8 s5_causal step 2900/4000: train bpb s5 0.0433; quiet KL s5 0.00031; 0.65 s/step; the job: 35.0 min spent, about 141.2 min of steps left; card peak 14.0 GB", "[ 2133.7s] stage 5/8 s5_causal step 2950/4000: train bpb s5 0.0433; quiet KL s5 0.00031; 0.65 s/step; the job: 35.6 min spent, about 140.7 min of steps left; card peak 14.0 GB", "[ 2165.9s] stage 5/8 s5_causal step 3000/4000: train bpb s5 0.0433; quiet KL s5 0.00032; 0.65 s/step; the job: 36.1 min spent, about 140.2 min of steps left; card peak 14.0 GB", "[ 2198.0s] stage 5/8 s5_causal step 3050/4000: train bpb s5 0.0432; quiet KL s5 0.00033; 0.65 s/step; the job: 36.6 min spent, about 139.6 min of steps left; card peak 14.0 GB", "[ 2230.3s] stage 5/8 s5_causal step 3100/4000: train bpb s5 0.0437; quiet KL s5 0.00040; 0.65 s/step; the job: 37.2 min spent, about 139.1 min of steps left; card peak 14.0 GB", "[ 2262.5s] stage 5/8 s5_causal step 3150/4000: train bpb s5 0.0436; quiet KL s5 0.00035; 0.65 s/step; the job: 37.7 min spent, about 138.5 min of steps left; card peak 14.0 GB", "[ 2294.8s] stage 5/8 s5_causal step 3200/4000: train bpb s5 0.0433; quiet KL s5 0.00034; 0.65 s/step; the job: 38.2 min spent, about 138.0 min of steps left; card peak 14.0 GB", "[ 2327.2s] stage 5/8 s5_causal step 3250/4000: train bpb s5 0.0430; quiet KL s5 0.00032; 0.65 s/step; the job: 38.8 min spent, about 137.4 min of steps left; card peak 14.0 GB", "[ 2359.5s] stage 5/8 s5_causal step 3300/4000: train bpb s5 0.0429; quiet KL s5 0.00032; 0.65 s/step; the job: 39.3 min spent, about 136.9 min of steps left; card peak 14.0 GB", "[ 2391.9s] stage 5/8 s5_causal step 3350/4000: train bpb s5 0.0428; quiet KL s5 0.00030; 0.65 s/step; the job: 39.9 min spent, about 136.4 min of steps left; card peak 14.0 GB", "[ 2424.3s] stage 5/8 s5_causal step 3400/4000: train bpb s5 0.0432; quiet KL s5 0.00031; 0.65 s/step; the job: 40.4 min spent, about 135.8 min of steps left; card peak 14.0 GB", "[ 2456.6s] stage 5/8 s5_causal step 3450/4000: train bpb s5 0.0428; quiet KL s5 0.00028; 0.65 s/step; the job: 40.9 min spent, about 135.3 min of steps left; card peak 14.0 GB", "[ 2488.8s] stage 5/8 s5_causal step 3500/4000: train bpb s5 0.0428; quiet KL s5 0.00029; 0.65 s/step; the job: 41.5 min spent, about 134.7 min of steps left; card peak 14.0 GB", "[ 2521.1s] stage 5/8 s5_causal step 3550/4000: train bpb s5 0.0429; quiet KL s5 0.00033; 0.65 s/step; the job: 42.0 min spent, about 134.2 min of steps left; card peak 14.0 GB", "[ 2553.1s] stage 5/8 s5_causal step 3600/4000: train bpb s5 0.0430; quiet KL s5 0.00033; 0.65 s/step; the job: 42.6 min spent, about 133.6 min of steps left; card peak 14.0 GB", "[ 2585.4s] stage 5/8 s5_causal step 3650/4000: train bpb s5 0.0430; quiet KL s5 0.00032; 0.65 s/step; the job: 43.1 min spent, about 133.1 min of steps left; card peak 14.0 GB", "[ 2617.8s] stage 5/8 s5_causal step 3700/4000: train bpb s5 0.0429; quiet KL s5 0.00033; 0.65 s/step; the job: 43.6 min spent, about 132.6 min of steps left; card peak 14.0 GB", "[ 2649.8s] stage 5/8 s5_causal step 3750/4000: train bpb s5 0.0428; quiet KL s5 0.00030; 0.65 s/step; the job: 44.2 min spent, about 132.0 min of steps left; card peak 14.0 GB", "[ 2682.0s] stage 5/8 s5_causal step 3800/4000: train bpb s5 0.0428; quiet KL s5 0.00030; 0.65 s/step; the job: 44.7 min spent, about 131.4 min of steps left; card peak 14.0 GB", "[ 2714.3s] stage 5/8 s5_causal step 3850/4000: train bpb s5 0.0425; quiet KL s5 0.00030; 0.65 s/step; the job: 45.2 min spent, about 130.9 min of steps left; card peak 14.0 GB", "[ 2746.4s] stage 5/8 s5_causal step 3900/4000: train bpb s5 0.0426; quiet KL s5 0.00032; 0.65 s/step; the job: 45.8 min spent, about 130.4 min of steps left; card peak 14.0 GB", "[ 2778.7s] stage 5/8 s5_causal step 3950/4000: train bpb s5 0.0426; quiet KL s5 0.00031; 0.65 s/step; the job: 46.3 min spent, about 129.8 min of steps left; card peak 14.0 GB", "[ 2811.1s] stage 5/8 s5_causal step 4000/4000: train bpb s5 0.0425; quiet KL s5 0.00030; 0.65 s/step; the job: 46.9 min spent, about 129.3 min of steps left; card peak 14.0 GB", "[ 2811.9s] anchors saved: /root/refit/group/anchors/gXB_close5_.safetensors for ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal']", "[ 2974.8s] CLOSE 5/8: the group ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal'] all on, against arms off:", "[ 2974.8s] s1_perspective rows: off 0.5538 -> all on 0.0479 (-0.5059 +- 0.0048); with s1_perspective masked 0.5187 (its share -0.4708); s1_perspective alone 0.0522 (-0.5017); gate mean 0.212 / widest 0.555", "[ 2974.8s] s2_concept rows: off 0.7212 -> all on 0.1212 (-0.6000 +- 0.0037); with s2_concept masked 0.2421 (its share -0.1209); s2_concept alone 0.3603 (-0.3608); gate mean 0.273 / widest 0.487", "[ 2974.8s] s3_rules rows: off 0.9296 -> all on 0.2226 (-0.7070 +- 0.0054); with s3_rules masked 0.3178 (its share -0.0952); s3_rules alone 0.8702 (-0.0594); gate mean 0.124 / widest 0.175", "[ 2974.8s] s4_arith rows: off 1.2339 -> all on 0.3379 (-0.8959 +- 0.0058); with s4_arith masked 0.5107 (its share -0.1728); s4_arith alone 1.1550 (-0.0789); gate mean 0.108 / widest 0.167", "[ 2974.8s] s5_causal rows: off 0.6604 -> all on 0.0426 (-0.6179 +- 0.0069); with s5_causal masked 0.4856 (its share -0.4430); s5_causal alone 0.1418 (-0.5187); gate mean 0.186 / widest 0.360", "[ 2974.8s] web holdout: off 0.9506 -> all on 0.9520 (+0.00138 +- 0.00042; the quiet bar +.012); each member's share: s1_perspective +0.00024, s2_concept -0.00036, s3_rules +0.00040, s4_arith +0.00021, s5_causal +0.00028; alone: s1_perspective +0.00134, s2_concept +0.00086, s3_rules +0.00034, s4_arith +0.00006, s5_causal +0.00000", "[ 2974.8s] a member's share on a PARTNER's rows (all on minus that member masked; near 0 = it stays out): s1_perspective on s2_concept -0.2370; s1_perspective on s3_rules -0.1968; s1_perspective on s4_arith -0.1597; s1_perspective on s5_causal -0.0428; s2_concept on s1_perspective -0.0042; s2_concept on s3_rules -0.1007; s2_concept on s4_arith -0.1135; s2_concept on s5_causal -0.0008; s3_rules on s1_perspective +0.0001; s3_rules on s2_concept -0.0006; s3_rules on s4_arith -0.0002; s3_rules on s5_causal -0.0001; s4_arith on s1_perspective +0.0001; s4_arith on s2_concept -0.0001; s4_arith on s3_rules +0.0002; s4_arith on s5_causal -0.0001; s5_causal on s1_perspective +0.0001; s5_causal on s2_concept +0.0000; s5_causal on s3_rules +0.0004; s5_causal on s4_arith +0.0001", "[ 2974.8s] a member ALONE on a partner's rows (alone minus off): s1_perspective on s2_concept -0.4772; s1_perspective on s3_rules -0.4262; s1_perspective on s4_arith -0.4807; s1_perspective on s5_causal -0.1919; s2_concept on s1_perspective -0.0300; s2_concept on s3_rules -0.3138; s2_concept on s4_arith -0.3577; s2_concept on s5_causal -0.0127; s3_rules on s1_perspective -0.0003; s3_rules on s2_concept +0.0016; s3_rules on s4_arith -0.0047; s3_rules on s5_causal +0.0060; s4_arith on s1_perspective -0.0013; s4_arith on s2_concept -0.0015; s4_arith on s3_rules +0.0005; s4_arith on s5_causal -0.0011; s5_causal on s1_perspective -0.0022; s5_causal on s2_concept -0.0010; s5_causal on s3_rules -0.0018; s5_causal on s4_arith -0.0022", "[ 2974.8s] the group on the rows of stages not yet attached: s6_tryfail -0.1851; s7_mixed -0.0062; s8_register -0.1113", "[ 2974.8s] s1_perspective battery accuracy, off -> all on: in 0.550 -> 1.000, transfer 0.463 -> 0.463; alone: in 0.925, transfer 0.463", "[ 2974.8s] s2_concept battery accuracy, off -> all on: in 0.298 -> 0.952, transfer 0.678 -> 0.780; alone: in 0.905, transfer 0.703", "[ 2974.8s] s3_rules battery accuracy, off -> all on: in 0.155 -> 0.880, transfer 0.070 -> 0.330, held-out 0.050 -> 0.475; alone: in 0.335, transfer 0.100, held-out 0.142", "[ 2974.8s] s4_arith battery accuracy, off -> all on: in 0.344 -> 0.594, transfer 0.242 -> 0.275, held-out 0.350 -> 0.287; alone: in 0.412, transfer 0.237, held-out 0.400", "[ 2974.8s] probes mean: off 0.411 -> all on 0.415", "[ 2974.8s] compiled chain: 13 cache entries (limit 256), 0 limit warnings; card peak 14.0 GB; host RSS 5.4 GB", "[ 2978.5s] STAGE 6/8: s6_tryfail attached over ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal'] (on, frozen); at birth the group reads exactly as before on its rows and on web text; 4000 steps of 1 task + 1 abstention chunks", "[ 2980.1s] stage 6/8 s6_tryfail step 1/4000: train bpb s6 0.4916; quiet KL s6 0.00017; 1.60 s/step; the job: 49.7 min spent, about 129.3 min of steps left; card peak 14.0 GB", "[ 3014.0s] stage 6/8 s6_tryfail step 50/4000: train bpb s6 0.4851; quiet KL s6 0.00020; 0.71 s/step; the job: 50.2 min spent, about 128.9 min of steps left; card peak 14.1 GB", "[ 3048.4s] stage 6/8 s6_tryfail step 100/4000: train bpb s6 0.4561; quiet KL s6 0.00042; 0.70 s/step; the job: 50.8 min spent, about 128.5 min of steps left; card peak 14.1 GB", "[ 3082.7s] stage 6/8 s6_tryfail step 150/4000: train bpb s6 0.4148; quiet KL s6 0.00052; 0.70 s/step; the job: 51.4 min spent, about 128.0 min of steps left; card peak 14.1 GB", "[ 3117.4s] stage 6/8 s6_tryfail step 200/4000: train bpb s6 0.3716; quiet KL s6 0.00069; 0.69 s/step; the job: 52.0 min spent, about 127.6 min of steps left; card peak 14.1 GB", "[ 3152.0s] stage 6/8 s6_tryfail step 250/4000: train bpb s6 0.3276; quiet KL s6 0.00079; 0.69 s/step; the job: 52.5 min spent, about 127.1 min of steps left; card peak 14.1 GB", "[ 3186.3s] stage 6/8 s6_tryfail step 300/4000: train bpb s6 0.2863; quiet KL s6 0.00081; 0.69 s/step; the job: 53.1 min spent, about 126.7 min of steps left; card peak 14.1 GB", "[ 3220.9s] stage 6/8 s6_tryfail step 350/4000: train bpb s6 0.2533; quiet KL s6 0.00088; 0.69 s/step; the job: 53.7 min spent, about 126.2 min of steps left; card peak 14.1 GB", "[ 3255.6s] stage 6/8 s6_tryfail step 400/4000: train bpb s6 0.2208; quiet KL s6 0.00072; 0.69 s/step; the job: 54.3 min spent, about 125.8 min of steps left; card peak 14.1 GB", "[ 3290.0s] stage 6/8 s6_tryfail step 450/4000: train bpb s6 0.1977; quiet KL s6 0.00064; 0.69 s/step; the job: 54.8 min spent, about 125.3 min of steps left; card peak 14.1 GB", "[ 3324.5s] stage 6/8 s6_tryfail step 500/4000: train bpb s6 0.1729; quiet KL s6 0.00067; 0.69 s/step; the job: 55.4 min spent, about 124.9 min of steps left; card peak 14.1 GB", "[ 3358.8s] stage 6/8 s6_tryfail step 550/4000: train bpb s6 0.1565; quiet KL s6 0.00076; 0.69 s/step; the job: 56.0 min spent, about 124.4 min of steps left; card peak 14.1 GB", "[ 3393.1s] stage 6/8 s6_tryfail step 600/4000: train bpb s6 0.1378; quiet KL s6 0.00071; 0.69 s/step; the job: 56.6 min spent, about 123.9 min of steps left; card peak 14.1 GB", "[ 3427.5s] stage 6/8 s6_tryfail step 650/4000: train bpb s6 0.1256; quiet KL s6 0.00067; 0.69 s/step; the job: 57.1 min spent, about 123.5 min of steps left; card peak 14.1 GB", "[ 3462.1s] stage 6/8 s6_tryfail step 700/4000: train bpb s6 0.1160; quiet KL s6 0.00073; 0.69 s/step; the job: 57.7 min spent, about 123.0 min of steps left; card peak 14.1 GB", "[ 3496.5s] stage 6/8 s6_tryfail step 750/4000: train bpb s6 0.1070; quiet KL s6 0.00074; 0.69 s/step; the job: 58.3 min spent, about 122.5 min of steps left; card peak 14.1 GB", "[ 3531.0s] stage 6/8 s6_tryfail step 800/4000: train bpb s6 0.1000; quiet KL s6 0.00080; 0.69 s/step; the job: 58.8 min spent, about 122.0 min of steps left; card peak 14.1 GB", "[ 3565.3s] stage 6/8 s6_tryfail step 850/4000: train bpb s6 0.0933; quiet KL s6 0.00076; 0.69 s/step; the job: 59.4 min spent, about 121.6 min of steps left; card peak 14.1 GB", "[ 3599.9s] stage 6/8 s6_tryfail step 900/4000: train bpb s6 0.0891; quiet KL s6 0.00073; 0.69 s/step; the job: 60.0 min spent, about 121.1 min of steps left; card peak 14.1 GB", "[ 3634.3s] stage 6/8 s6_tryfail step 950/4000: train bpb s6 0.0876; quiet KL s6 0.00060; 0.69 s/step; the job: 60.6 min spent, about 120.6 min of steps left; card peak 14.1 GB", "[ 3669.0s] stage 6/8 s6_tryfail step 1000/4000: train bpb s6 0.0855; quiet KL s6 0.00060; 0.69 s/step; the job: 61.1 min spent, about 120.1 min of steps left; card peak 14.1 GB", "[ 3703.5s] stage 6/8 s6_tryfail step 1050/4000: train bpb s6 0.0852; quiet KL s6 0.00064; 0.69 s/step; the job: 61.7 min spent, about 119.6 min of steps left; card peak 14.1 GB", "[ 3738.0s] stage 6/8 s6_tryfail step 1100/4000: train bpb s6 0.0812; quiet KL s6 0.00059; 0.69 s/step; the job: 62.3 min spent, about 119.2 min of steps left; card peak 14.1 GB", "[ 3772.5s] stage 6/8 s6_tryfail step 1150/4000: train bpb s6 0.0794; quiet KL s6 0.00053; 0.69 s/step; the job: 62.9 min spent, about 118.7 min of steps left; card peak 14.1 GB", "[ 3807.1s] stage 6/8 s6_tryfail step 1200/4000: train bpb s6 0.0786; quiet KL s6 0.00060; 0.69 s/step; the job: 63.5 min spent, about 118.2 min of steps left; card peak 14.1 GB", "[ 3841.5s] stage 6/8 s6_tryfail step 1250/4000: train bpb s6 0.0742; quiet KL s6 0.00050; 0.69 s/step; the job: 64.0 min spent, about 117.7 min of steps left; card peak 14.1 GB", "[ 3876.0s] stage 6/8 s6_tryfail step 1300/4000: train bpb s6 0.0750; quiet KL s6 0.00054; 0.69 s/step; the job: 64.6 min spent, about 117.2 min of steps left; card peak 14.1 GB", "[ 3910.6s] stage 6/8 s6_tryfail step 1350/4000: train bpb s6 0.0781; quiet KL s6 0.00061; 0.69 s/step; the job: 65.2 min spent, about 116.7 min of steps left; card peak 14.1 GB", "[ 3945.0s] stage 6/8 s6_tryfail step 1400/4000: train bpb s6 0.0748; quiet KL s6 0.00054; 0.69 s/step; the job: 65.7 min spent, about 116.2 min of steps left; card peak 14.1 GB", "[ 3979.5s] stage 6/8 s6_tryfail step 1450/4000: train bpb s6 0.0718; quiet KL s6 0.00052; 0.69 s/step; the job: 66.3 min spent, about 115.7 min of steps left; card peak 14.1 GB", "[ 4014.0s] stage 6/8 s6_tryfail step 1500/4000: train bpb s6 0.0745; quiet KL s6 0.00051; 0.69 s/step; the job: 66.9 min spent, about 115.2 min of steps left; card peak 14.1 GB", "[ 4048.5s] stage 6/8 s6_tryfail step 1550/4000: train bpb s6 0.0728; quiet KL s6 0.00051; 0.69 s/step; the job: 67.5 min spent, about 114.7 min of steps left; card peak 14.1 GB", "[ 4082.9s] stage 6/8 s6_tryfail step 1600/4000: train bpb s6 0.0721; quiet KL s6 0.00050; 0.69 s/step; the job: 68.0 min spent, about 114.2 min of steps left; card peak 14.1 GB", "[ 4117.9s] stage 6/8 s6_tryfail step 1650/4000: train bpb s6 0.0734; quiet KL s6 0.00045; 0.69 s/step; the job: 68.6 min spent, about 113.7 min of steps left; card peak 14.1 GB", "[ 4152.6s] stage 6/8 s6_tryfail step 1700/4000: train bpb s6 0.0684; quiet KL s6 0.00048; 0.69 s/step; the job: 69.2 min spent, about 113.2 min of steps left; card peak 14.1 GB", "[ 4187.0s] stage 6/8 s6_tryfail step 1750/4000: train bpb s6 0.0703; quiet KL s6 0.00050; 0.69 s/step; the job: 69.8 min spent, about 112.7 min of steps left; card peak 14.1 GB", "[ 4221.9s] stage 6/8 s6_tryfail step 1800/4000: train bpb s6 0.0681; quiet KL s6 0.00051; 0.69 s/step; the job: 70.4 min spent, about 112.2 min of steps left; card peak 14.1 GB", "[ 4257.4s] stage 6/8 s6_tryfail step 1850/4000: train bpb s6 0.0702; quiet KL s6 0.00048; 0.69 s/step; the job: 71.0 min spent, about 111.8 min of steps left; card peak 14.1 GB", "[ 4294.3s] stage 6/8 s6_tryfail step 1900/4000: train bpb s6 0.0676; quiet KL s6 0.00046; 0.69 s/step; the job: 71.6 min spent, about 111.3 min of steps left; card peak 14.1 GB", "[ 4329.6s] stage 6/8 s6_tryfail step 1950/4000: train bpb s6 0.0682; quiet KL s6 0.00049; 0.69 s/step; the job: 72.2 min spent, about 110.8 min of steps left; card peak 14.1 GB", "[ 4364.6s] stage 6/8 s6_tryfail step 2000/4000: train bpb s6 0.0688; quiet KL s6 0.00049; 0.69 s/step; the job: 72.7 min spent, about 110.3 min of steps left; card peak 14.1 GB", "[ 4399.0s] stage 6/8 s6_tryfail step 2050/4000: train bpb s6 0.0677; quiet KL s6 0.00045; 0.69 s/step; the job: 73.3 min spent, about 109.8 min of steps left; card peak 14.1 GB", "[ 4433.7s] stage 6/8 s6_tryfail step 2100/4000: train bpb s6 0.0646; quiet KL s6 0.00051; 0.69 s/step; the job: 73.9 min spent, about 109.3 min of steps left; card peak 14.1 GB", "[ 4468.3s] stage 6/8 s6_tryfail step 2150/4000: train bpb s6 0.0658; quiet KL s6 0.00046; 0.69 s/step; the job: 74.5 min spent, about 108.8 min of steps left; card peak 14.1 GB", "[ 4504.2s] stage 6/8 s6_tryfail step 2200/4000: train bpb s6 0.0663; quiet KL s6 0.00051; 0.69 s/step; the job: 75.1 min spent, about 108.3 min of steps left; card peak 14.1 GB", "[ 4540.6s] stage 6/8 s6_tryfail step 2250/4000: train bpb s6 0.0638; quiet KL s6 0.00042; 0.69 s/step; the job: 75.7 min spent, about 107.8 min of steps left; card peak 14.1 GB", "[ 4577.2s] stage 6/8 s6_tryfail step 2300/4000: train bpb s6 0.0629; quiet KL s6 0.00047; 0.70 s/step; the job: 76.3 min spent, about 107.4 min of steps left; card peak 14.1 GB", "[ 4613.7s] stage 6/8 s6_tryfail step 2350/4000: train bpb s6 0.0646; quiet KL s6 0.00044; 0.70 s/step; the job: 76.9 min spent, about 106.9 min of steps left; card peak 14.1 GB", "[ 4649.8s] stage 6/8 s6_tryfail step 2400/4000: train bpb s6 0.0642; quiet KL s6 0.00045; 0.70 s/step; the job: 77.5 min spent, about 106.4 min of steps left; card peak 14.1 GB", "[ 4686.1s] stage 6/8 s6_tryfail step 2450/4000: train bpb s6 0.0607; quiet KL s6 0.00044; 0.70 s/step; the job: 78.1 min spent, about 105.9 min of steps left; card peak 14.1 GB", "[ 4722.1s] stage 6/8 s6_tryfail step 2500/4000: train bpb s6 0.0641; quiet KL s6 0.00043; 0.70 s/step; the job: 78.7 min spent, about 105.5 min of steps left; card peak 14.1 GB", "[ 4756.2s] stage 6/8 s6_tryfail step 2550/4000: train bpb s6 0.0633; quiet KL s6 0.00039; 0.70 s/step; the job: 79.3 min spent, about 104.9 min of steps left; card peak 14.1 GB", "[ 4790.6s] stage 6/8 s6_tryfail step 2600/4000: train bpb s6 0.0645; quiet KL s6 0.00040; 0.70 s/step; the job: 79.8 min spent, about 104.4 min of steps left; card peak 14.1 GB", "[ 4824.9s] stage 6/8 s6_tryfail step 2650/4000: train bpb s6 0.0638; quiet KL s6 0.00041; 0.70 s/step; the job: 80.4 min spent, about 103.9 min of steps left; card peak 14.1 GB", "[ 4858.9s] stage 6/8 s6_tryfail step 2700/4000: train bpb s6 0.0614; quiet KL s6 0.00041; 0.70 s/step; the job: 81.0 min spent, about 103.3 min of steps left; card peak 14.1 GB", "[ 4893.3s] stage 6/8 s6_tryfail step 2750/4000: train bpb s6 0.0629; quiet KL s6 0.00044; 0.70 s/step; the job: 81.6 min spent, about 102.8 min of steps left; card peak 14.1 GB", "[ 4927.5s] stage 6/8 s6_tryfail step 2800/4000: train bpb s6 0.0624; quiet KL s6 0.00042; 0.70 s/step; the job: 82.1 min spent, about 102.3 min of steps left; card peak 14.1 GB", "[ 4962.0s] stage 6/8 s6_tryfail step 2850/4000: train bpb s6 0.0639; quiet KL s6 0.00041; 0.70 s/step; the job: 82.7 min spent, about 101.7 min of steps left; card peak 14.1 GB", "[ 4996.1s] stage 6/8 s6_tryfail step 2900/4000: train bpb s6 0.0613; quiet KL s6 0.00043; 0.70 s/step; the job: 83.3 min spent, about 101.2 min of steps left; card peak 14.1 GB", "[ 5030.4s] stage 6/8 s6_tryfail step 2950/4000: train bpb s6 0.0599; quiet KL s6 0.00040; 0.70 s/step; the job: 83.8 min spent, about 100.6 min of steps left; card peak 14.1 GB", "[ 5064.7s] stage 6/8 s6_tryfail step 3000/4000: train bpb s6 0.0638; quiet KL s6 0.00040; 0.70 s/step; the job: 84.4 min spent, about 100.1 min of steps left; card peak 14.1 GB", "[ 5098.7s] stage 6/8 s6_tryfail step 3050/4000: train bpb s6 0.0600; quiet KL s6 0.00041; 0.70 s/step; the job: 85.0 min spent, about 99.6 min of steps left; card peak 14.1 GB", "[ 5133.0s] stage 6/8 s6_tryfail step 3100/4000: train bpb s6 0.0609; quiet KL s6 0.00041; 0.70 s/step; the job: 85.6 min spent, about 99.0 min of steps left; card peak 14.1 GB", "[ 5167.2s] stage 6/8 s6_tryfail step 3150/4000: train bpb s6 0.0604; quiet KL s6 0.00042; 0.69 s/step; the job: 86.1 min spent, about 98.5 min of steps left; card peak 14.1 GB", "[ 5200.9s] stage 6/8 s6_tryfail step 3200/4000: train bpb s6 0.0620; quiet KL s6 0.00039; 0.69 s/step; the job: 86.7 min spent, about 97.9 min of steps left; card peak 14.1 GB", "[ 5234.8s] stage 6/8 s6_tryfail step 3250/4000: train bpb s6 0.0595; quiet KL s6 0.00041; 0.69 s/step; the job: 87.2 min spent, about 97.4 min of steps left; card peak 14.1 GB", "[ 5268.7s] stage 6/8 s6_tryfail step 3300/4000: train bpb s6 0.0595; quiet KL s6 0.00044; 0.69 s/step; the job: 87.8 min spent, about 96.9 min of steps left; card peak 14.1 GB", "[ 5302.4s] stage 6/8 s6_tryfail step 3350/4000: train bpb s6 0.0603; quiet KL s6 0.00044; 0.69 s/step; the job: 88.4 min spent, about 96.3 min of steps left; card peak 14.1 GB", "[ 5336.2s] stage 6/8 s6_tryfail step 3400/4000: train bpb s6 0.0622; quiet KL s6 0.00045; 0.69 s/step; the job: 88.9 min spent, about 95.8 min of steps left; card peak 14.1 GB", "[ 5370.1s] stage 6/8 s6_tryfail step 3450/4000: train bpb s6 0.0580; quiet KL s6 0.00042; 0.69 s/step; the job: 89.5 min spent, about 95.2 min of steps left; card peak 14.1 GB", "[ 5404.0s] stage 6/8 s6_tryfail step 3500/4000: train bpb s6 0.0603; quiet KL s6 0.00040; 0.69 s/step; the job: 90.1 min spent, about 94.7 min of steps left; card peak 14.1 GB", "[ 5437.7s] stage 6/8 s6_tryfail step 3550/4000: train bpb s6 0.0610; quiet KL s6 0.00040; 0.69 s/step; the job: 90.6 min spent, about 94.1 min of steps left; card peak 14.1 GB", "[ 5471.6s] stage 6/8 s6_tryfail step 3600/4000: train bpb s6 0.0593; quiet KL s6 0.00038; 0.69 s/step; the job: 91.2 min spent, about 93.6 min of steps left; card peak 14.1 GB", "[ 5505.8s] stage 6/8 s6_tryfail step 3650/4000: train bpb s6 0.0591; quiet KL s6 0.00043; 0.69 s/step; the job: 91.8 min spent, about 93.0 min of steps left; card peak 14.1 GB", "[ 5539.9s] stage 6/8 s6_tryfail step 3700/4000: train bpb s6 0.0571; quiet KL s6 0.00041; 0.69 s/step; the job: 92.3 min spent, about 92.5 min of steps left; card peak 14.1 GB", "[ 5574.1s] stage 6/8 s6_tryfail step 3750/4000: train bpb s6 0.0578; quiet KL s6 0.00041; 0.69 s/step; the job: 92.9 min spent, about 91.9 min of steps left; card peak 14.1 GB", "[ 5608.5s] stage 6/8 s6_tryfail step 3800/4000: train bpb s6 0.0572; quiet KL s6 0.00041; 0.69 s/step; the job: 93.5 min spent, about 91.4 min of steps left; card peak 14.1 GB", "[ 5642.5s] stage 6/8 s6_tryfail step 3850/4000: train bpb s6 0.0593; quiet KL s6 0.00040; 0.69 s/step; the job: 94.0 min spent, about 90.8 min of steps left; card peak 14.1 GB", "[ 5676.7s] stage 6/8 s6_tryfail step 3900/4000: train bpb s6 0.0614; quiet KL s6 0.00040; 0.69 s/step; the job: 94.6 min spent, about 90.3 min of steps left; card peak 14.1 GB", "[ 5710.9s] stage 6/8 s6_tryfail step 3950/4000: train bpb s6 0.0590; quiet KL s6 0.00037; 0.69 s/step; the job: 95.2 min spent, about 89.7 min of steps left; card peak 14.1 GB", "[ 5744.9s] stage 6/8 s6_tryfail step 4000/4000: train bpb s6 0.0607; quiet KL s6 0.00042; 0.69 s/step; the job: 95.7 min spent, about 89.2 min of steps left; card peak 14.1 GB", "[ 5746.0s] anchors saved: /root/refit/group/anchors/gXB_close6_.safetensors for ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail']", "[ 5930.9s] CLOSE 6/8: the group ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail'] all on, against arms off:", "[ 5930.9s] s1_perspective rows: off 0.5538 -> all on 0.0481 (-0.5057 +- 0.0048); with s1_perspective masked 0.5211 (its share -0.4730); s1_perspective alone 0.0522 (-0.5017); gate mean 0.212 / widest 0.555", "[ 5930.9s] s2_concept rows: off 0.7212 -> all on 0.1215 (-0.5997 +- 0.0037); with s2_concept masked 0.2431 (its share -0.1216); s2_concept alone 0.3603 (-0.3608); gate mean 0.273 / widest 0.487", "[ 5930.9s] s3_rules rows: off 0.9296 -> all on 0.2225 (-0.7071 +- 0.0055); with s3_rules masked 0.3181 (its share -0.0956); s3_rules alone 0.8702 (-0.0594); gate mean 0.124 / widest 0.175", "[ 5930.9s] s4_arith rows: off 1.2339 -> all on 0.3364 (-0.8975 +- 0.0058); with s4_arith masked 0.5119 (its share -0.1755); s4_arith alone 1.1550 (-0.0789); gate mean 0.108 / widest 0.167", "[ 5930.9s] s5_causal rows: off 0.6604 -> all on 0.0426 (-0.6178 +- 0.0069); with s5_causal masked 0.4881 (its share -0.4455); s5_causal alone 0.1418 (-0.5187); gate mean 0.186 / widest 0.360", "[ 5930.9s] s6_tryfail rows: off 0.6728 -> all on 0.0546 (-0.6182 +- 0.0036); with s6_tryfail masked 0.4878 (its share -0.4332); s6_tryfail alone 0.1437 (-0.5291); gate mean 0.229 / widest 0.581", "[ 5930.9s] web holdout: off 0.9506 -> all on 0.9521 (+0.00147 +- 0.00034; the quiet bar +.012); each member's share: s1_perspective +0.00006, s2_concept -0.00053, s3_rules +0.00037, s4_arith +0.00016, s5_causal +0.00001, s6_tryfail +0.00009; alone: s1_perspective +0.00134, s2_concept +0.00086, s3_rules +0.00034, s4_arith +0.00006, s5_causal +0.00000, s6_tryfail +0.00011", "[ 5930.9s] a member's share on a PARTNER's rows (all on minus that member masked; near 0 = it stays out): s1_perspective on s2_concept -0.2369; s1_perspective on s3_rules -0.1977; s1_perspective on s4_arith -0.1493; s1_perspective on s5_causal -0.0459; s1_perspective on s6_tryfail -0.0241; s2_concept on s1_perspective -0.0043; s2_concept on s3_rules -0.1012; s2_concept on s4_arith -0.1118; s2_concept on s5_causal -0.0008; s2_concept on s6_tryfail -0.0025; s3_rules on s1_perspective +0.0002; s3_rules on s2_concept -0.0005; s3_rules on s4_arith -0.0005; s3_rules on s5_causal -0.0001; s3_rules on s6_tryfail -0.0002; s4_arith on s1_perspective +0.0001; s4_arith on s2_concept -0.0002; s4_arith on s3_rules -0.0001; s4_arith on s5_causal -0.0001; s4_arith on s6_tryfail -0.0041; s5_causal on s1_perspective +0.0001; s5_causal on s2_concept +0.0000; s5_causal on s3_rules +0.0003; s5_causal on s4_arith -0.0001; s5_causal on s6_tryfail -0.0003; s6_tryfail on s1_perspective +0.0002; s6_tryfail on s2_concept +0.0003; s6_tryfail on s3_rules -0.0001; s6_tryfail on s4_arith -0.0016; s6_tryfail on s5_causal +0.0000", "[ 5930.9s] a member ALONE on a partner's rows (alone minus off): s1_perspective on s2_concept -0.4772; s1_perspective on s3_rules -0.4262; s1_perspective on s4_arith -0.4807; s1_perspective on s5_causal -0.1919; s1_perspective on s6_tryfail -0.1690; s2_concept on s1_perspective -0.0300; s2_concept on s3_rules -0.3138; s2_concept on s4_arith -0.3577; s2_concept on s5_causal -0.0127; s2_concept on s6_tryfail -0.0062; s3_rules on s1_perspective -0.0003; s3_rules on s2_concept +0.0016; s3_rules on s4_arith -0.0047; s3_rules on s5_causal +0.0060; s3_rules on s6_tryfail -0.0011; s4_arith on s1_perspective -0.0013; s4_arith on s2_concept -0.0015; s4_arith on s3_rules +0.0005; s4_arith on s5_causal -0.0011; s4_arith on s6_tryfail -0.0134; s5_causal on s1_perspective -0.0022; s5_causal on s2_concept -0.0010; s5_causal on s3_rules -0.0018; s5_causal on s4_arith -0.0022; s5_causal on s6_tryfail -0.0023; s6_tryfail on s1_perspective +0.0029; s6_tryfail on s2_concept +0.0013; s6_tryfail on s3_rules -0.0003; s6_tryfail on s4_arith -0.0010; s6_tryfail on s5_causal -0.0026", "[ 5930.9s] the group on the rows of stages not yet attached: s7_mixed -0.0066; s8_register -0.1112", "[ 5930.9s] s1_perspective battery accuracy, off -> all on: in 0.550 -> 1.000, transfer 0.463 -> 0.475; alone: in 0.925, transfer 0.463", "[ 5930.9s] s2_concept battery accuracy, off -> all on: in 0.298 -> 0.964, transfer 0.678 -> 0.737; alone: in 0.905, transfer 0.703", "[ 5930.9s] s3_rules battery accuracy, off -> all on: in 0.155 -> 0.890, transfer 0.070 -> 0.330, held-out 0.050 -> 0.467; alone: in 0.335, transfer 0.100, held-out 0.142", "[ 5930.9s] s4_arith battery accuracy, off -> all on: in 0.344 -> 0.619, transfer 0.242 -> 0.283, held-out 0.350 -> 0.325; alone: in 0.412, transfer 0.237, held-out 0.400", "[ 5930.9s] probes mean: off 0.411 -> all on 0.426", "[ 5930.9s] compiled chain: 18 cache entries (limit 256), 0 limit warnings; card peak 14.1 GB; host RSS 5.4 GB", "[ 5935.0s] STAGE 7/8: s7_mixed attached over ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail'] (on, frozen); at birth the group reads exactly as before on its rows and on web text; 4000 steps of 1 task + 1 abstention chunks", "[ 5936.7s] stage 7/8 s7_mixed step 1/4000: train bpb s7 0.1839; quiet KL s7 0.00011; 1.77 s/step; the job: 98.9 min spent, about 89.2 min of steps left; card peak 14.2 GB", "[ 6007.3s] stage 7/8 s7_mixed step 50/4000: train bpb s7 0.4395; quiet KL s7 0.00014; 1.45 s/step; the job: 100.1 min spent, about 89.3 min of steps left; card peak 14.3 GB", "[ 6045.4s] stage 7/8 s7_mixed step 100/4000: train bpb s7 0.5081; quiet KL s7 0.00016; 1.10 s/step; the job: 100.8 min spent, about 88.8 min of steps left; card peak 14.3 GB", "[ 6081.5s] stage 7/8 s7_mixed step 150/4000: train bpb s7 0.4529; quiet KL s7 0.00020; 0.98 s/step; the job: 101.4 min spent, about 88.3 min of steps left; card peak 14.3 GB", "[ 6117.7s] stage 7/8 s7_mixed step 200/4000: train bpb s7 0.4662; quiet KL s7 0.00022; 0.91 s/step; the job: 102.0 min spent, about 87.7 min of steps left; card peak 14.3 GB", "[ 6154.0s] stage 7/8 s7_mixed step 250/4000: train bpb s7 0.4931; quiet KL s7 0.00031; 0.88 s/step; the job: 102.6 min spent, about 87.2 min of steps left; card peak 14.3 GB", "[ 6190.2s] stage 7/8 s7_mixed step 300/4000: train bpb s7 0.5269; quiet KL s7 0.00033; 0.85 s/step; the job: 103.2 min spent, about 86.7 min of steps left; card peak 14.3 GB", "[ 6226.6s] stage 7/8 s7_mixed step 350/4000: train bpb s7 0.3769; quiet KL s7 0.00038; 0.83 s/step; the job: 103.8 min spent, about 86.2 min of steps left; card peak 14.3 GB", "[ 6262.9s] stage 7/8 s7_mixed step 400/4000: train bpb s7 0.4406; quiet KL s7 0.00036; 0.82 s/step; the job: 104.4 min spent, about 85.7 min of steps left; card peak 14.3 GB", "[ 6299.2s] stage 7/8 s7_mixed step 450/4000: train bpb s7 0.4244; quiet KL s7 0.00035; 0.81 s/step; the job: 105.0 min spent, about 85.1 min of steps left; card peak 14.3 GB", "[ 6335.5s] stage 7/8 s7_mixed step 500/4000: train bpb s7 0.4225; quiet KL s7 0.00045; 0.80 s/step; the job: 105.6 min spent, about 84.6 min of steps left; card peak 14.3 GB", "[ 6371.6s] stage 7/8 s7_mixed step 550/4000: train bpb s7 0.5462; quiet KL s7 0.00045; 0.79 s/step; the job: 106.2 min spent, about 84.1 min of steps left; card peak 14.3 GB", "[ 6407.8s] stage 7/8 s7_mixed step 600/4000: train bpb s7 0.3887; quiet KL s7 0.00040; 0.79 s/step; the job: 106.8 min spent, about 83.5 min of steps left; card peak 14.3 GB", "[ 6444.2s] stage 7/8 s7_mixed step 650/4000: train bpb s7 0.4559; quiet KL s7 0.00041; 0.78 s/step; the job: 107.4 min spent, about 83.0 min of steps left; card peak 14.3 GB", "[ 6480.7s] stage 7/8 s7_mixed step 700/4000: train bpb s7 0.5294; quiet KL s7 0.00056; 0.78 s/step; the job: 108.0 min spent, about 82.5 min of steps left; card peak 14.3 GB", "[ 6517.2s] stage 7/8 s7_mixed step 750/4000: train bpb s7 0.4394; quiet KL s7 0.00057; 0.78 s/step; the job: 108.6 min spent, about 82.0 min of steps left; card peak 14.3 GB", "[ 6553.7s] stage 7/8 s7_mixed step 800/4000: train bpb s7 0.4267; quiet KL s7 0.00070; 0.77 s/step; the job: 109.2 min spent, about 81.4 min of steps left; card peak 14.3 GB", "[ 6590.1s] stage 7/8 s7_mixed step 850/4000: train bpb s7 0.4690; quiet KL s7 0.00063; 0.77 s/step; the job: 109.8 min spent, about 80.9 min of steps left; card peak 14.3 GB", "[ 6626.4s] stage 7/8 s7_mixed step 900/4000: train bpb s7 0.5084; quiet KL s7 0.00079; 0.77 s/step; the job: 110.4 min spent, about 80.4 min of steps left; card peak 14.3 GB", "[ 6662.7s] stage 7/8 s7_mixed step 950/4000: train bpb s7 0.4823; quiet KL s7 0.00078; 0.77 s/step; the job: 111.0 min spent, about 79.8 min of steps left; card peak 14.3 GB", "[ 6699.2s] stage 7/8 s7_mixed step 1000/4000: train bpb s7 0.4056; quiet KL s7 0.00066; 0.76 s/step; the job: 111.7 min spent, about 79.3 min of steps left; card peak 14.3 GB", "[ 6735.7s] stage 7/8 s7_mixed step 1050/4000: train bpb s7 0.4017; quiet KL s7 0.00067; 0.76 s/step; the job: 112.3 min spent, about 78.8 min of steps left; card peak 14.3 GB", "[ 6772.2s] stage 7/8 s7_mixed step 1100/4000: train bpb s7 0.3963; quiet KL s7 0.00068; 0.76 s/step; the job: 112.9 min spent, about 78.2 min of steps left; card peak 14.3 GB", "[ 6808.8s] stage 7/8 s7_mixed step 1150/4000: train bpb s7 0.4763; quiet KL s7 0.00067; 0.76 s/step; the job: 113.5 min spent, about 77.7 min of steps left; card peak 14.3 GB", "[ 6845.2s] stage 7/8 s7_mixed step 1200/4000: train bpb s7 0.4271; quiet KL s7 0.00056; 0.76 s/step; the job: 114.1 min spent, about 77.1 min of steps left; card peak 14.3 GB", "[ 6881.7s] stage 7/8 s7_mixed step 1250/4000: train bpb s7 0.4475; quiet KL s7 0.00055; 0.76 s/step; the job: 114.7 min spent, about 76.6 min of steps left; card peak 14.3 GB", "[ 6918.2s] stage 7/8 s7_mixed step 1300/4000: train bpb s7 0.4104; quiet KL s7 0.00071; 0.76 s/step; the job: 115.3 min spent, about 76.1 min of steps left; card peak 14.3 GB", "[ 6954.7s] stage 7/8 s7_mixed step 1350/4000: train bpb s7 0.5277; quiet KL s7 0.00085; 0.76 s/step; the job: 115.9 min spent, about 75.5 min of steps left; card peak 14.3 GB", "[ 6990.9s] stage 7/8 s7_mixed step 1400/4000: train bpb s7 0.4393; quiet KL s7 0.00078; 0.75 s/step; the job: 116.5 min spent, about 75.0 min of steps left; card peak 14.3 GB", "[ 7027.1s] stage 7/8 s7_mixed step 1450/4000: train bpb s7 0.3759; quiet KL s7 0.00058; 0.75 s/step; the job: 117.1 min spent, about 74.4 min of steps left; card peak 14.3 GB", "[ 7063.3s] stage 7/8 s7_mixed step 1500/4000: train bpb s7 0.4307; quiet KL s7 0.00066; 0.75 s/step; the job: 117.7 min spent, about 73.9 min of steps left; card peak 14.3 GB", "[ 7099.6s] stage 7/8 s7_mixed step 1550/4000: train bpb s7 0.3765; quiet KL s7 0.00056; 0.75 s/step; the job: 118.3 min spent, about 73.4 min of steps left; card peak 14.3 GB", "[ 7135.8s] stage 7/8 s7_mixed step 1600/4000: train bpb s7 0.4728; quiet KL s7 0.00054; 0.75 s/step; the job: 118.9 min spent, about 72.8 min of steps left; card peak 14.3 GB", "[ 7172.1s] stage 7/8 s7_mixed step 1650/4000: train bpb s7 0.4451; quiet KL s7 0.00060; 0.75 s/step; the job: 119.5 min spent, about 72.3 min of steps left; card peak 14.3 GB", "[ 7208.4s] stage 7/8 s7_mixed step 1700/4000: train bpb s7 0.3963; quiet KL s7 0.00060; 0.75 s/step; the job: 120.1 min spent, about 71.7 min of steps left; card peak 14.3 GB", "[ 7244.7s] stage 7/8 s7_mixed step 1750/4000: train bpb s7 0.4506; quiet KL s7 0.00063; 0.75 s/step; the job: 120.7 min spent, about 71.2 min of steps left; card peak 14.3 GB", "[ 7281.4s] stage 7/8 s7_mixed step 1800/4000: train bpb s7 0.4440; quiet KL s7 0.00060; 0.75 s/step; the job: 121.4 min spent, about 70.6 min of steps left; card peak 14.3 GB", "[ 7317.9s] stage 7/8 s7_mixed step 1850/4000: train bpb s7 0.4507; quiet KL s7 0.00061; 0.75 s/step; the job: 122.0 min spent, about 70.1 min of steps left; card peak 14.3 GB", "[ 7354.5s] stage 7/8 s7_mixed step 1900/4000: train bpb s7 0.4530; quiet KL s7 0.00064; 0.75 s/step; the job: 122.6 min spent, about 69.5 min of steps left; card peak 14.3 GB", "[ 7391.5s] stage 7/8 s7_mixed step 1950/4000: train bpb s7 0.4510; quiet KL s7 0.00067; 0.75 s/step; the job: 123.2 min spent, about 69.0 min of steps left; card peak 14.3 GB", "[ 7428.3s] stage 7/8 s7_mixed step 2000/4000: train bpb s7 0.4289; quiet KL s7 0.00069; 0.75 s/step; the job: 123.8 min spent, about 68.5 min of steps left; card peak 14.3 GB", "[ 7465.1s] stage 7/8 s7_mixed step 2050/4000: train bpb s7 0.3810; quiet KL s7 0.00064; 0.75 s/step; the job: 124.4 min spent, about 67.9 min of steps left; card peak 14.3 GB", "[ 7501.9s] stage 7/8 s7_mixed step 2100/4000: train bpb s7 0.3504; quiet KL s7 0.00061; 0.75 s/step; the job: 125.0 min spent, about 67.4 min of steps left; card peak 14.3 GB", "[ 7538.7s] stage 7/8 s7_mixed step 2150/4000: train bpb s7 0.4724; quiet KL s7 0.00057; 0.75 s/step; the job: 125.6 min spent, about 66.8 min of steps left; card peak 14.3 GB", "[ 7575.4s] stage 7/8 s7_mixed step 2200/4000: train bpb s7 0.4324; quiet KL s7 0.00060; 0.75 s/step; the job: 126.3 min spent, about 66.3 min of steps left; card peak 14.3 GB", "[ 7612.2s] stage 7/8 s7_mixed step 2250/4000: train bpb s7 0.3504; quiet KL s7 0.00059; 0.75 s/step; the job: 126.9 min spent, about 65.7 min of steps left; card peak 14.3 GB", "[ 7649.0s] stage 7/8 s7_mixed step 2300/4000: train bpb s7 0.4314; quiet KL s7 0.00064; 0.75 s/step; the job: 127.5 min spent, about 65.2 min of steps left; card peak 14.3 GB", "[ 7685.7s] stage 7/8 s7_mixed step 2350/4000: train bpb s7 0.4042; quiet KL s7 0.00068; 0.75 s/step; the job: 128.1 min spent, about 64.6 min of steps left; card peak 14.3 GB", "[ 7722.4s] stage 7/8 s7_mixed step 2400/4000: train bpb s7 0.4466; quiet KL s7 0.00062; 0.74 s/step; the job: 128.7 min spent, about 64.1 min of steps left; card peak 14.3 GB", "[ 7759.1s] stage 7/8 s7_mixed step 2450/4000: train bpb s7 0.4690; quiet KL s7 0.00071; 0.74 s/step; the job: 129.3 min spent, about 63.5 min of steps left; card peak 14.3 GB", "[ 7795.7s] stage 7/8 s7_mixed step 2500/4000: train bpb s7 0.4238; quiet KL s7 0.00065; 0.74 s/step; the job: 129.9 min spent, about 63.0 min of steps left; card peak 14.3 GB", "[ 7832.4s] stage 7/8 s7_mixed step 2550/4000: train bpb s7 0.3321; quiet KL s7 0.00055; 0.74 s/step; the job: 130.5 min spent, about 62.4 min of steps left; card peak 14.3 GB", "[ 7869.1s] stage 7/8 s7_mixed step 2600/4000: train bpb s7 0.4020; quiet KL s7 0.00068; 0.74 s/step; the job: 131.2 min spent, about 61.9 min of steps left; card peak 14.3 GB", "[ 7905.8s] stage 7/8 s7_mixed step 2650/4000: train bpb s7 0.3683; quiet KL s7 0.00064; 0.74 s/step; the job: 131.8 min spent, about 61.3 min of steps left; card peak 14.3 GB", "[ 7942.5s] stage 7/8 s7_mixed step 2700/4000: train bpb s7 0.4454; quiet KL s7 0.00065; 0.74 s/step; the job: 132.4 min spent, about 60.8 min of steps left; card peak 14.3 GB", "[ 7979.2s] stage 7/8 s7_mixed step 2750/4000: train bpb s7 0.3968; quiet KL s7 0.00063; 0.74 s/step; the job: 133.0 min spent, about 60.2 min of steps left; card peak 14.3 GB", "[ 8015.9s] stage 7/8 s7_mixed step 2800/4000: train bpb s7 0.4423; quiet KL s7 0.00063; 0.74 s/step; the job: 133.6 min spent, about 59.6 min of steps left; card peak 14.3 GB", "[ 8052.7s] stage 7/8 s7_mixed step 2850/4000: train bpb s7 0.3865; quiet KL s7 0.00067; 0.74 s/step; the job: 134.2 min spent, about 59.1 min of steps left; card peak 14.3 GB", "[ 8089.4s] stage 7/8 s7_mixed step 2900/4000: train bpb s7 0.3749; quiet KL s7 0.00087; 0.74 s/step; the job: 134.8 min spent, about 58.5 min of steps left; card peak 14.3 GB", "[ 8126.1s] stage 7/8 s7_mixed step 2950/4000: train bpb s7 0.3731; quiet KL s7 0.00071; 0.74 s/step; the job: 135.4 min spent, about 58.0 min of steps left; card peak 14.3 GB", "[ 8163.1s] stage 7/8 s7_mixed step 3000/4000: train bpb s7 0.4007; quiet KL s7 0.00063; 0.74 s/step; the job: 136.1 min spent, about 57.4 min of steps left; card peak 14.3 GB", "[ 8200.1s] stage 7/8 s7_mixed step 3050/4000: train bpb s7 0.3594; quiet KL s7 0.00063; 0.74 s/step; the job: 136.7 min spent, about 56.9 min of steps left; card peak 14.3 GB", "[ 8236.9s] stage 7/8 s7_mixed step 3100/4000: train bpb s7 0.4813; quiet KL s7 0.00066; 0.74 s/step; the job: 137.3 min spent, about 56.3 min of steps left; card peak 14.3 GB", "[ 8273.8s] stage 7/8 s7_mixed step 3150/4000: train bpb s7 0.4742; quiet KL s7 0.00060; 0.74 s/step; the job: 137.9 min spent, about 55.8 min of steps left; card peak 14.3 GB", "[ 8310.8s] stage 7/8 s7_mixed step 3200/4000: train bpb s7 0.4433; quiet KL s7 0.00059; 0.74 s/step; the job: 138.5 min spent, about 55.2 min of steps left; card peak 14.3 GB", "[ 8347.7s] stage 7/8 s7_mixed step 3250/4000: train bpb s7 0.4342; quiet KL s7 0.00061; 0.74 s/step; the job: 139.1 min spent, about 54.6 min of steps left; card peak 14.3 GB", "[ 8384.7s] stage 7/8 s7_mixed step 3300/4000: train bpb s7 0.3662; quiet KL s7 0.00068; 0.74 s/step; the job: 139.7 min spent, about 54.1 min of steps left; card peak 14.3 GB", "[ 8421.6s] stage 7/8 s7_mixed step 3350/4000: train bpb s7 0.3878; quiet KL s7 0.00063; 0.74 s/step; the job: 140.4 min spent, about 53.5 min of steps left; card peak 14.3 GB", "[ 8458.4s] stage 7/8 s7_mixed step 3400/4000: train bpb s7 0.4314; quiet KL s7 0.00070; 0.74 s/step; the job: 141.0 min spent, about 53.0 min of steps left; card peak 14.3 GB", "[ 8495.3s] stage 7/8 s7_mixed step 3450/4000: train bpb s7 0.3234; quiet KL s7 0.00072; 0.74 s/step; the job: 141.6 min spent, about 52.4 min of steps left; card peak 14.3 GB", "[ 8532.1s] stage 7/8 s7_mixed step 3500/4000: train bpb s7 0.3902; quiet KL s7 0.00062; 0.74 s/step; the job: 142.2 min spent, about 51.8 min of steps left; card peak 14.3 GB", "[ 8568.9s] stage 7/8 s7_mixed step 3550/4000: train bpb s7 0.4182; quiet KL s7 0.00058; 0.74 s/step; the job: 142.8 min spent, about 51.3 min of steps left; card peak 14.3 GB", "[ 8605.7s] stage 7/8 s7_mixed step 3600/4000: train bpb s7 0.3988; quiet KL s7 0.00064; 0.74 s/step; the job: 143.4 min spent, about 50.7 min of steps left; card peak 14.3 GB", "[ 8642.5s] stage 7/8 s7_mixed step 3650/4000: train bpb s7 0.4450; quiet KL s7 0.00072; 0.74 s/step; the job: 144.0 min spent, about 50.2 min of steps left; card peak 14.3 GB", "[ 8679.3s] stage 7/8 s7_mixed step 3700/4000: train bpb s7 0.4566; quiet KL s7 0.00069; 0.74 s/step; the job: 144.7 min spent, about 49.6 min of steps left; card peak 14.3 GB", "[ 8716.2s] stage 7/8 s7_mixed step 3750/4000: train bpb s7 0.5194; quiet KL s7 0.00071; 0.74 s/step; the job: 145.3 min spent, about 49.0 min of steps left; card peak 14.3 GB", "[ 8753.0s] stage 7/8 s7_mixed step 3800/4000: train bpb s7 0.3606; quiet KL s7 0.00060; 0.74 s/step; the job: 145.9 min spent, about 48.5 min of steps left; card peak 14.3 GB", "[ 8789.8s] stage 7/8 s7_mixed step 3850/4000: train bpb s7 0.3989; quiet KL s7 0.00065; 0.74 s/step; the job: 146.5 min spent, about 47.9 min of steps left; card peak 14.3 GB", "[ 8826.6s] stage 7/8 s7_mixed step 3900/4000: train bpb s7 0.4600; quiet KL s7 0.00092; 0.74 s/step; the job: 147.1 min spent, about 47.3 min of steps left; card peak 14.3 GB", "[ 8863.2s] stage 7/8 s7_mixed step 3950/4000: train bpb s7 0.3999; quiet KL s7 0.00064; 0.74 s/step; the job: 147.7 min spent, about 46.8 min of steps left; card peak 14.3 GB", "[ 8900.1s] stage 7/8 s7_mixed step 4000/4000: train bpb s7 0.4241; quiet KL s7 0.00065; 0.74 s/step; the job: 148.3 min spent, about 46.2 min of steps left; card peak 14.3 GB", "[ 8901.2s] anchors saved: /root/refit/group/anchors/gXB_close7_.safetensors for ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail', 's7_mixed']", "[ 9115.8s] CLOSE 7/8: the group ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail', 's7_mixed'] all on, against arms off:", "[ 9115.8s] s1_perspective rows: off 0.5538 -> all on 0.0482 (-0.5056 +- 0.0048); with s1_perspective masked 0.4921 (its share -0.4439); s1_perspective alone 0.0522 (-0.5017); gate mean 0.212 / widest 0.555", "[ 9115.8s] s2_concept rows: off 0.7212 -> all on 0.1212 (-0.6000 +- 0.0038); with s2_concept masked 0.2420 (its share -0.1208); s2_concept alone 0.3603 (-0.3608); gate mean 0.273 / widest 0.487", "[ 9115.8s] s3_rules rows: off 0.9296 -> all on 0.2217 (-0.7079 +- 0.0055); with s3_rules masked 0.3142 (its share -0.0925); s3_rules alone 0.8702 (-0.0594); gate mean 0.124 / widest 0.175", "[ 9115.8s] s4_arith rows: off 1.2339 -> all on 0.3265 (-0.9074 +- 0.0063); with s4_arith masked 0.4916 (its share -0.1651); s4_arith alone 1.1550 (-0.0789); gate mean 0.108 / widest 0.167", "[ 9115.8s] s5_causal rows: off 0.6604 -> all on 0.0426 (-0.6178 +- 0.0070); with s5_causal masked 0.4702 (its share -0.4276); s5_causal alone 0.1418 (-0.5187); gate mean 0.186 / widest 0.360", "[ 9115.8s] s6_tryfail rows: off 0.6728 -> all on 0.0542 (-0.6186 +- 0.0036); with s6_tryfail masked 0.4688 (its share -0.4146); s6_tryfail alone 0.1437 (-0.5291); gate mean 0.229 / widest 0.581", "[ 9115.8s] s7_mixed rows: off 0.7830 -> all on 0.7196 (-0.0634 +- 0.0390); with s7_mixed masked 0.7764 (its share -0.0568); s7_mixed alone 0.7254 (-0.0576); gate mean 0.253 / widest 0.467", "[ 9115.8s] web holdout: off 0.9506 -> all on 0.9512 (+0.00057 +- 0.00032; the quiet bar +.012); each member's share: s1_perspective +0.00001, s2_concept -0.00103, s3_rules +0.00014, s4_arith -0.00015, s5_causal +0.00011, s6_tryfail -0.00008, s7_mixed -0.00090; alone: s1_perspective +0.00134, s2_concept +0.00086, s3_rules +0.00034, s4_arith +0.00006, s5_causal +0.00000, s6_tryfail +0.00011, s7_mixed -0.00032", "[ 9115.8s] a member's share on a PARTNER's rows (all on minus that member masked; near 0 = it stays out): s1_perspective on s2_concept -0.2305; s1_perspective on s3_rules -0.1953; s1_perspective on s4_arith -0.1345; s1_perspective on s5_causal -0.0308; s1_perspective on s6_tryfail -0.0212; s1_perspective on s7_mixed -0.0043; s2_concept on s1_perspective -0.0041; s2_concept on s3_rules -0.1009; s2_concept on s4_arith -0.0990; s2_concept on s5_causal -0.0009; s2_concept on s6_tryfail -0.0024; s2_concept on s7_mixed -0.0002; s3_rules on s1_perspective +0.0001; s3_rules on s2_concept -0.0005; s3_rules on s4_arith -0.0003; s3_rules on s5_causal -0.0001; s3_rules on s6_tryfail -0.0003; s3_rules on s7_mixed -0.0000; s4_arith on s1_perspective +0.0001; s4_arith on s2_concept -0.0000; s4_arith on s3_rules +0.0001; s4_arith on s5_causal -0.0002; s4_arith on s6_tryfail -0.0038; s4_arith on s7_mixed +0.0002; s5_causal on s1_perspective +0.0000; s5_causal on s2_concept +0.0002; s5_causal on s3_rules +0.0003; s5_causal on s4_arith +0.0001; s5_causal on s6_tryfail -0.0004; s5_causal on s7_mixed +0.0001; s6_tryfail on s1_perspective +0.0002; s6_tryfail on s2_concept +0.0004; s6_tryfail on s3_rules +0.0001; s6_tryfail on s4_arith -0.0015; s6_tryfail on s5_causal +0.0001; s6_tryfail on s7_mixed -0.0000; s7_mixed on s1_perspective +0.0001; s7_mixed on s2_concept -0.0004; s7_mixed on s3_rules -0.0008; s7_mixed on s4_arith -0.0099; s7_mixed on s5_causal +0.0000; s7_mixed on s6_tryfail -0.0004", "[ 9115.8s] a member ALONE on a partner's rows (alone minus off): s1_perspective on s2_concept -0.4772; s1_perspective on s3_rules -0.4262; s1_perspective on s4_arith -0.4807; s1_perspective on s5_causal -0.1919; s1_perspective on s6_tryfail -0.1690; s1_perspective on s7_mixed -0.0088; s2_concept on s1_perspective -0.0300; s2_concept on s3_rules -0.3138; s2_concept on s4_arith -0.3577; s2_concept on s5_causal -0.0127; s2_concept on s6_tryfail -0.0062; s2_concept on s7_mixed +0.0002; s3_rules on s1_perspective -0.0003; s3_rules on s2_concept +0.0016; s3_rules on s4_arith -0.0047; s3_rules on s5_causal +0.0060; s3_rules on s6_tryfail -0.0011; s3_rules on s7_mixed -0.0007; s4_arith on s1_perspective -0.0013; s4_arith on s2_concept -0.0015; s4_arith on s3_rules +0.0005; s4_arith on s5_causal -0.0011; s4_arith on s6_tryfail -0.0134; s4_arith on s7_mixed -0.0003; s5_causal on s1_perspective -0.0022; s5_causal on s2_concept -0.0010; s5_causal on s3_rules -0.0018; s5_causal on s4_arith -0.0022; s5_causal on s6_tryfail -0.0023; s5_causal on s7_mixed -0.0001; s6_tryfail on s1_perspective +0.0029; s6_tryfail on s2_concept +0.0013; s6_tryfail on s3_rules -0.0003; s6_tryfail on s4_arith -0.0010; s6_tryfail on s5_causal -0.0026; s6_tryfail on s7_mixed +0.0001; s7_mixed on s1_perspective -0.0237; s7_mixed on s2_concept -0.0022; s7_mixed on s3_rules -0.0529; s7_mixed on s4_arith -0.0414; s7_mixed on s5_causal -0.0976; s7_mixed on s6_tryfail -0.0774", "[ 9115.8s] the group on the rows of stages not yet attached: s8_register -0.1125", "[ 9115.8s] s1_perspective battery accuracy, off -> all on: in 0.550 -> 1.000, transfer 0.463 -> 0.425; alone: in 0.925, transfer 0.463", "[ 9115.8s] s2_concept battery accuracy, off -> all on: in 0.298 -> 0.952, transfer 0.678 -> 0.737; alone: in 0.905, transfer 0.703", "[ 9115.8s] s3_rules battery accuracy, off -> all on: in 0.155 -> 0.925, transfer 0.070 -> 0.350, held-out 0.050 -> 0.483; alone: in 0.335, transfer 0.100, held-out 0.142", "[ 9115.8s] s4_arith battery accuracy, off -> all on: in 0.344 -> 0.625, transfer 0.242 -> 0.258, held-out 0.350 -> 0.287; alone: in 0.412, transfer 0.237, held-out 0.400", "[ 9115.8s] probes mean: off 0.411 -> all on 0.415", "[ 9115.8s] compiled chain: 23 cache entries (limit 256), 0 limit warnings; card peak 14.3 GB; host RSS 11.9 GB", "[ 9120.1s] STAGE 8/8: s8_register attached over ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail', 's7_mixed'] (on, frozen); at birth the group reads exactly as before on its rows and on web text; 4000 steps of 1 task + 1 abstention chunks", "[ 9122.0s] stage 8/8 s8_register step 1/4000: train bpb s8 0.3816; quiet KL s8 0.00016; 1.89 s/step; the job: 152.0 min spent, about 46.2 min of steps left; card peak 14.4 GB", "[ 9159.8s] stage 8/8 s8_register step 50/4000: train bpb s8 0.3760; quiet KL s8 0.00015; 0.79 s/step; the job: 152.7 min spent, about 45.7 min of steps left; card peak 14.5 GB", "[ 9198.6s] stage 8/8 s8_register step 100/4000: train bpb s8 0.3565; quiet KL s8 0.00026; 0.79 s/step; the job: 153.3 min spent, about 45.1 min of steps left; card peak 14.5 GB", "[ 9237.2s] stage 8/8 s8_register step 150/4000: train bpb s8 0.3341; quiet KL s8 0.00044; 0.78 s/step; the job: 154.0 min spent, about 44.5 min of steps left; card peak 14.5 GB", "[ 9276.1s] stage 8/8 s8_register step 200/4000: train bpb s8 0.3108; quiet KL s8 0.00055; 0.78 s/step; the job: 154.6 min spent, about 44.0 min of steps left; card peak 14.5 GB", "[ 9314.7s] stage 8/8 s8_register step 250/4000: train bpb s8 0.2774; quiet KL s8 0.00069; 0.78 s/step; the job: 155.2 min spent, about 43.4 min of steps left; card peak 14.5 GB", "[ 9353.6s] stage 8/8 s8_register step 300/4000: train bpb s8 0.2435; quiet KL s8 0.00078; 0.78 s/step; the job: 155.9 min spent, about 42.9 min of steps left; card peak 14.5 GB", "[ 9392.2s] stage 8/8 s8_register step 350/4000: train bpb s8 0.2065; quiet KL s8 0.00084; 0.78 s/step; the job: 156.5 min spent, about 42.3 min of steps left; card peak 14.5 GB", "[ 9431.1s] stage 8/8 s8_register step 400/4000: train bpb s8 0.1763; quiet KL s8 0.00078; 0.78 s/step; the job: 157.2 min spent, about 41.7 min of steps left; card peak 14.5 GB", "[ 9469.8s] stage 8/8 s8_register step 450/4000: train bpb s8 0.1446; quiet KL s8 0.00080; 0.78 s/step; the job: 157.8 min spent, about 41.2 min of steps left; card peak 14.5 GB", "[ 9508.8s] stage 8/8 s8_register step 500/4000: train bpb s8 0.1194; quiet KL s8 0.00067; 0.78 s/step; the job: 158.5 min spent, about 40.6 min of steps left; card peak 14.5 GB", "[ 9547.5s] stage 8/8 s8_register step 550/4000: train bpb s8 0.0959; quiet KL s8 0.00073; 0.78 s/step; the job: 159.1 min spent, about 40.1 min of steps left; card peak 14.5 GB", "[ 9586.4s] stage 8/8 s8_register step 600/4000: train bpb s8 0.0803; quiet KL s8 0.00074; 0.78 s/step; the job: 159.8 min spent, about 39.5 min of steps left; card peak 14.5 GB", "[ 9625.1s] stage 8/8 s8_register step 650/4000: train bpb s8 0.0669; quiet KL s8 0.00072; 0.78 s/step; the job: 160.4 min spent, about 38.9 min of steps left; card peak 14.5 GB", "[ 9664.1s] stage 8/8 s8_register step 700/4000: train bpb s8 0.0574; quiet KL s8 0.00070; 0.78 s/step; the job: 161.1 min spent, about 38.4 min of steps left; card peak 14.5 GB", "[ 9702.9s] stage 8/8 s8_register step 750/4000: train bpb s8 0.0489; quiet KL s8 0.00068; 0.78 s/step; the job: 161.7 min spent, about 37.8 min of steps left; card peak 14.5 GB", "[ 9741.8s] stage 8/8 s8_register step 800/4000: train bpb s8 0.0437; quiet KL s8 0.00072; 0.78 s/step; the job: 162.4 min spent, about 37.2 min of steps left; card peak 14.5 GB", "[ 9780.5s] stage 8/8 s8_register step 850/4000: train bpb s8 0.0395; quiet KL s8 0.00064; 0.78 s/step; the job: 163.0 min spent, about 36.7 min of steps left; card peak 14.5 GB", "[ 9819.5s] stage 8/8 s8_register step 900/4000: train bpb s8 0.0366; quiet KL s8 0.00060; 0.78 s/step; the job: 163.7 min spent, about 36.1 min of steps left; card peak 14.5 GB", "[ 9858.2s] stage 8/8 s8_register step 950/4000: train bpb s8 0.0337; quiet KL s8 0.00062; 0.78 s/step; the job: 164.3 min spent, about 35.5 min of steps left; card peak 14.5 GB", "[ 9896.9s] stage 8/8 s8_register step 1000/4000: train bpb s8 0.0324; quiet KL s8 0.00056; 0.78 s/step; the job: 164.9 min spent, about 35.0 min of steps left; card peak 14.5 GB", "[ 9935.9s] stage 8/8 s8_register step 1050/4000: train bpb s8 0.0302; quiet KL s8 0.00054; 0.78 s/step; the job: 165.6 min spent, about 34.4 min of steps left; card peak 14.5 GB", "[ 9974.6s] stage 8/8 s8_register step 1100/4000: train bpb s8 0.0336; quiet KL s8 0.00090; 0.78 s/step; the job: 166.2 min spent, about 33.8 min of steps left; card peak 14.5 GB", "[10013.6s] stage 8/8 s8_register step 1150/4000: train bpb s8 0.0288; quiet KL s8 0.00059; 0.78 s/step; the job: 166.9 min spent, about 33.3 min of steps left; card peak 14.5 GB", "[10052.4s] stage 8/8 s8_register step 1200/4000: train bpb s8 0.0281; quiet KL s8 0.00051; 0.78 s/step; the job: 167.5 min spent, about 32.7 min of steps left; card peak 14.5 GB", "[10091.4s] stage 8/8 s8_register step 1250/4000: train bpb s8 0.0271; quiet KL s8 0.00047; 0.78 s/step; the job: 168.2 min spent, about 32.1 min of steps left; card peak 14.5 GB", "[10130.1s] stage 8/8 s8_register step 1300/4000: train bpb s8 0.0274; quiet KL s8 0.00045; 0.78 s/step; the job: 168.8 min spent, about 31.6 min of steps left; card peak 14.5 GB", "[10169.0s] stage 8/8 s8_register step 1350/4000: train bpb s8 0.0261; quiet KL s8 0.00045; 0.78 s/step; the job: 169.5 min spent, about 31.0 min of steps left; card peak 14.5 GB", "[10207.8s] stage 8/8 s8_register step 1400/4000: train bpb s8 0.0258; quiet KL s8 0.00045; 0.78 s/step; the job: 170.1 min spent, about 30.4 min of steps left; card peak 14.5 GB", "[10246.8s] stage 8/8 s8_register step 1450/4000: train bpb s8 0.0258; quiet KL s8 0.00046; 0.78 s/step; the job: 170.8 min spent, about 29.8 min of steps left; card peak 14.5 GB", "[10285.2s] stage 8/8 s8_register step 1500/4000: train bpb s8 0.0254; quiet KL s8 0.00045; 0.78 s/step; the job: 171.4 min spent, about 29.3 min of steps left; card peak 14.5 GB", "[10323.8s] stage 8/8 s8_register step 1550/4000: train bpb s8 0.0246; quiet KL s8 0.00040; 0.78 s/step; the job: 172.1 min spent, about 28.7 min of steps left; card peak 14.5 GB", "[10362.2s] stage 8/8 s8_register step 1600/4000: train bpb s8 0.0251; quiet KL s8 0.00041; 0.78 s/step; the job: 172.7 min spent, about 28.1 min of steps left; card peak 14.5 GB", "[10400.7s] stage 8/8 s8_register step 1650/4000: train bpb s8 0.0250; quiet KL s8 0.00043; 0.78 s/step; the job: 173.3 min spent, about 27.5 min of steps left; card peak 14.5 GB", "[10439.1s] stage 8/8 s8_register step 1700/4000: train bpb s8 0.0250; quiet KL s8 0.00040; 0.78 s/step; the job: 174.0 min spent, about 27.0 min of steps left; card peak 14.5 GB", "[10477.9s] stage 8/8 s8_register step 1750/4000: train bpb s8 0.0247; quiet KL s8 0.00041; 0.78 s/step; the job: 174.6 min spent, about 26.4 min of steps left; card peak 14.5 GB", "[10516.3s] stage 8/8 s8_register step 1800/4000: train bpb s8 0.0245; quiet KL s8 0.00043; 0.78 s/step; the job: 175.3 min spent, about 25.8 min of steps left; card peak 14.5 GB", "[10555.2s] stage 8/8 s8_register step 1850/4000: train bpb s8 0.0246; quiet KL s8 0.00041; 0.78 s/step; the job: 175.9 min spent, about 25.2 min of steps left; card peak 14.5 GB", "[10593.9s] stage 8/8 s8_register step 1900/4000: train bpb s8 0.0243; quiet KL s8 0.00036; 0.78 s/step; the job: 176.6 min spent, about 24.7 min of steps left; card peak 14.5 GB", "[10632.8s] stage 8/8 s8_register step 1950/4000: train bpb s8 0.0243; quiet KL s8 0.00036; 0.78 s/step; the job: 177.2 min spent, about 24.1 min of steps left; card peak 14.5 GB", "[10671.5s] stage 8/8 s8_register step 2000/4000: train bpb s8 0.0240; quiet KL s8 0.00040; 0.78 s/step; the job: 177.9 min spent, about 23.5 min of steps left; card peak 14.5 GB", "[10710.6s] stage 8/8 s8_register step 2050/4000: train bpb s8 0.0244; quiet KL s8 0.00036; 0.78 s/step; the job: 178.5 min spent, about 22.9 min of steps left; card peak 14.5 GB", "[10749.2s] stage 8/8 s8_register step 2100/4000: train bpb s8 0.0240; quiet KL s8 0.00039; 0.78 s/step; the job: 179.2 min spent, about 22.3 min of steps left; card peak 14.5 GB", "[10788.3s] stage 8/8 s8_register step 2150/4000: train bpb s8 0.0238; quiet KL s8 0.00037; 0.78 s/step; the job: 179.8 min spent, about 21.8 min of steps left; card peak 14.5 GB", "[10827.0s] stage 8/8 s8_register step 2200/4000: train bpb s8 0.0235; quiet KL s8 0.00038; 0.78 s/step; the job: 180.5 min spent, about 21.2 min of steps left; card peak 14.5 GB", "[10865.8s] stage 8/8 s8_register step 2250/4000: train bpb s8 0.0233; quiet KL s8 0.00036; 0.78 s/step; the job: 181.1 min spent, about 20.6 min of steps left; card peak 14.5 GB", "[10905.1s] stage 8/8 s8_register step 2300/4000: train bpb s8 0.0232; quiet KL s8 0.00037; 0.78 s/step; the job: 181.8 min spent, about 20.0 min of steps left; card peak 14.5 GB", "[10943.8s] stage 8/8 s8_register step 2350/4000: train bpb s8 0.0233; quiet KL s8 0.00036; 0.78 s/step; the job: 182.4 min spent, about 19.4 min of steps left; card peak 14.5 GB", "[10983.1s] stage 8/8 s8_register step 2400/4000: train bpb s8 0.0236; quiet KL s8 0.00035; 0.78 s/step; the job: 183.1 min spent, about 18.9 min of steps left; card peak 14.5 GB", "[11022.0s] stage 8/8 s8_register step 2450/4000: train bpb s8 0.0238; quiet KL s8 0.00036; 0.78 s/step; the job: 183.7 min spent, about 18.3 min of steps left; card peak 14.5 GB", "[11061.2s] stage 8/8 s8_register step 2500/4000: train bpb s8 0.0230; quiet KL s8 0.00036; 0.78 s/step; the job: 184.4 min spent, about 17.7 min of steps left; card peak 14.5 GB", "[11100.0s] stage 8/8 s8_register step 2550/4000: train bpb s8 0.0230; quiet KL s8 0.00033; 0.78 s/step; the job: 185.0 min spent, about 17.1 min of steps left; card peak 14.5 GB", "[11138.8s] stage 8/8 s8_register step 2600/4000: train bpb s8 0.0235; quiet KL s8 0.00032; 0.78 s/step; the job: 185.6 min spent, about 16.5 min of steps left; card peak 14.5 GB", "[11177.2s] stage 8/8 s8_register step 2650/4000: train bpb s8 0.0233; quiet KL s8 0.00038; 0.78 s/step; the job: 186.3 min spent, about 15.9 min of steps left; card peak 14.5 GB", "[11215.9s] stage 8/8 s8_register step 2700/4000: train bpb s8 0.0227; quiet KL s8 0.00035; 0.78 s/step; the job: 186.9 min spent, about 15.3 min of steps left; card peak 14.5 GB", "[11254.3s] stage 8/8 s8_register step 2750/4000: train bpb s8 0.0233; quiet KL s8 0.00035; 0.78 s/step; the job: 187.6 min spent, about 14.8 min of steps left; card peak 14.5 GB", "[11293.0s] stage 8/8 s8_register step 2800/4000: train bpb s8 0.0223; quiet KL s8 0.00031; 0.78 s/step; the job: 188.2 min spent, about 14.2 min of steps left; card peak 14.5 GB", "[11331.4s] stage 8/8 s8_register step 2850/4000: train bpb s8 0.0227; quiet KL s8 0.00035; 0.78 s/step; the job: 188.9 min spent, about 13.6 min of steps left; card peak 14.5 GB", "[11370.0s] stage 8/8 s8_register step 2900/4000: train bpb s8 0.0228; quiet KL s8 0.00037; 0.78 s/step; the job: 189.5 min spent, about 13.0 min of steps left; card peak 14.5 GB", "[11408.4s] stage 8/8 s8_register step 2950/4000: train bpb s8 0.0245; quiet KL s8 0.00072; 0.78 s/step; the job: 190.1 min spent, about 12.4 min of steps left; card peak 14.5 GB", "[11447.1s] stage 8/8 s8_register step 3000/4000: train bpb s8 0.0237; quiet KL s8 0.00042; 0.78 s/step; the job: 190.8 min spent, about 11.8 min of steps left; card peak 14.5 GB", "[11485.5s] stage 8/8 s8_register step 3050/4000: train bpb s8 0.0232; quiet KL s8 0.00040; 0.78 s/step; the job: 191.4 min spent, about 11.2 min of steps left; card peak 14.5 GB", "[11524.2s] stage 8/8 s8_register step 3100/4000: train bpb s8 0.0227; quiet KL s8 0.00035; 0.78 s/step; the job: 192.1 min spent, about 10.7 min of steps left; card peak 14.5 GB", "[11562.6s] stage 8/8 s8_register step 3150/4000: train bpb s8 0.0221; quiet KL s8 0.00031; 0.78 s/step; the job: 192.7 min spent, about 10.1 min of steps left; card peak 14.5 GB", "[11601.0s] stage 8/8 s8_register step 3200/4000: train bpb s8 0.0231; quiet KL s8 0.00035; 0.78 s/step; the job: 193.3 min spent, about 9.5 min of steps left; card peak 14.5 GB", "[11639.7s] stage 8/8 s8_register step 3250/4000: train bpb s8 0.0227; quiet KL s8 0.00033; 0.78 s/step; the job: 194.0 min spent, about 8.9 min of steps left; card peak 14.5 GB", "[11678.2s] stage 8/8 s8_register step 3300/4000: train bpb s8 0.0226; quiet KL s8 0.00032; 0.78 s/step; the job: 194.6 min spent, about 8.3 min of steps left; card peak 14.5 GB", "[11716.9s] stage 8/8 s8_register step 3350/4000: train bpb s8 0.0228; quiet KL s8 0.00034; 0.78 s/step; the job: 195.3 min spent, about 7.7 min of steps left; card peak 14.5 GB", "[11755.3s] stage 8/8 s8_register step 3400/4000: train bpb s8 0.0221; quiet KL s8 0.00032; 0.78 s/step; the job: 195.9 min spent, about 7.1 min of steps left; card peak 14.5 GB", "[11794.0s] stage 8/8 s8_register step 3450/4000: train bpb s8 0.0227; quiet KL s8 0.00034; 0.78 s/step; the job: 196.6 min spent, about 6.5 min of steps left; card peak 14.5 GB", "[11832.4s] stage 8/8 s8_register step 3500/4000: train bpb s8 0.0224; quiet KL s8 0.00034; 0.77 s/step; the job: 197.2 min spent, about 5.9 min of steps left; card peak 14.5 GB", "[11871.2s] stage 8/8 s8_register step 3550/4000: train bpb s8 0.0224; quiet KL s8 0.00035; 0.77 s/step; the job: 197.9 min spent, about 5.3 min of steps left; card peak 14.5 GB", "[11909.7s] stage 8/8 s8_register step 3600/4000: train bpb s8 0.0227; quiet KL s8 0.00035; 0.77 s/step; the job: 198.5 min spent, about 4.7 min of steps left; card peak 14.5 GB", "[11948.4s] stage 8/8 s8_register step 3650/4000: train bpb s8 0.0227; quiet KL s8 0.00030; 0.77 s/step; the job: 199.1 min spent, about 4.2 min of steps left; card peak 14.5 GB", "[11986.9s] stage 8/8 s8_register step 3700/4000: train bpb s8 0.0222; quiet KL s8 0.00033; 0.77 s/step; the job: 199.8 min spent, about 3.6 min of steps left; card peak 14.5 GB", "[12025.7s] stage 8/8 s8_register step 3750/4000: train bpb s8 0.0221; quiet KL s8 0.00035; 0.77 s/step; the job: 200.4 min spent, about 3.0 min of steps left; card peak 14.5 GB", "[12064.1s] stage 8/8 s8_register step 3800/4000: train bpb s8 0.0224; quiet KL s8 0.00034; 0.77 s/step; the job: 201.1 min spent, about 2.4 min of steps left; card peak 14.5 GB", "[12102.9s] stage 8/8 s8_register step 3850/4000: train bpb s8 0.0230; quiet KL s8 0.00042; 0.77 s/step; the job: 201.7 min spent, about 1.8 min of steps left; card peak 14.5 GB", "[12141.5s] stage 8/8 s8_register step 3900/4000: train bpb s8 0.0221; quiet KL s8 0.00035; 0.77 s/step; the job: 202.4 min spent, about 1.2 min of steps left; card peak 14.5 GB", "[12180.3s] stage 8/8 s8_register step 3950/4000: train bpb s8 0.0224; quiet KL s8 0.00033; 0.77 s/step; the job: 203.0 min spent, about 0.6 min of steps left; card peak 14.5 GB", "[12218.9s] stage 8/8 s8_register step 4000/4000: train bpb s8 0.0225; quiet KL s8 0.00034; 0.77 s/step; the job: 203.6 min spent, about 0.0 min of steps left; card peak 14.5 GB", "[12220.1s] anchors saved: /root/refit/group/anchors/gXB_close8_.safetensors for ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail', 's7_mixed', 's8_register']", "[12467.9s] CLOSE 8/8: the group ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail', 's7_mixed', 's8_register'] all on, against arms off:", "[12467.9s] s1_perspective rows: off 0.5538 -> all on 0.0484 (-0.5055 +- 0.0048); with s1_perspective masked 0.4889 (its share -0.4405); s1_perspective alone 0.0522 (-0.5017); gate mean 0.212 / widest 0.555", "[12467.9s] s2_concept rows: off 0.7212 -> all on 0.1212 (-0.6000 +- 0.0038); with s2_concept masked 0.2402 (its share -0.1190); s2_concept alone 0.3603 (-0.3608); gate mean 0.273 / widest 0.487", "[12467.9s] s3_rules rows: off 0.9296 -> all on 0.2222 (-0.7074 +- 0.0055); with s3_rules masked 0.3137 (its share -0.0915); s3_rules alone 0.8702 (-0.0594); gate mean 0.124 / widest 0.175", "[12467.9s] s4_arith rows: off 1.2339 -> all on 0.3266 (-0.9073 +- 0.0062); with s4_arith masked 0.4909 (its share -0.1643); s4_arith alone 1.1550 (-0.0789); gate mean 0.108 / widest 0.167", "[12467.9s] s5_causal rows: off 0.6604 -> all on 0.0428 (-0.6176 +- 0.0069); with s5_causal masked 0.4673 (its share -0.4245); s5_causal alone 0.1418 (-0.5187); gate mean 0.186 / widest 0.360", "[12467.9s] s6_tryfail rows: off 0.6728 -> all on 0.0543 (-0.6185 +- 0.0036); with s6_tryfail masked 0.4663 (its share -0.4120); s6_tryfail alone 0.1437 (-0.5291); gate mean 0.229 / widest 0.581", "[12467.9s] s7_mixed rows: off 0.7830 -> all on 0.7198 (-0.0632 +- 0.0390); with s7_mixed masked 0.7767 (its share -0.0569); s7_mixed alone 0.7254 (-0.0576); gate mean 0.253 / widest 0.467", "[12467.9s] s8_register rows: off 0.4907 -> all on 0.0208 (-0.4699 +- 0.0086); with s8_register masked 0.3781 (its share -0.3573); s8_register alone 0.0516 (-0.4390); gate mean 0.205 / widest 0.470", "[12467.9s] web holdout: off 0.9506 -> all on 0.9516 (+0.00103 +- 0.00020; the quiet bar +.012); each member's share: s1_perspective +0.00043, s2_concept -0.00073, s3_rules +0.00059, s4_arith +0.00017, s5_causal +0.00029, s6_tryfail +0.00025, s7_mixed -0.00070, s8_register +0.00046; alone: s1_perspective +0.00134, s2_concept +0.00086, s3_rules +0.00034, s4_arith +0.00006, s5_causal +0.00000, s6_tryfail +0.00011, s7_mixed -0.00032, s8_register +0.00005", "[12467.9s] a member's share on a PARTNER's rows (all on minus that member masked; near 0 = it stays out): s1_perspective on s2_concept -0.2289; s1_perspective on s3_rules -0.1943; s1_perspective on s4_arith -0.1300; s1_perspective on s5_causal -0.0282; s1_perspective on s6_tryfail -0.0206; s1_perspective on s7_mixed -0.0043; s1_perspective on s8_register -0.0214; s2_concept on s1_perspective -0.0042; s2_concept on s3_rules -0.1006; s2_concept on s4_arith -0.0957; s2_concept on s5_causal -0.0009; s2_concept on s6_tryfail -0.0023; s2_concept on s7_mixed -0.0004; s2_concept on s8_register -0.0005; s3_rules on s1_perspective +0.0000; s3_rules on s2_concept -0.0005; s3_rules on s4_arith -0.0004; s3_rules on s5_causal -0.0001; s3_rules on s6_tryfail -0.0003; s3_rules on s7_mixed -0.0001; s3_rules on s8_register +0.0000; s4_arith on s1_perspective +0.0000; s4_arith on s2_concept -0.0002; s4_arith on s3_rules +0.0001; s4_arith on s5_causal -0.0002; s4_arith on s6_tryfail -0.0039; s4_arith on s7_mixed -0.0000; s4_arith on s8_register -0.0001; s5_causal on s1_perspective +0.0001; s5_causal on s2_concept +0.0002; s5_causal on s3_rules +0.0004; s5_causal on s4_arith +0.0001; s5_causal on s6_tryfail -0.0003; s5_causal on s7_mixed +0.0001; s5_causal on s8_register -0.0001; s6_tryfail on s1_perspective +0.0000; s6_tryfail on s2_concept +0.0003; s6_tryfail on s3_rules +0.0000; s6_tryfail on s4_arith -0.0016; s6_tryfail on s5_causal +0.0000; s6_tryfail on s7_mixed -0.0001; s6_tryfail on s8_register -0.0005; s7_mixed on s1_perspective +0.0001; s7_mixed on s2_concept -0.0003; s7_mixed on s3_rules -0.0007; s7_mixed on s4_arith -0.0099; s7_mixed on s5_causal +0.0000; s7_mixed on s6_tryfail -0.0003; s7_mixed on s8_register -0.0002; s8_register on s1_perspective +0.0001; s8_register on s2_concept +0.0001; s8_register on s3_rules +0.0005; s8_register on s4_arith +0.0001; s8_register on s5_causal +0.0002; s8_register on s6_tryfail +0.0001; s8_register on s7_mixed +0.0002", "[12467.9s] a member ALONE on a partner's rows (alone minus off): s1_perspective on s2_concept -0.4772; s1_perspective on s3_rules -0.4262; s1_perspective on s4_arith -0.4807; s1_perspective on s5_causal -0.1919; s1_perspective on s6_tryfail -0.1690; s1_perspective on s7_mixed -0.0088; s1_perspective on s8_register -0.1105; s2_concept on s1_perspective -0.0300; s2_concept on s3_rules -0.3138; s2_concept on s4_arith -0.3577; s2_concept on s5_causal -0.0127; s2_concept on s6_tryfail -0.0062; s2_concept on s7_mixed +0.0002; s2_concept on s8_register -0.0193; s3_rules on s1_perspective -0.0003; s3_rules on s2_concept +0.0016; s3_rules on s4_arith -0.0047; s3_rules on s5_causal +0.0060; s3_rules on s6_tryfail -0.0011; s3_rules on s7_mixed -0.0007; s3_rules on s8_register -0.0044; s4_arith on s1_perspective -0.0013; s4_arith on s2_concept -0.0015; s4_arith on s3_rules +0.0005; s4_arith on s5_causal -0.0011; s4_arith on s6_tryfail -0.0134; s4_arith on s7_mixed -0.0003; s4_arith on s8_register -0.0014; s5_causal on s1_perspective -0.0022; s5_causal on s2_concept -0.0010; s5_causal on s3_rules -0.0018; s5_causal on s4_arith -0.0022; s5_causal on s6_tryfail -0.0023; s5_causal on s7_mixed -0.0001; s5_causal on s8_register -0.0009; s6_tryfail on s1_perspective +0.0029; s6_tryfail on s2_concept +0.0013; s6_tryfail on s3_rules -0.0003; s6_tryfail on s4_arith -0.0010; s6_tryfail on s5_causal -0.0026; s6_tryfail on s7_mixed +0.0001; s6_tryfail on s8_register -0.0001; s7_mixed on s1_perspective -0.0237; s7_mixed on s2_concept -0.0022; s7_mixed on s3_rules -0.0529; s7_mixed on s4_arith -0.0414; s7_mixed on s5_causal -0.0976; s7_mixed on s6_tryfail -0.0774; s7_mixed on s8_register -0.0061; s8_register on s1_perspective -0.0045; s8_register on s2_concept -0.0050; s8_register on s3_rules -0.0063; s8_register on s4_arith -0.0064; s8_register on s5_causal -0.0139; s8_register on s6_tryfail -0.0187; s8_register on s7_mixed +0.0001", "[12467.9s] s1_perspective battery accuracy, off -> all on: in 0.550 -> 1.000, transfer 0.463 -> 0.425; alone: in 0.925, transfer 0.463", "[12467.9s] s2_concept battery accuracy, off -> all on: in 0.298 -> 0.952, transfer 0.678 -> 0.720; alone: in 0.905, transfer 0.703", "[12467.9s] s3_rules battery accuracy, off -> all on: in 0.155 -> 0.920, transfer 0.070 -> 0.370, held-out 0.050 -> 0.458; alone: in 0.335, transfer 0.100, held-out 0.142", "[12467.9s] s4_arith battery accuracy, off -> all on: in 0.344 -> 0.656, transfer 0.242 -> 0.271, held-out 0.350 -> 0.325; alone: in 0.412, transfer 0.237, held-out 0.400", "[12467.9s] probes mean: off 0.411 -> all on 0.430", "[12467.9s] compiled chain: 28 cache entries (limit 256), 0 limit warnings; card peak 14.5 GB; host RSS 11.9 GB", "[12471.8s] the write of s1_perspective (mean |delta| / |x|, all members live): own rows 1.64e-02; web 6.14e-03 (own over web 2.67x); partners' rows s2_concept 1.31e-02, s3_rules 1.50e-02, s4_arith 1.57e-02, s5_causal 1.42e-02, s6_tryfail 1.41e-02, s7_mixed 6.09e-03, s8_register 1.26e-02", "[12471.8s] the write of s2_concept (mean |delta| / |x|, all members live): own rows 1.12e-02; web 4.23e-03 (own over web 2.65x); partners' rows s1_perspective 8.16e-03, s3_rules 1.11e-02, s4_arith 1.14e-02, s5_causal 8.44e-03, s6_tryfail 8.74e-03, s7_mixed 4.29e-03, s8_register 8.23e-03", "[12471.8s] the write of s3_rules (mean |delta| / |x|, all members live): own rows 8.97e-03; web 1.50e-03 (own over web 5.97x); partners' rows s1_perspective 1.89e-03, s2_concept 1.94e-03, s4_arith 1.52e-03, s5_causal 2.20e-03, s6_tryfail 1.99e-03, s7_mixed 1.47e-03, s8_register 1.81e-03", "[12471.8s] the write of s4_arith (mean |delta| / |x|, all members live): own rows 8.04e-03; web 1.23e-03 (own over web 6.55x); partners' rows s1_perspective 1.64e-03, s2_concept 1.54e-03, s3_rules 1.63e-03, s5_causal 1.99e-03, s6_tryfail 3.98e-03, s7_mixed 1.24e-03, s8_register 1.79e-03", "[12471.8s] the write of s5_causal (mean |delta| / |x|, all members live): own rows 1.49e-02; web 3.35e-03 (own over web 4.43x); partners' rows s1_perspective 4.64e-03, s2_concept 3.67e-03, s3_rules 4.01e-03, s4_arith 3.97e-03, s6_tryfail 4.55e-03, s7_mixed 3.37e-03, s8_register 4.76e-03", "[12471.8s] the write of s6_tryfail (mean |delta| / |x|, all members live): own rows 2.06e-02; web 4.95e-03 (own over web 4.16x); partners' rows s1_perspective 6.90e-03, s2_concept 5.40e-03, s3_rules 5.94e-03, s4_arith 6.91e-03, s5_causal 7.94e-03, s7_mixed 4.94e-03, s8_register 7.25e-03", "[12471.8s] the write of s7_mixed (mean |delta| / |x|, all members live): own rows 3.75e-03; web 3.84e-03 (own over web 0.98x); partners' rows s1_perspective 6.99e-03, s2_concept 5.11e-03, s3_rules 7.15e-03, s4_arith 6.53e-03, s5_causal 6.55e-03, s6_tryfail 6.59e-03, s8_register 5.28e-03", "[12471.8s] the write of s8_register (mean |delta| / |x|, all members live): own rows 1.61e-02; web 3.78e-03 (own over web 4.26x); partners' rows s1_perspective 5.60e-03, s2_concept 4.38e-03, s3_rules 4.42e-03, s4_arith 4.65e-03, s5_causal 6.37e-03, s6_tryfail 5.93e-03, s7_mixed 3.74e-03"], "write_matrix": {"s1_perspective": {"s1_perspective": 0.016407812356192153, "s2_concept": 0.013148073403499438, "s3_rules": 0.01496687993494561, "s4_arith": 0.015742732775834156, "s5_causal": 0.01423543636337854, "s6_tryfail": 0.01406089482770767, "s7_mixed": 0.0060945838604311575, "s8_register": 0.012577386100019794, "web": 0.0061354582157946425}, "s2_concept": {"s1_perspective": 0.00816054860672466, "s2_concept": 0.011237053117810092, "s3_rules": 0.011118454229517738, "s4_arith": 0.011361750353160005, "s5_causal": 0.008444922649005093, "s6_tryfail": 0.008738936410736642, "s7_mixed": 0.004292263970199883, "s8_register": 0.008233623639512189, "web": 0.004234066907173428}, "s3_rules": {"s1_perspective": 0.0018889328348450363, "s2_concept": 0.0019443760966169066, "s3_rules": 0.008971445154656976, "s4_arith": 0.0015229402324621333, "s5_causal": 0.002200605088546581, "s6_tryfail": 0.0019919419346479117, "s7_mixed": 0.0014710288423884776, "s8_register": 0.0018111583221980254, "web": 0.001502687053630325}, "s4_arith": {"s1_perspective": 0.0016448525429950678, "s2_concept": 0.0015432865905040671, "s3_rules": 0.0016303738225360576, "s4_arith": 0.008038323701839545, "s5_causal": 0.0019917135577998124, "s6_tryfail": 0.0039813058137951884, "s7_mixed": 0.001239521655406861, "s8_register": 0.00178723558474303, "web": 0.001227573652386127}, "s5_causal": {"s1_perspective": 0.004642169897124404, "s2_concept": 0.003670334017897403, "s3_rules": 0.00400549333608069, "s4_arith": 0.003966731132095447, "s5_causal": 0.014856689507723786, "s6_tryfail": 0.004549177976514329, "s7_mixed": 0.003368857461055086, "s8_register": 0.004758934819619753, "web": 0.0033513079695239867}, "s6_tryfail": {"s1_perspective": 0.006902354203702998, "s2_concept": 0.005396462267071911, "s3_rules": 0.005935972416409641, "s4_arith": 0.006907044384206529, "s5_causal": 0.007936711757793091, "s6_tryfail": 0.02059511511833989, "s7_mixed": 0.004935832144838059, "s8_register": 0.007252299757965375, "web": 0.00495316595197437}, "s7_mixed": {"s1_perspective": 0.006985741669268464, "s2_concept": 0.005107633487568819, "s3_rules": 0.007149678403948201, "s4_arith": 0.006534146387821238, "s5_causal": 0.006550839418196119, "s6_tryfail": 0.006587349109395291, "s7_mixed": 0.0037504364399865153, "s8_register": 0.0052752542032976635, "web": 0.003835762816834176}, "s8_register": {"s1_perspective": 0.0055966403961065225, "s2_concept": 0.004383555567528674, "s3_rules": 0.004417165027916781, "s4_arith": 0.00465424353023991, "s5_causal": 0.00636857239442179, "s6_tryfail": 0.005933548196480842, "s7_mixed": 0.0037431221721817565, "s8_register": 0.016130906937178224, "web": 0.0037828754370821116}}, "finished": true} \ No newline at end of file diff --git a/mini-beatrix-3/arm_refit/group/results/lines_gX_close8.json b/mini-beatrix-3/arm_refit/group/results/lines_gX_close8.json new file mode 100644 index 0000000000000000000000000000000000000000..64c8dbeb25e259b235233cafb82624aae7ec891d --- /dev/null +++ b/mini-beatrix-3/arm_refit/group/results/lines_gX_close8.json @@ -0,0 +1,872 @@ +{ + "A": { + "run": "gXA", + "close": 8, + "stages": { + "s1_perspective": { + "off": 0.5538333013185592, + "all_on": 0.049211667943427066, + "gain": 0.5046216333751322, + "own_share": 0.3832365148723068, + "share_of_gain": 0.7594532012213822, + "alone": 0.05251879489204885, + "alone_gain": 0.5013145064265104, + "solo_gain_800": 0.4736533805080021, + "group_over_solo": 1.0653816781248686, + "gate_mean": 0.2099039478571495, + "gate_max": 0.5712027580899203 + }, + "s2_concept": { + "off": 0.7211839326967642, + "all_on": 0.11045867104531781, + "gain": 0.6107252616514464, + "own_share": 0.12024064920374317, + "share_of_gain": 0.19688173513341095, + "alone": 0.35026574371684355, + "alone_gain": 0.3709181889799206, + "solo_gain_800": 0.6012941939378897, + "group_over_solo": 1.0156846146339653, + "gate_mean": 0.27674460361666475, + "gate_max": 0.4977012170865557 + }, + "s3_rules": { + "off": 0.9296131361815035, + "all_on": 0.2237152122866791, + "gain": 0.7058979238948244, + "own_share": 0.09000000454687146, + "share_of_gain": 0.12749719399979573, + "alone": 0.8236434891082838, + "alone_gain": 0.10596964707321976, + "solo_gain_800": 0.5558898295491429, + "group_over_solo": 1.2698522015906395, + "gate_mean": 0.12033936123137423, + "gate_max": 0.1738736117222015 + }, + "s4_arith": { + "off": 1.2338748670942346, + "all_on": 0.32794756855052143, + "gain": 0.9059272985437131, + "own_share": 0.14313402303310482, + "share_of_gain": 0.15799725128406455, + "alone": 1.126693075548818, + "alone_gain": 0.10718179154541652, + "solo_gain_800": 0.8454966312483596, + "group_over_solo": 1.0714735754844213, + "gate_mean": 0.10767731599767934, + "gate_max": 0.15932411779931022 + }, + "s5_causal": { + "off": 0.6604376826115352, + "all_on": 0.04305856436254439, + "gain": 0.6173791182489908, + "own_share": 0.4149917942760501, + "share_of_gain": 0.6721830752116282, + "alone": 0.09109720145841224, + "alone_gain": 0.569340481153123, + "solo_gain_800": 0.5923695896024492, + "group_over_solo": 1.042219467517443, + "gate_mean": 0.18518384792916404, + "gate_max": 0.3681551185947585 + }, + "s6_tryfail": { + "off": 0.6728212611401647, + "all_on": 0.05483553152977349, + "gain": 0.6179857296103912, + "own_share": 0.41378190338120613, + "share_of_gain": 0.6695654665716548, + "alone": 0.16377145340459293, + "alone_gain": 0.5090498077355718, + "solo_gain_800": 0.5586712149393783, + "group_over_solo": 1.1061707012727497, + "gate_mean": 0.23037076111374397, + "gate_max": 0.5373235543789295 + }, + "s7_mixed": { + "off": 0.7829759286650222, + "all_on": 0.719981644980245, + "gain": 0.06299428368477722, + "own_share": 0.05679246030767622, + "share_of_gain": 0.9015494261648428, + "alone": 0.7241658592632068, + "alone_gain": 0.058810069401815435, + "solo_gain_800": 0.027998275735886195, + "group_over_solo": 2.2499343987828375, + "gate_mean": 0.2511517146927655, + "gate_max": 0.463826973625139 + }, + "s8_register": { + "off": 0.4906504977761471, + "all_on": 0.02118320337400121, + "gain": 0.4694672944021459, + "own_share": 0.3545607166744309, + "share_of_gain": 0.7552405053603458, + "alone": 0.06783037737981043, + "alone_gain": 0.42282012039633665, + "solo_gain_800": 0.4396141805701983, + "group_over_solo": 1.067907531538739, + "gate_mean": 0.20274797688933244, + "gate_max": 0.4520048633318294 + } + }, + "web": { + "off": 0.9505967609111062, + "all_on": 0.9524873016985469, + "change": 0.0018905407874406643, + "member_shares": { + "s1_perspective": 0.0002528467435153753, + "s2_concept": -0.0004267749481160349, + "s3_rules": 9.024789604025751e-05, + "s4_arith": 0.00019910216512075696, + "s5_causal": 0.00035185500583397034, + "s6_tryfail": -0.00024723580953098234, + "s7_mixed": -0.0009560945518090752, + "s8_register": 0.0004544534059892774 + } + }, + "lines": { + "1_all_on_works": { + "s1_perspective": true, + "s2_concept": true, + "s3_rules": true, + "s4_arith": true, + "s5_causal": true, + "s6_tryfail": true, + "s7_mixed": true, + "s8_register": true + }, + "2_each_member_carries_its_stage": { + "s1_perspective": false, + "s2_concept": false, + "s3_rules": false, + "s4_arith": false, + "s5_causal": false, + "s6_tryfail": false, + "s7_mixed": true, + "s8_register": false + }, + "3_quiet": true, + "4_each_member_stays_out": { + "s1_perspective": true, + "s2_concept": true, + "s3_rules": true, + "s4_arith": true, + "s5_causal": true, + "s6_tryfail": true, + "s7_mixed": true, + "s8_register": true + }, + "5_survival": { + "s1_perspective": true, + "s2_concept": true, + "s3_rules": true, + "s4_arith": true, + "s5_causal": true, + "s6_tryfail": true, + "s7_mixed": true + } + }, + "partner_shares": { + "s1_perspective": { + "s2_concept": -0.21735774998500507, + "s3_rules": -0.18315040610285027, + "s4_arith": -0.12801225497551014, + "s5_causal": -0.012769430003430584, + "s6_tryfail": -0.028017711119048133, + "s7_mixed": -0.0028699766096466142, + "s8_register": -0.027170217061138438 + }, + "s2_concept": { + "s1_perspective": -0.005727841878552048, + "s3_rules": -0.1086460948305252, + "s4_arith": -0.10603784356874471, + "s5_causal": -0.0013008304164274188, + "s6_tryfail": -0.003474366640441262, + "s7_mixed": -0.0008228079973903665, + "s8_register": -0.0007025314938347883 + }, + "s3_rules": { + "s1_perspective": 7.567404589770878e-05, + "s2_concept": 0.0005395982543109808, + "s4_arith": -0.0001519816060132559, + "s5_causal": -0.00011356061462977282, + "s6_tryfail": -0.00028407099994818327, + "s7_mixed": 0.00010996946908219118, + "s8_register": -9.199996929592363e-05 + }, + "s4_arith": { + "s1_perspective": 1.874678075127357e-05, + "s2_concept": -0.0005979097782546922, + "s3_rules": -7.197205215716401e-05, + "s5_causal": -3.0840654504526825e-05, + "s6_tryfail": -0.003640219050330934, + "s7_mixed": 0.00014866556552631943, + "s8_register": -2.9318675225114936e-05 + }, + "s5_causal": { + "s1_perspective": 0.0006818527634420302, + "s2_concept": -0.0004982041800314398, + "s3_rules": 0.0004600240315398921, + "s4_arith": 0.0002281484225141045, + "s6_tryfail": -0.00047271111107717763, + "s7_mixed": 0.00038759515046637993, + "s8_register": -0.0002568485312322683 + }, + "s6_tryfail": { + "s1_perspective": -8.356677313857853e-05, + "s2_concept": 0.00016495689085217063, + "s3_rules": -0.00015880179301152458, + "s4_arith": -0.000901089663051094, + "s5_causal": -8.137634566538106e-05, + "s7_mixed": 0.00012510662958704205, + "s8_register": -0.0005701179529114088 + }, + "s7_mixed": { + "s1_perspective": -0.0003909192526400859, + "s2_concept": -0.0005283253289927077, + "s3_rules": -0.001002543303686633, + "s4_arith": -0.010266286677712871, + "s5_causal": -0.00015841651156515892, + "s6_tryfail": -0.00035021176534953333, + "s8_register": -5.3690665864426e-05 + }, + "s8_register": { + "s1_perspective": 0.0007188001456666475, + "s2_concept": -0.000785752454201755, + "s3_rules": 0.0004901478677300786, + "s4_arith": 6.941918468340857e-05, + "s5_causal": 0.00017443004460145507, + "s6_tryfail": 0.00024964356664304316, + "s7_mixed": 0.0012158582225638426 + } + }, + "survival": { + "s1_perspective": [ + { + "from_close": 4, + "to_close": 5, + "loss_frac": 0.0013851492749361704 + }, + { + "from_close": 5, + "to_close": 6, + "loss_frac": -0.00012215275085882178 + }, + { + "from_close": 6, + "to_close": 7, + "loss_frac": -0.000602886106127756 + }, + { + "from_close": 7, + "to_close": 8, + "loss_frac": 0.001422407743347584 + } + ], + "s2_concept": [ + { + "from_close": 4, + "to_close": 5, + "loss_frac": -0.0010328987080300191 + }, + { + "from_close": 5, + "to_close": 6, + "loss_frac": 0.0003895345762453068 + }, + { + "from_close": 6, + "to_close": 7, + "loss_frac": -0.0010456909592767348 + }, + { + "from_close": 7, + "to_close": 8, + "loss_frac": -0.001288246526669353 + } + ], + "s3_rules": [ + { + "from_close": 4, + "to_close": 5, + "loss_frac": 0.0007115904427607917 + }, + { + "from_close": 5, + "to_close": 6, + "loss_frac": -8.352652715358385e-05 + }, + { + "from_close": 6, + "to_close": 7, + "loss_frac": -0.0012160915878870514 + }, + { + "from_close": 7, + "to_close": 8, + "loss_frac": 0.0006938790267326601 + } + ], + "s4_arith": [ + { + "from_close": 4, + "to_close": 5, + "loss_frac": 0.000148594817064465 + }, + { + "from_close": 5, + "to_close": 6, + "loss_frac": -0.0006794155565193974 + }, + { + "from_close": 6, + "to_close": 7, + "loss_frac": -0.011188445022368616 + }, + { + "from_close": 7, + "to_close": 8, + "loss_frac": 7.662189423538891e-05 + } + ], + "s5_causal": [ + { + "from_close": 5, + "to_close": 6, + "loss_frac": -0.0002721223020169556 + }, + { + "from_close": 6, + "to_close": 7, + "loss_frac": -0.00021497324071434576 + }, + { + "from_close": 7, + "to_close": 8, + "loss_frac": 0.00028245331126907594 + } + ], + "s6_tryfail": [ + { + "from_close": 6, + "to_close": 7, + "loss_frac": -0.0005170999958412308 + }, + { + "from_close": 7, + "to_close": 8, + "loss_frac": 0.0004038001988791823 + } + ], + "s7_mixed": [ + { + "from_close": 7, + "to_close": 8, + "loss_frac": 0.018935610270390358 + } + ] + }, + "probes": { + "off": 0.41111111111111115, + "all_on": 0.4222222222222222 + }, + "battery": { + "s1_perspective": { + "acc_off": { + "in": 0.55, + "transfer": 0.4625 + }, + "acc_all_on": { + "in": 0.8875, + "transfer": 0.4375 + }, + "acc_member_alone": { + "in": 0.8375, + "transfer": 0.4625 + } + }, + "s2_concept": { + "acc_off": { + "in": 0.2976190476190476, + "transfer": 0.6779661016949152 + }, + "acc_all_on": { + "in": 0.9761904761904762, + "transfer": 0.7203389830508474 + }, + "acc_member_alone": { + "in": 0.9642857142857143, + "transfer": 0.7203389830508474 + } + }, + "s3_rules": { + "acc_off": { + "in": 0.155, + "transfer": 0.07, + "held-out": 0.05 + }, + "acc_all_on": { + "in": 0.855, + "transfer": 0.31, + "held-out": 0.43333333333333335 + }, + "acc_member_alone": { + "in": 0.32, + "transfer": 0.08, + "held-out": 0.10833333333333334 + } + }, + "s4_arith": { + "acc_off": { + "in": 0.34375, + "transfer": 0.24166666666666667, + "held-out": 0.35 + }, + "acc_all_on": { + "in": 0.64375, + "transfer": 0.24166666666666667, + "held-out": 0.425 + }, + "acc_member_alone": { + "in": 0.38125, + "transfer": 0.24583333333333332, + "held-out": 0.3875 + } + } + } + }, + "B": { + "run": "gXB", + "close": 8, + "stages": { + "s1_perspective": { + "off": 0.5538333013185592, + "all_on": 0.048373615296375044, + "gain": 0.5054596860221842, + "own_share": 0.4405018871614219, + "share_of_gain": 0.8714876761548273, + "alone": 0.05217516482213097, + "alone_gain": 0.5016581364964283, + "solo_gain_800": 0.4736533805080021, + "group_over_solo": 1.067151015538133, + "gate_mean": 0.21163099743248942, + "gate_max": 0.554569513244894 + }, + "s2_concept": { + "off": 0.7211839326967642, + "all_on": 0.12122929684668586, + "gain": 0.5999546358500784, + "own_share": 0.11900558535609002, + "share_of_gain": 0.1983576394696417, + "alone": 0.36033454325401637, + "alone_gain": 0.3608493894427478, + "solo_gain_800": 0.6012941939378897, + "group_over_solo": 0.9977722085107149, + "gate_mean": 0.27282737975382454, + "gate_max": 0.48724048599867004 + }, + "s3_rules": { + "off": 0.9296131361815035, + "all_on": 0.22223428311624405, + "gain": 0.7073788530652595, + "own_share": 0.09148521984743348, + "share_of_gain": 0.1293298767004468, + "alone": 0.8701872076558641, + "alone_gain": 0.05942592852563944, + "solo_gain_800": 0.5558898295491429, + "group_over_solo": 1.2725162711449183, + "gate_mean": 0.12374002504886239, + "gate_max": 0.17456351278495152 + }, + "s4_arith": { + "off": 1.2338748670942346, + "all_on": 0.3266083503971684, + "gain": 0.9072665166970661, + "own_share": 0.16432224262374015, + "share_of_gain": 0.1811179400976471, + "alone": 1.1550061816504822, + "alone_gain": 0.07886868544375236, + "solo_gain_800": 0.8454966312483596, + "group_over_solo": 1.0730575181092141, + "gate_mean": 0.10761241921992425, + "gate_max": 0.16724209369972712 + }, + "s5_causal": { + "off": 0.6604376826115352, + "all_on": 0.042796246816605044, + "gain": 0.6176414357949302, + "own_share": 0.4245476673136512, + "share_of_gain": 0.6873691477114722, + "alone": 0.1417724646021276, + "alone_gain": 0.5186652180094076, + "solo_gain_800": 0.5923695896024492, + "group_over_solo": 1.0426622950199746, + "gate_mean": 0.18640250725757604, + "gate_max": 0.3602895342887907 + }, + "s6_tryfail": { + "off": 0.6728212611401647, + "all_on": 0.05430685618921398, + "gain": 0.6185144049509508, + "own_share": 0.4120385804147112, + "share_of_gain": 0.6661745904646902, + "alone": 0.14373820614727262, + "alone_gain": 0.5290830549928921, + "solo_gain_800": 0.5586712149393783, + "group_over_solo": 1.1071170098106202, + "gate_mean": 0.2293551327826076, + "gate_max": 0.580794738809484 + }, + "s7_mixed": { + "off": 0.7829759286650222, + "all_on": 0.7197570947572544, + "gain": 0.06321883390776784, + "own_share": 0.056922807033656736, + "share_of_gain": 0.9004090002150847, + "alone": 0.7253837275330026, + "alone_gain": 0.05759220113201957, + "solo_gain_800": 0.027998275735886195, + "group_over_solo": 2.257954543491349, + "gate_mean": 0.2532132549279672, + "gate_max": 0.46651751205856823 + }, + "s8_register": { + "off": 0.4906504977761471, + "all_on": 0.020783710349540914, + "gain": 0.4698667874266062, + "own_share": 0.35733589204658456, + "share_of_gain": 0.7605046826222015, + "alone": 0.05160327413789158, + "alone_gain": 0.43904722363825555, + "solo_gain_800": 0.4396141805701983, + "group_over_solo": 1.0688162670666561, + "gate_mean": 0.20503783160560274, + "gate_max": 0.4696574997215004 + } + }, + "web": { + "off": 0.9505967609111062, + "all_on": 0.9516240885271194, + "change": 0.0010273276160132072, + "member_shares": { + "s1_perspective": 0.000430085614145162, + "s2_concept": -0.00072688467387165, + "s3_rules": 0.0005892018129402959, + "s4_arith": 0.00017075727447540823, + "s5_causal": 0.000285383711275497, + "s6_tryfail": 0.0002547707994218912, + "s7_mixed": -0.0007026458690156806, + "s8_register": 0.000457076141414936 + } + }, + "lines": { + "1_all_on_works": { + "s1_perspective": true, + "s2_concept": true, + "s3_rules": true, + "s4_arith": true, + "s5_causal": true, + "s6_tryfail": true, + "s7_mixed": true, + "s8_register": true + }, + "2_each_member_carries_its_stage": { + "s1_perspective": true, + "s2_concept": false, + "s3_rules": false, + "s4_arith": false, + "s5_causal": false, + "s6_tryfail": false, + "s7_mixed": true, + "s8_register": false + }, + "3_quiet": true, + "4_each_member_stays_out": { + "s1_perspective": true, + "s2_concept": true, + "s3_rules": true, + "s4_arith": true, + "s5_causal": true, + "s6_tryfail": true, + "s7_mixed": true, + "s8_register": true + }, + "5_survival": { + "s1_perspective": true, + "s2_concept": true, + "s3_rules": true, + "s4_arith": true, + "s5_causal": true, + "s6_tryfail": true, + "s7_mixed": true + } + }, + "partner_shares": { + "s1_perspective": { + "s2_concept": -0.22886177967760402, + "s3_rules": -0.19434419521666627, + "s4_arith": -0.13002179163397448, + "s5_causal": -0.028244990894813943, + "s6_tryfail": -0.020635761602328953, + "s7_mixed": -0.004294089699031799, + "s8_register": -0.021379535509961124 + }, + "s2_concept": { + "s1_perspective": -0.004184658347540812, + "s3_rules": -0.10059162865543836, + "s4_arith": -0.09568419239162246, + "s5_causal": -0.0008903840789384939, + "s6_tryfail": -0.002257523548478361, + "s7_mixed": -0.00037046406610310306, + "s8_register": -0.0005194366475334668 + }, + "s3_rules": { + "s1_perspective": 4.275448392015643e-05, + "s2_concept": -0.0005125459207760454, + "s4_arith": -0.00037896377117621616, + "s5_causal": -0.00014991311155228455, + "s6_tryfail": -0.0002914884235738725, + "s7_mixed": -0.00013408734863678412, + "s8_register": 1.4759940745239127e-05 + }, + "s4_arith": { + "s1_perspective": 1.474986363679022e-05, + "s2_concept": -0.00017278478869534736, + "s3_rules": 0.00011795322620268556, + "s5_causal": -0.00023599479505611535, + "s6_tryfail": -0.00393898180259052, + "s7_mixed": -4.9963647304585024e-05, + "s8_register": -5.915229069197118e-05 + }, + "s5_causal": { + "s1_perspective": 8.648611145625068e-05, + "s2_concept": 0.00019777736126330887, + "s3_rules": 0.0004248562666673665, + "s4_arith": 9.970962906662162e-05, + "s6_tryfail": -0.00032014436096665985, + "s7_mixed": 5.702031044778344e-05, + "s8_register": -6.247202611868495e-05 + }, + "s6_tryfail": { + "s1_perspective": 4.160770897866043e-05, + "s2_concept": 0.00033898989738073704, + "s3_rules": 2.885815136898992e-05, + "s4_arith": -0.0015790130260052507, + "s5_causal": 4.245385018476072e-05, + "s7_mixed": -0.0001090181900446005, + "s8_register": -0.0004942220426760128 + }, + "s7_mixed": { + "s1_perspective": 6.147472828584976e-05, + "s2_concept": -0.00032884426459428573, + "s3_rules": -0.0007004933986509837, + "s4_arith": -0.009933408882510063, + "s5_causal": 4.162786319555911e-05, + "s6_tryfail": -0.00032225114843972835, + "s8_register": -0.0001553863250545475 + }, + "s8_register": { + "s1_perspective": 0.00014360013901260679, + "s2_concept": 7.064859191420123e-05, + "s3_rules": 0.0004899624489346187, + "s4_arith": 0.00012165891488298552, + "s5_causal": 0.00015175319155507055, + "s6_tryfail": 9.433584303439931e-05, + "s7_mixed": 0.00017436891014353112 + } + }, + "survival": { + "s1_perspective": [ + { + "from_close": 4, + "to_close": 5, + "loss_frac": 0.0001658717235832137 + }, + { + "from_close": 5, + "to_close": 6, + "loss_frac": 0.0003397902673758371 + }, + { + "from_close": 6, + "to_close": 7, + "loss_frac": 0.0002201609499956208 + }, + { + "from_close": 7, + "to_close": 8, + "loss_frac": 0.00028401741630852066 + } + ], + "s2_concept": [ + { + "from_close": 4, + "to_close": 5, + "loss_frac": 1.909461799511504e-05 + }, + { + "from_close": 5, + "to_close": 6, + "loss_frac": 0.0005461182940285014 + }, + { + "from_close": 6, + "to_close": 7, + "loss_frac": -0.0005876254296242997 + }, + { + "from_close": 7, + "to_close": 8, + "loss_frac": 0.00011774269142656566 + } + ], + "s3_rules": [ + { + "from_close": 4, + "to_close": 5, + "loss_frac": 0.0005470956157656302 + }, + { + "from_close": 5, + "to_close": 6, + "loss_frac": -0.00017980816081613094 + }, + { + "from_close": 6, + "to_close": 7, + "loss_frac": -0.0010671363786990155 + }, + { + "from_close": 7, + "to_close": 8, + "loss_frac": 0.0006921656078022235 + } + ], + "s4_arith": [ + { + "from_close": 4, + "to_close": 5, + "loss_frac": 6.215724030901868e-05 + }, + { + "from_close": 5, + "to_close": 6, + "loss_frac": -0.0017313569603456984 + }, + { + "from_close": 6, + "to_close": 7, + "loss_frac": -0.011023761338668215 + }, + { + "from_close": 7, + "to_close": 8, + "loss_frac": 0.00013407593150623602 + } + ], + "s5_causal": [ + { + "from_close": 5, + "to_close": 6, + "loss_frac": 7.135242187850962e-05 + }, + { + "from_close": 6, + "to_close": 7, + "loss_frac": 4.8903712970199415e-05 + }, + { + "from_close": 7, + "to_close": 8, + "loss_frac": 0.0002456375276716527 + } + ], + "s6_tryfail": [ + { + "from_close": 6, + "to_close": 7, + "loss_frac": -0.0005945380561072665 + }, + { + "from_close": 7, + "to_close": 8, + "loss_frac": 0.00015249678320618265 + } + ], + "s7_mixed": [ + { + "from_close": 7, + "to_close": 8, + "loss_frac": 0.002750593161295283 + } + ] + }, + "probes": { + "off": 0.41111111111111115, + "all_on": 0.4296296296296296 + }, + "battery": { + "s1_perspective": { + "acc_off": { + "in": 0.55, + "transfer": 0.4625 + }, + "acc_all_on": { + "in": 1.0, + "transfer": 0.425 + }, + "acc_member_alone": { + "in": 0.925, + "transfer": 0.4625 + } + }, + "s2_concept": { + "acc_off": { + "in": 0.2976190476190476, + "transfer": 0.6779661016949152 + }, + "acc_all_on": { + "in": 0.9523809523809523, + "transfer": 0.7203389830508474 + }, + "acc_member_alone": { + "in": 0.9047619047619048, + "transfer": 0.7033898305084746 + } + }, + "s3_rules": { + "acc_off": { + "in": 0.155, + "transfer": 0.07, + "held-out": 0.05 + }, + "acc_all_on": { + "in": 0.92, + "transfer": 0.37, + "held-out": 0.4583333333333333 + }, + "acc_member_alone": { + "in": 0.335, + "transfer": 0.1, + "held-out": 0.14166666666666666 + } + }, + "s4_arith": { + "acc_off": { + "in": 0.34375, + "transfer": 0.24166666666666667, + "held-out": 0.35 + }, + "acc_all_on": { + "in": 0.65625, + "transfer": 0.2708333333333333, + "held-out": 0.325 + }, + "acc_member_alone": { + "in": 0.4125, + "transfer": 0.2375, + "held-out": 0.4 + } + } + } + } +} \ No newline at end of file diff --git a/mini-beatrix-3/arm_refit/logs/chain10_card0.log b/mini-beatrix-3/arm_refit/logs/chain10_card0.log new file mode 100644 index 0000000000000000000000000000000000000000..fe91de1a8ad1b07306307e358b6affedd180585b --- /dev/null +++ b/mini-beatrix-3/arm_refit/logs/chain10_card0.log @@ -0,0 +1 @@ +[chain10 card 0] 01:07:04Z armed (4000 steps); waiting for chain9 to end on this card diff --git a/mini-beatrix-3/arm_refit/logs/chain10_card1.log b/mini-beatrix-3/arm_refit/logs/chain10_card1.log new file mode 100644 index 0000000000000000000000000000000000000000..043869123870acf060c1c6733fc5a8310ba87625 --- /dev/null +++ b/mini-beatrix-3/arm_refit/logs/chain10_card1.log @@ -0,0 +1 @@ +[chain10 card 1] 01:07:04Z armed (4000 steps); waiting for chain9 to end on this card diff --git a/mini-beatrix-3/arm_refit/logs/chain11_card0.log b/mini-beatrix-3/arm_refit/logs/chain11_card0.log new file mode 100644 index 0000000000000000000000000000000000000000..bc9d2b34856ccd01b3a156700ebe1441d029989e --- /dev/null +++ b/mini-beatrix-3/arm_refit/logs/chain11_card0.log @@ -0,0 +1 @@ +[chain11 card 0] 02:50:44Z armed (4000 steps); waiting for chain10 to end on this card diff --git a/mini-beatrix-3/arm_refit/logs/chain11_card1.log b/mini-beatrix-3/arm_refit/logs/chain11_card1.log new file mode 100644 index 0000000000000000000000000000000000000000..63191f1758e6df306c0158b08b9b85a470b64f8e --- /dev/null +++ b/mini-beatrix-3/arm_refit/logs/chain11_card1.log @@ -0,0 +1 @@ +[chain11 card 1] 02:50:44Z armed (4000 steps); waiting for chain10 to end on this card diff --git a/mini-beatrix-3/arm_refit/logs/chain7_card0.log b/mini-beatrix-3/arm_refit/logs/chain7_card0.log index 5ee2239de7768d3a87bca0b7a15a3601b6382459..096df5508e21393417f85cace7ecc4d561e513bd 100644 --- a/mini-beatrix-3/arm_refit/logs/chain7_card0.log +++ b/mini-beatrix-3/arm_refit/logs/chain7_card0.log @@ -3,3 +3,5 @@ [chain7 card 0] 00:39:21Z routed read start: gSA [chain7 card 0] 00:46:41Z routed read rc=0: gSA [chain7 card 0] 00:46:41Z long solo start: --arm=s3_rules --seed=3 --data-seed=20261005 --steps=3200 --reads=0,200,400,800,1200,1600,2000,2400,2800,3200 +[chain7 card 0] 01:15:18Z long solo rc=0: --arm=s3_rules --seed=3 --data-seed=20261005 --steps=3200 --reads=0,200,400,800,1200,1600,2000,2400,2800,3200 +[chain7 card 0] 01:15:18Z done diff --git a/mini-beatrix-3/arm_refit/logs/chain7_card1.log b/mini-beatrix-3/arm_refit/logs/chain7_card1.log index 75891c275fe7f8f2a955bec6026778d9195fef5e..ec735be7c9e06420ac6e3ecbffd3941c30552994 100644 --- a/mini-beatrix-3/arm_refit/logs/chain7_card1.log +++ b/mini-beatrix-3/arm_refit/logs/chain7_card1.log @@ -3,3 +3,5 @@ [chain7 card 1] 00:39:31Z routed read start: gSB [chain7 card 1] 00:46:49Z routed read rc=0: gSB [chain7 card 1] 00:46:49Z long solo start: --arm=s4_arith --seed=4 --data-seed=20261005 --steps=3200 --reads=0,200,400,800,1200,1600,2000,2400,2800,3200 +[chain7 card 1] 01:14:49Z long solo rc=0: --arm=s4_arith --seed=4 --data-seed=20261005 --steps=3200 --reads=0,200,400,800,1200,1600,2000,2400,2800,3200 +[chain7 card 1] 01:14:49Z done diff --git a/mini-beatrix-3/arm_refit/logs/chain8_card0.log b/mini-beatrix-3/arm_refit/logs/chain8_card0.log new file mode 100644 index 0000000000000000000000000000000000000000..ec91c5b05a5b751c78d5d14867b21396a18b2d5d --- /dev/null +++ b/mini-beatrix-3/arm_refit/logs/chain8_card0.log @@ -0,0 +1,6 @@ +[chain8 card 0] 01:00:08Z armed (4000 steps per stage); waiting for chain7 to end on this card +[chain8 card 0] 01:15:19Z smoke start: --stages=s1_perspective,s2_concept,s3_rules,s4_arith,s5_causal,s6_tryfail,s7_mixed,s8_register --resume=gRA:4 --resume-members=4 --freeze-earlier=1 --offdomain=roster --no-partner-rows-of=s7_mixed --seed-offset=0 --data-seed=20261011 --tag=gXA +[chain8 card 0] 01:19:08Z smoke rc=0 +[chain8 card 0] 01:19:09Z gXA start: --stages=s1_perspective,s2_concept,s3_rules,s4_arith,s5_causal,s6_tryfail,s7_mixed,s8_register --resume=gRA:4 --resume-members=4 --freeze-earlier=1 --offdomain=roster --no-partner-rows-of=s7_mixed --seed-offset=0 --data-seed=20261011 --tag=gXA --steps=4000 +[chain8 card 0] 04:42:54Z gXA rc=0 +[chain8 card 0] 04:42:54Z done diff --git a/mini-beatrix-3/arm_refit/logs/chain8_card1.log b/mini-beatrix-3/arm_refit/logs/chain8_card1.log new file mode 100644 index 0000000000000000000000000000000000000000..6e76dc96f996c72468024776809b2f905a7ac0dc --- /dev/null +++ b/mini-beatrix-3/arm_refit/logs/chain8_card1.log @@ -0,0 +1,6 @@ +[chain8 card 1] 01:00:08Z armed (4000 steps per stage); waiting for chain7 to end on this card +[chain8 card 1] 01:14:59Z smoke start: --stages=s1_perspective,s2_concept,s3_rules,s4_arith,s5_causal,s6_tryfail,s7_mixed,s8_register --resume=gRB:4 --resume-members=4 --freeze-earlier=1 --offdomain=roster --no-partner-rows-of=s7_mixed --seed-offset=100 --data-seed=20261012 --tag=gXB +[chain8 card 1] 01:19:08Z smoke rc=0 +[chain8 card 1] 01:19:08Z gXB start: --stages=s1_perspective,s2_concept,s3_rules,s4_arith,s5_causal,s6_tryfail,s7_mixed,s8_register --resume=gRB:4 --resume-members=4 --freeze-earlier=1 --offdomain=roster --no-partner-rows-of=s7_mixed --seed-offset=100 --data-seed=20261012 --tag=gXB --steps=4000 +[chain8 card 1] 04:47:02Z gXB rc=0 +[chain8 card 1] 04:47:02Z done diff --git a/mini-beatrix-3/arm_refit/logs/chain9_card0.log b/mini-beatrix-3/arm_refit/logs/chain9_card0.log new file mode 100644 index 0000000000000000000000000000000000000000..df5ba687b4119e2578fd8c83cd8c12e2c79ceaa7 --- /dev/null +++ b/mini-beatrix-3/arm_refit/logs/chain9_card0.log @@ -0,0 +1,4 @@ +[chain9 card 0] 01:02:29Z armed (4000 steps per stage); waiting for chain8 to end on this card +[chain9 card 0] 04:42:57Z smoke start: --stages=s1_perspective,s2_concept,s3_rules,s4_arith,s5_causal,s6_tryfail,s7_mixed,s8_register --freeze-earlier=1 --offdomain=roster --no-partner-rows-of=s7_mixed --seed-offset=0 --data-seed=20261013 --tag=gEA +[chain9 card 0] 04:46:32Z smoke rc=0 +[chain9 card 0] 04:46:32Z gEA start: --stages=s1_perspective,s2_concept,s3_rules,s4_arith,s5_causal,s6_tryfail,s7_mixed,s8_register --freeze-earlier=1 --offdomain=roster --no-partner-rows-of=s7_mixed --seed-offset=0 --data-seed=20261013 --tag=gEA --steps=4000 diff --git a/mini-beatrix-3/arm_refit/logs/chain9_card1.log b/mini-beatrix-3/arm_refit/logs/chain9_card1.log new file mode 100644 index 0000000000000000000000000000000000000000..6a83ccc7e7abf87d7400449d3e46901291536094 --- /dev/null +++ b/mini-beatrix-3/arm_refit/logs/chain9_card1.log @@ -0,0 +1,4 @@ +[chain9 card 1] 01:02:29Z armed (4000 steps per stage); waiting for chain8 to end on this card +[chain9 card 1] 04:47:07Z smoke start: --stages=s1_perspective,s2_concept,s3_rules,s4_arith,s5_causal,s6_tryfail,s7_mixed,s8_register --freeze-earlier=1 --offdomain=roster --no-partner-rows-of=s7_mixed --seed-offset=100 --data-seed=20261014 --tag=gEB +[chain9 card 1] 04:50:38Z smoke rc=0 +[chain9 card 1] 04:50:38Z gEB start: --stages=s1_perspective,s2_concept,s3_rules,s4_arith,s5_causal,s6_tryfail,s7_mixed,s8_register --freeze-earlier=1 --offdomain=roster --no-partner-rows-of=s7_mixed --seed-offset=100 --data-seed=20261014 --tag=gEB --steps=4000 diff --git a/mini-beatrix-3/arm_refit/logs/group_EA.log b/mini-beatrix-3/arm_refit/logs/group_EA.log new file mode 100644 index 0000000000000000000000000000000000000000..9b6411bd79cffd2a296fe8ca1fd8dad266c3eb9f --- /dev/null +++ b/mini-beatrix-3/arm_refit/logs/group_EA.log @@ -0,0 +1,41 @@ +[ 0.2s] THE JOB: gEA: the group ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail', 's7_mixed', 's8_register'], staged, EARLIER ARMS FROZEN (on, not training); 4000 steps per stage; each step = one task chunk + one abstention chunk of 2 x 4096 bytes per newest member; lambda 2.0; pure Adam lr 0.001; the compiled chain with recompute; trunk step 245,674; card 31.4 GB with 16.4 GB free; host RSS 0.6 GB; a member's partner rows come from every other member of the roster, attached or not +Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads. +[ 14.9s] mini-beatrix-3 @ 245,674: trunk loaded from /root/v3/runs/mission/mini-beatrix-3/checkpoints/step_00245674.safetensors and frozen (376.0M); amoe-lora 0.2.11, torch 2.8.0+cu128 +[stream] opening 'fineweb-good' (HuggingFaceFW/fineweb-edu) … +[stream] opening 'simple-wiki' (wikimedia/wikipedia) … +[stream] opening 'babi-prose' (Muennighoff/babi) … +Repo card metadata block was not found. Setting CardData to empty. +[stream] opening 'tinystories' (roneneldan/TinyStories) … +[stream] opening 'proofwriter-prose' (tasksource/proofwriter) … +[ 77.8s] ARMS OFF (the bare trunk): s1_perspective rows 0.5538; s2_concept rows 0.7212; s3_rules rows 0.9296; s4_arith rows 1.2339; s5_causal rows 0.6604; s6_tryfail rows 0.6728; s7_mixed rows 0.7830; s8_register rows 0.4907; web rows 0.9506; s1_perspective battery in 0.550, transfer 0.463; s2_concept battery in 0.298, transfer 0.678; s3_rules battery in 0.155, transfer 0.070, held-out 0.050; s4_arith battery in 0.344, transfer 0.242, held-out 0.350; probes mean 0.411 +[arms] attached 's1_perspective' for phase 'curriculum_s1': 13.69M params x 32 blocks, lambda 2.0, spec {'n_slots': 16, 'K': 16, 'D': 8, 'hidden': 256}, recompute on, compile on +[ 81.6s] STAGE 1/8: s1_perspective attached over the bare trunk; at birth the group reads exactly as before on its rows and on web text; 4000 steps of 1 task + 1 abstention chunks +[stream] opening 'fineweb-edu' (HuggingFaceFW/fineweb-edu) … +[ 94.7s] stage 1/8 s1_perspective step 1/4000: train bpb s1 0.5699; quiet KL s1 0.00000; 13.18 s/step; the job: 1.6 min spent, about 7031.2 min of steps left; card peak 13.6 GB +[ 118.8s] stage 1/8 s1_perspective step 50/4000: train bpb s1 0.5495; quiet KL s1 0.00050; 0.74 s/step; the job: 2.0 min spent, about 396.1 min of steps left; card peak 13.7 GB +[ 142.3s] stage 1/8 s1_perspective step 100/4000: train bpb s1 0.5024; quiet KL s1 0.00162; 0.61 s/step; the job: 2.4 min spent, about 322.7 min of steps left; card peak 13.7 GB +[ 165.8s] stage 1/8 s1_perspective step 150/4000: train bpb s1 0.4354; quiet KL s1 0.00136; 0.56 s/step; the job: 2.8 min spent, about 298.2 min of steps left; card peak 13.7 GB +[ 189.2s] stage 1/8 s1_perspective step 200/4000: train bpb s1 0.3557; quiet KL s1 0.00187; 0.54 s/step; the job: 3.2 min spent, about 285.1 min of steps left; card peak 13.7 GB +[ 212.7s] stage 1/8 s1_perspective step 250/4000: train bpb s1 0.3016; quiet KL s1 0.00197; 0.52 s/step; the job: 3.5 min spent, about 277.6 min of steps left; card peak 13.7 GB +[ 236.0s] stage 1/8 s1_perspective step 300/4000: train bpb s1 0.2624; quiet KL s1 0.00184; 0.51 s/step; the job: 3.9 min spent, about 272.0 min of steps left; card peak 13.7 GB +[ 259.5s] stage 1/8 s1_perspective step 350/4000: train bpb s1 0.2324; quiet KL s1 0.00145; 0.51 s/step; the job: 4.3 min spent, about 268.2 min of steps left; card peak 13.7 GB +[ 282.9s] stage 1/8 s1_perspective step 400/4000: train bpb s1 0.2032; quiet KL s1 0.00109; 0.50 s/step; the job: 4.7 min spent, about 265.1 min of steps left; card peak 13.7 GB +[ 306.2s] stage 1/8 s1_perspective step 450/4000: train bpb s1 0.1767; quiet KL s1 0.00090; 0.50 s/step; the job: 5.1 min spent, about 262.5 min of steps left; card peak 13.7 GB +[ 329.8s] stage 1/8 s1_perspective step 500/4000: train bpb s1 0.1563; quiet KL s1 0.00077; 0.50 s/step; the job: 5.5 min spent, about 260.6 min of steps left; card peak 13.7 GB +[ 353.2s] stage 1/8 s1_perspective step 550/4000: train bpb s1 0.1396; quiet KL s1 0.00074; 0.49 s/step; the job: 5.9 min spent, about 258.8 min of steps left; card peak 13.7 GB +[ 376.7s] stage 1/8 s1_perspective step 600/4000: train bpb s1 0.1260; quiet KL s1 0.00067; 0.49 s/step; the job: 6.3 min spent, about 257.4 min of steps left; card peak 13.7 GB +[ 400.1s] stage 1/8 s1_perspective step 650/4000: train bpb s1 0.1122; quiet KL s1 0.00066; 0.49 s/step; the job: 6.7 min spent, about 256.1 min of steps left; card peak 13.7 GB +[ 423.6s] stage 1/8 s1_perspective step 700/4000: train bpb s1 0.1026; quiet KL s1 0.00067; 0.49 s/step; the job: 7.1 min spent, about 254.9 min of steps left; card peak 13.7 GB +[ 447.0s] stage 1/8 s1_perspective step 750/4000: train bpb s1 0.0936; quiet KL s1 0.00069; 0.49 s/step; the job: 7.5 min spent, about 253.8 min of steps left; card peak 13.7 GB +[ 470.6s] stage 1/8 s1_perspective step 800/4000: train bpb s1 0.0866; quiet KL s1 0.00063; 0.49 s/step; the job: 7.8 min spent, about 252.9 min of steps left; card peak 13.7 GB +[ 494.0s] stage 1/8 s1_perspective step 850/4000: train bpb s1 0.0799; quiet KL s1 0.00059; 0.49 s/step; the job: 8.2 min spent, about 251.9 min of steps left; card peak 13.7 GB +[ 517.4s] stage 1/8 s1_perspective step 900/4000: train bpb s1 0.0753; quiet KL s1 0.00056; 0.48 s/step; the job: 8.6 min spent, about 251.0 min of steps left; card peak 13.7 GB +[ 541.6s] stage 1/8 s1_perspective step 950/4000: train bpb s1 0.0702; quiet KL s1 0.00056; 0.48 s/step; the job: 9.0 min spent, about 250.6 min of steps left; card peak 13.7 GB +[ 566.0s] stage 1/8 s1_perspective step 1000/4000: train bpb s1 0.0675; quiet KL s1 0.00059; 0.48 s/step; the job: 9.4 min spent, about 250.3 min of steps left; card peak 13.7 GB +[ 590.1s] stage 1/8 s1_perspective step 1050/4000: train bpb s1 0.0649; quiet KL s1 0.00058; 0.48 s/step; the job: 9.8 min spent, about 249.8 min of steps left; card peak 13.7 GB +[ 614.0s] stage 1/8 s1_perspective step 1100/4000: train bpb s1 0.0622; quiet KL s1 0.00054; 0.48 s/step; the job: 10.2 min spent, about 249.3 min of steps left; card peak 13.7 GB +[ 638.2s] stage 1/8 s1_perspective step 1150/4000: train bpb s1 0.0603; quiet KL s1 0.00056; 0.48 s/step; the job: 10.6 min spent, about 248.9 min of steps left; card peak 13.7 GB +[ 662.0s] stage 1/8 s1_perspective step 1200/4000: train bpb s1 0.0595; quiet KL s1 0.00058; 0.48 s/step; the job: 11.0 min spent, about 248.3 min of steps left; card peak 13.7 GB +[ 686.0s] stage 1/8 s1_perspective step 1250/4000: train bpb s1 0.0582; quiet KL s1 0.00054; 0.48 s/step; the job: 11.4 min spent, about 247.8 min of steps left; card peak 13.7 GB +[ 709.9s] stage 1/8 s1_perspective step 1300/4000: train bpb s1 0.0573; quiet KL s1 0.00053; 0.48 s/step; the job: 11.8 min spent, about 247.3 min of steps left; card peak 13.7 GB +[ 733.6s] stage 1/8 s1_perspective step 1350/4000: train bpb s1 0.0564; quiet KL s1 0.00051; 0.48 s/step; the job: 12.2 min spent, about 246.7 min of steps left; card peak 13.7 GB diff --git a/mini-beatrix-3/arm_refit/logs/group_EB.log b/mini-beatrix-3/arm_refit/logs/group_EB.log new file mode 100644 index 0000000000000000000000000000000000000000..c62cc1cf328c4ea481dfba520ca42de8f94d8c3a --- /dev/null +++ b/mini-beatrix-3/arm_refit/logs/group_EB.log @@ -0,0 +1,31 @@ +[ 0.2s] THE JOB: gEB: the group ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail', 's7_mixed', 's8_register'], staged, EARLIER ARMS FROZEN (on, not training); 4000 steps per stage; each step = one task chunk + one abstention chunk of 2 x 4096 bytes per newest member; lambda 2.0; pure Adam lr 0.001; the compiled chain with recompute; trunk step 245,674; card 31.4 GB with 16.4 GB free; host RSS 0.6 GB; a member's partner rows come from every other member of the roster, attached or not +Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads. +[ 14.7s] mini-beatrix-3 @ 245,674: trunk loaded from /root/v3/runs/mission/mini-beatrix-3/checkpoints/step_00245674.safetensors and frozen (376.0M); amoe-lora 0.2.11, torch 2.8.0+cu128 +[stream] opening 'fineweb-good' (HuggingFaceFW/fineweb-edu) … +[stream] opening 'simple-wiki' (wikimedia/wikipedia) … +[stream] opening 'babi-prose' (Muennighoff/babi) … +Repo card metadata block was not found. Setting CardData to empty. +[stream] opening 'tinystories' (roneneldan/TinyStories) … +[stream] opening 'proofwriter-prose' (tasksource/proofwriter) … +[ 85.1s] ARMS OFF (the bare trunk): s1_perspective rows 0.5538; s2_concept rows 0.7212; s3_rules rows 0.9296; s4_arith rows 1.2339; s5_causal rows 0.6604; s6_tryfail rows 0.6728; s7_mixed rows 0.7830; s8_register rows 0.4907; web rows 0.9506; s1_perspective battery in 0.550, transfer 0.463; s2_concept battery in 0.298, transfer 0.678; s3_rules battery in 0.155, transfer 0.070, held-out 0.050; s4_arith battery in 0.344, transfer 0.242, held-out 0.350; probes mean 0.411 +[arms] attached 's1_perspective' for phase 'curriculum_s1': 13.69M params x 32 blocks, lambda 2.0, spec {'n_slots': 16, 'K': 16, 'D': 8, 'hidden': 256}, recompute on, compile on +[ 88.9s] STAGE 1/8: s1_perspective attached over the bare trunk; at birth the group reads exactly as before on its rows and on web text; 4000 steps of 1 task + 1 abstention chunks +[stream] opening 'fineweb-edu' (HuggingFaceFW/fineweb-edu) … +[ 94.1s] stage 1/8 s1_perspective step 1/4000: train bpb s1 0.5753; quiet KL s1 0.00000; 5.27 s/step; the job: 1.6 min spent, about 2809.6 min of steps left; card peak 13.6 GB +[ 118.0s] stage 1/8 s1_perspective step 50/4000: train bpb s1 0.5493; quiet KL s1 0.00059; 0.58 s/step; the job: 2.0 min spent, about 310.1 min of steps left; card peak 13.7 GB +[ 141.7s] stage 1/8 s1_perspective step 100/4000: train bpb s1 0.5042; quiet KL s1 0.00148; 0.53 s/step; the job: 2.4 min spent, about 280.7 min of steps left; card peak 13.7 GB +[ 165.1s] stage 1/8 s1_perspective step 150/4000: train bpb s1 0.4349; quiet KL s1 0.00145; 0.51 s/step; the job: 2.8 min spent, about 269.9 min of steps left; card peak 13.7 GB +[ 188.8s] stage 1/8 s1_perspective step 200/4000: train bpb s1 0.3551; quiet KL s1 0.00166; 0.50 s/step; the job: 3.1 min spent, about 264.7 min of steps left; card peak 13.7 GB +[ 212.2s] stage 1/8 s1_perspective step 250/4000: train bpb s1 0.2999; quiet KL s1 0.00217; 0.49 s/step; the job: 3.5 min spent, about 261.1 min of steps left; card peak 13.7 GB +[ 235.7s] stage 1/8 s1_perspective step 300/4000: train bpb s1 0.2626; quiet KL s1 0.00185; 0.49 s/step; the job: 3.9 min spent, about 258.6 min of steps left; card peak 13.7 GB +[ 259.3s] stage 1/8 s1_perspective step 350/4000: train bpb s1 0.2305; quiet KL s1 0.00137; 0.49 s/step; the job: 4.3 min spent, about 256.9 min of steps left; card peak 13.7 GB +[ 282.9s] stage 1/8 s1_perspective step 400/4000: train bpb s1 0.2013; quiet KL s1 0.00116; 0.49 s/step; the job: 4.7 min spent, about 255.5 min of steps left; card peak 13.7 GB +[ 307.2s] stage 1/8 s1_perspective step 450/4000: train bpb s1 0.1764; quiet KL s1 0.00099; 0.49 s/step; the job: 5.1 min spent, about 255.1 min of steps left; card peak 13.7 GB +[ 330.7s] stage 1/8 s1_perspective step 500/4000: train bpb s1 0.1572; quiet KL s1 0.00079; 0.48 s/step; the job: 5.5 min spent, about 253.9 min of steps left; card peak 13.7 GB +[ 354.6s] stage 1/8 s1_perspective step 550/4000: train bpb s1 0.1388; quiet KL s1 0.00074; 0.48 s/step; the job: 5.9 min spent, about 253.2 min of steps left; card peak 13.7 GB +[ 379.0s] stage 1/8 s1_perspective step 600/4000: train bpb s1 0.1259; quiet KL s1 0.00067; 0.48 s/step; the job: 6.3 min spent, about 253.0 min of steps left; card peak 13.7 GB +[ 403.4s] stage 1/8 s1_perspective step 650/4000: train bpb s1 0.1131; quiet KL s1 0.00056; 0.48 s/step; the job: 6.7 min spent, about 252.8 min of steps left; card peak 13.7 GB +[ 427.3s] stage 1/8 s1_perspective step 700/4000: train bpb s1 0.1024; quiet KL s1 0.00064; 0.48 s/step; the job: 7.1 min spent, about 252.2 min of steps left; card peak 13.7 GB +[ 451.6s] stage 1/8 s1_perspective step 750/4000: train bpb s1 0.0944; quiet KL s1 0.00062; 0.48 s/step; the job: 7.5 min spent, about 251.9 min of steps left; card peak 13.7 GB +[ 476.0s] stage 1/8 s1_perspective step 800/4000: train bpb s1 0.0863; quiet KL s1 0.00059; 0.48 s/step; the job: 7.9 min spent, about 251.7 min of steps left; card peak 13.7 GB +[ 500.3s] stage 1/8 s1_perspective step 850/4000: train bpb s1 0.0805; quiet KL s1 0.00057; 0.48 s/step; the job: 8.3 min spent, about 251.3 min of steps left; card peak 13.7 GB diff --git a/mini-beatrix-3/arm_refit/logs/group_XA.log b/mini-beatrix-3/arm_refit/logs/group_XA.log new file mode 100644 index 0000000000000000000000000000000000000000..e1ec4535514c1385dd57b8d79180921d6fdb1733 --- /dev/null +++ b/mini-beatrix-3/arm_refit/logs/group_XA.log @@ -0,0 +1,454 @@ +[ 0.2s] THE JOB: gXA: the group ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail', 's7_mixed', 's8_register'], staged, EARLIER ARMS FROZEN (on, not training); 4000 steps per stage; each step = one task chunk + one abstention chunk of 2 x 4096 bytes per newest member; lambda 2.0; pure Adam lr 0.001; the compiled chain with recompute; trunk step 245,674; card 31.4 GB with 16.4 GB free; host RSS 0.6 GB; the first 4 members loaded from gRA:4 (the later members attach fresh); a member's partner rows come from every other member of the roster, attached or not +Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads. +[ 15.7s] mini-beatrix-3 @ 245,674: trunk loaded from /root/v3/runs/mission/mini-beatrix-3/checkpoints/step_00245674.safetensors and frozen (376.0M); amoe-lora 0.2.11, torch 2.8.0+cu128 +[stream] opening 'fineweb-good' (HuggingFaceFW/fineweb-edu) … +[stream] opening 'simple-wiki' (wikimedia/wikipedia) … +[stream] opening 'babi-prose' (Muennighoff/babi) … +Repo card metadata block was not found. Setting CardData to empty. +[stream] opening 'tinystories' (roneneldan/TinyStories) … +[stream] opening 'proofwriter-prose' (tasksource/proofwriter) … +[ 82.1s] ARMS OFF (the bare trunk): s1_perspective rows 0.5538; s2_concept rows 0.7212; s3_rules rows 0.9296; s4_arith rows 1.2339; s5_causal rows 0.6604; s6_tryfail rows 0.6728; s7_mixed rows 0.7830; s8_register rows 0.4907; web rows 0.9506; s1_perspective battery in 0.550, transfer 0.463; s2_concept battery in 0.298, transfer 0.678; s3_rules battery in 0.155, transfer 0.070, held-out 0.050; s4_arith battery in 0.344, transfer 0.242, held-out 0.350; probes mean 0.411 +[arms] attached 's1_perspective' for phase 'curriculum_s1': 13.69M params x 32 blocks, lambda 2.0, spec {'n_slots': 16, 'K': 16, 'D': 8, 'hidden': 256}, recompute on, compile on +[arms] attached 's2_concept' for phase 'curriculum_s2': 13.69M params x 32 blocks, lambda 2.0, spec {'n_slots': 16, 'K': 16, 'D': 8, 'hidden': 256}, recompute on, compile on +[arms] attached 's3_rules' for phase 'curriculum_s3': 13.69M params x 32 blocks, lambda 2.0, spec {'n_slots': 16, 'K': 16, 'D': 8, 'hidden': 256}, recompute on, compile on +[arms] attached 's4_arith' for phase 'curriculum_s4': 13.69M params x 32 blocks, lambda 2.0, spec {'n_slots': 16, 'K': 16, 'D': 8, 'hidden': 256}, recompute on, compile on +[ 86.8s] RESUMED from gRA at its close 4: ['s1_perspective', 's2_concept', 's3_rules', 's4_arith'] loaded (on; frozen while the later arms train); the next stage attaches over them +[ 223.4s] CLOSE 4/8: the group ['s1_perspective', 's2_concept', 's3_rules', 's4_arith'] all on, against arms off: +[ 223.4s] s1_perspective rows: off 0.5538 -> all on 0.0482 (-0.5057 +- 0.0048); with s1_perspective masked 0.5249 (its share -0.4767); s1_perspective alone 0.0525 (-0.5013); gate mean 0.210 / widest 0.571 +[ 223.4s] s2_concept rows: off 0.7212 -> all on 0.1123 (-0.6089 +- 0.0038); with s2_concept masked 0.2405 (its share -0.1283); s2_concept alone 0.3503 (-0.3709); gate mean 0.277 / widest 0.498 +[ 223.4s] s3_rules rows: off 0.9296 -> all on 0.2236 (-0.7060 +- 0.0052); with s3_rules masked 0.3189 (its share -0.0952); s3_rules alone 0.8236 (-0.1060); gate mean 0.120 / widest 0.174 +[ 223.4s] s4_arith rows: off 1.2339 -> all on 0.3384 (-0.8955 +- 0.0059); with s4_arith masked 0.4861 (its share -0.1477); s4_arith alone 1.1267 (-0.1072); gate mean 0.108 / widest 0.159 +[ 223.4s] web holdout: off 0.9506 -> all on 0.9523 (+0.00171 +- 0.00035; the quiet bar +.012); each member's share: s1_perspective +0.00035, s2_concept -0.00010, s3_rules +0.00040, s4_arith +0.00010; alone: s1_perspective +0.00143, s2_concept +0.00105, s3_rules +0.00053, s4_arith +0.00011 +[ 223.4s] a member's share on a PARTNER's rows (all on minus that member masked; near 0 = it stays out): s1_perspective on s2_concept -0.2405; s1_perspective on s3_rules -0.1848; s1_perspective on s4_arith -0.1575; s2_concept on s1_perspective -0.0043; s2_concept on s3_rules -0.1064; s2_concept on s4_arith -0.1204; s3_rules on s1_perspective +0.0001; s3_rules on s2_concept +0.0007; s3_rules on s4_arith +0.0000; s4_arith on s1_perspective -0.0001; s4_arith on s2_concept -0.0006; s4_arith on s3_rules +0.0000 +[ 223.4s] a member ALONE on a partner's rows (alone minus off): s1_perspective on s2_concept -0.4817; s1_perspective on s3_rules -0.4209; s1_perspective on s4_arith -0.5113; s2_concept on s1_perspective -0.0273; s2_concept on s3_rules -0.3108; s2_concept on s4_arith -0.3537; s3_rules on s1_perspective -0.0003; s3_rules on s2_concept +0.0031; s3_rules on s4_arith +0.0001; s4_arith on s1_perspective -0.0009; s4_arith on s2_concept -0.0004; s4_arith on s3_rules -0.0016 +[ 223.4s] the group on the rows of stages not yet attached: s5_causal -0.1818; s6_tryfail -0.1849; s7_mixed -0.0082; s8_register -0.1085 +[ 223.4s] s1_perspective battery accuracy, off -> all on: in 0.550 -> 0.963, transfer 0.463 -> 0.475; alone: in 0.838, transfer 0.463 +[ 223.4s] s2_concept battery accuracy, off -> all on: in 0.298 -> 0.976, transfer 0.678 -> 0.780; alone: in 0.964, transfer 0.720 +[ 223.4s] s3_rules battery accuracy, off -> all on: in 0.155 -> 0.905, transfer 0.070 -> 0.300, held-out 0.050 -> 0.458; alone: in 0.320, transfer 0.080, held-out 0.108 +[ 223.4s] s4_arith battery accuracy, off -> all on: in 0.344 -> 0.613, transfer 0.242 -> 0.237, held-out 0.350 -> 0.350; alone: in 0.381, transfer 0.246, held-out 0.388 +[ 223.4s] probes mean: off 0.411 -> all on 0.419 +[ 223.4s] compiled chain: 8 cache entries (limit 256), 0 limit warnings; card peak 3.0 GB; host RSS 5.3 GB +[arms] attached 's5_causal' for phase 'curriculum_s5': 13.69M params x 32 blocks, lambda 2.0, spec {'n_slots': 16, 'K': 16, 'D': 8, 'hidden': 256}, recompute on, compile on +[ 226.9s] STAGE 5/8: s5_causal attached over ['s1_perspective', 's2_concept', 's3_rules', 's4_arith'] (on, frozen); at birth the group reads exactly as before on its rows and on web text; 4000 steps of 1 task + 1 abstention chunks +[stream] opening 'fineweb-edu' (HuggingFaceFW/fineweb-edu) … +[ 237.0s] stage 5/8 s5_causal step 1/4000: train bpb s5 0.4826; quiet KL s5 0.00013; 10.03 s/step; the job: 3.9 min spent, about 2674.3 min of steps left; card peak 13.9 GB +[ 269.8s] stage 5/8 s5_causal step 50/4000: train bpb s5 0.4627; quiet KL s5 0.00023; 0.86 s/step; the job: 4.5 min spent, about 228.1 min of steps left; card peak 14.0 GB +[ 301.3s] stage 5/8 s5_causal step 100/4000: train bpb s5 0.4353; quiet KL s5 0.00058; 0.74 s/step; the job: 5.0 min spent, about 197.2 min of steps left; card peak 14.0 GB +[ 332.7s] stage 5/8 s5_causal step 150/4000: train bpb s5 0.4018; quiet KL s5 0.00065; 0.71 s/step; the job: 5.5 min spent, about 186.4 min of steps left; card peak 14.0 GB +[ 364.3s] stage 5/8 s5_causal step 200/4000: train bpb s5 0.3648; quiet KL s5 0.00068; 0.69 s/step; the job: 6.1 min spent, about 180.9 min of steps left; card peak 14.0 GB +[ 395.7s] stage 5/8 s5_causal step 250/4000: train bpb s5 0.3277; quiet KL s5 0.00077; 0.68 s/step; the job: 6.6 min spent, about 177.2 min of steps left; card peak 14.0 GB +[ 427.1s] stage 5/8 s5_causal step 300/4000: train bpb s5 0.2895; quiet KL s5 0.00074; 0.67 s/step; the job: 7.1 min spent, about 174.6 min of steps left; card peak 14.0 GB +[ 458.6s] stage 5/8 s5_causal step 350/4000: train bpb s5 0.2472; quiet KL s5 0.00088; 0.66 s/step; the job: 7.6 min spent, about 172.7 min of steps left; card peak 14.0 GB +[ 490.0s] stage 5/8 s5_causal step 400/4000: train bpb s5 0.2082; quiet KL s5 0.00084; 0.66 s/step; the job: 8.2 min spent, about 171.0 min of steps left; card peak 14.0 GB +[ 521.3s] stage 5/8 s5_causal step 450/4000: train bpb s5 0.1748; quiet KL s5 0.00085; 0.65 s/step; the job: 8.7 min spent, about 169.5 min of steps left; card peak 14.0 GB +[ 552.8s] stage 5/8 s5_causal step 500/4000: train bpb s5 0.1429; quiet KL s5 0.00086; 0.65 s/step; the job: 9.2 min spent, about 168.4 min of steps left; card peak 14.0 GB +[ 584.2s] stage 5/8 s5_causal step 550/4000: train bpb s5 0.1156; quiet KL s5 0.00079; 0.65 s/step; the job: 9.7 min spent, about 167.3 min of steps left; card peak 14.0 GB +[ 615.6s] stage 5/8 s5_causal step 600/4000: train bpb s5 0.0952; quiet KL s5 0.00074; 0.65 s/step; the job: 10.3 min spent, about 166.2 min of steps left; card peak 14.0 GB +[ 647.1s] stage 5/8 s5_causal step 650/4000: train bpb s5 0.0807; quiet KL s5 0.00072; 0.65 s/step; the job: 10.8 min spent, about 165.4 min of steps left; card peak 14.0 GB +[ 678.5s] stage 5/8 s5_causal step 700/4000: train bpb s5 0.0719; quiet KL s5 0.00073; 0.65 s/step; the job: 11.3 min spent, about 164.5 min of steps left; card peak 14.0 GB +[ 709.9s] stage 5/8 s5_causal step 750/4000: train bpb s5 0.0631; quiet KL s5 0.00070; 0.64 s/step; the job: 11.8 min spent, about 163.7 min of steps left; card peak 14.0 GB +[ 741.5s] stage 5/8 s5_causal step 800/4000: train bpb s5 0.0586; quiet KL s5 0.00065; 0.64 s/step; the job: 12.4 min spent, about 163.0 min of steps left; card peak 14.0 GB +[ 772.9s] stage 5/8 s5_causal step 850/4000: train bpb s5 0.0560; quiet KL s5 0.00061; 0.64 s/step; the job: 12.9 min spent, about 162.2 min of steps left; card peak 14.0 GB +[ 804.3s] stage 5/8 s5_causal step 900/4000: train bpb s5 0.0534; quiet KL s5 0.00058; 0.64 s/step; the job: 13.4 min spent, about 161.4 min of steps left; card peak 14.0 GB +[ 835.7s] stage 5/8 s5_causal step 950/4000: train bpb s5 0.0519; quiet KL s5 0.00055; 0.64 s/step; the job: 13.9 min spent, about 160.7 min of steps left; card peak 14.0 GB +[ 867.3s] stage 5/8 s5_causal step 1000/4000: train bpb s5 0.0505; quiet KL s5 0.00052; 0.64 s/step; the job: 14.5 min spent, about 160.1 min of steps left; card peak 14.0 GB +[ 898.6s] stage 5/8 s5_causal step 1050/4000: train bpb s5 0.0496; quiet KL s5 0.00049; 0.64 s/step; the job: 15.0 min spent, about 159.4 min of steps left; card peak 14.0 GB +[ 930.1s] stage 5/8 s5_causal step 1100/4000: train bpb s5 0.0497; quiet KL s5 0.00047; 0.64 s/step; the job: 15.5 min spent, about 158.7 min of steps left; card peak 14.0 GB +[ 961.6s] stage 5/8 s5_causal step 1150/4000: train bpb s5 0.0495; quiet KL s5 0.00049; 0.64 s/step; the job: 16.0 min spent, about 158.1 min of steps left; card peak 14.0 GB +[ 993.0s] stage 5/8 s5_causal step 1200/4000: train bpb s5 0.0479; quiet KL s5 0.00048; 0.64 s/step; the job: 16.6 min spent, about 157.5 min of steps left; card peak 14.0 GB +[ 1024.6s] stage 5/8 s5_causal step 1250/4000: train bpb s5 0.0476; quiet KL s5 0.00045; 0.64 s/step; the job: 17.1 min spent, about 156.9 min of steps left; card peak 14.0 GB +[ 1056.0s] stage 5/8 s5_causal step 1300/4000: train bpb s5 0.0471; quiet KL s5 0.00041; 0.64 s/step; the job: 17.6 min spent, about 156.2 min of steps left; card peak 14.0 GB +[ 1087.4s] stage 5/8 s5_causal step 1350/4000: train bpb s5 0.0469; quiet KL s5 0.00044; 0.64 s/step; the job: 18.1 min spent, about 155.6 min of steps left; card peak 14.0 GB +[ 1118.8s] stage 5/8 s5_causal step 1400/4000: train bpb s5 0.0470; quiet KL s5 0.00042; 0.64 s/step; the job: 18.6 min spent, about 155.0 min of steps left; card peak 14.0 GB +[ 1150.4s] stage 5/8 s5_causal step 1450/4000: train bpb s5 0.0460; quiet KL s5 0.00039; 0.64 s/step; the job: 19.2 min spent, about 154.4 min of steps left; card peak 14.0 GB +[ 1182.2s] stage 5/8 s5_causal step 1500/4000: train bpb s5 0.0460; quiet KL s5 0.00042; 0.64 s/step; the job: 19.7 min spent, about 153.9 min of steps left; card peak 14.0 GB +[ 1214.0s] stage 5/8 s5_causal step 1550/4000: train bpb s5 0.0461; quiet KL s5 0.00039; 0.64 s/step; the job: 20.2 min spent, about 153.4 min of steps left; card peak 14.0 GB +[ 1245.8s] stage 5/8 s5_causal step 1600/4000: train bpb s5 0.0457; quiet KL s5 0.00038; 0.64 s/step; the job: 20.8 min spent, about 152.8 min of steps left; card peak 14.0 GB +[ 1277.4s] stage 5/8 s5_causal step 1650/4000: train bpb s5 0.0452; quiet KL s5 0.00039; 0.64 s/step; the job: 21.3 min spent, about 152.3 min of steps left; card peak 14.0 GB +[ 1308.7s] stage 5/8 s5_causal step 1700/4000: train bpb s5 0.0453; quiet KL s5 0.00037; 0.64 s/step; the job: 21.8 min spent, about 151.7 min of steps left; card peak 14.0 GB +[ 1340.3s] stage 5/8 s5_causal step 1750/4000: train bpb s5 0.0451; quiet KL s5 0.00038; 0.64 s/step; the job: 22.3 min spent, about 151.1 min of steps left; card peak 14.0 GB +[ 1371.7s] stage 5/8 s5_causal step 1800/4000: train bpb s5 0.0449; quiet KL s5 0.00035; 0.64 s/step; the job: 22.9 min spent, about 150.5 min of steps left; card peak 14.0 GB +[ 1403.2s] stage 5/8 s5_causal step 1850/4000: train bpb s5 0.0450; quiet KL s5 0.00041; 0.64 s/step; the job: 23.4 min spent, about 149.9 min of steps left; card peak 14.0 GB +[ 1434.8s] stage 5/8 s5_causal step 1900/4000: train bpb s5 0.0450; quiet KL s5 0.00038; 0.64 s/step; the job: 23.9 min spent, about 149.4 min of steps left; card peak 14.0 GB +[ 1466.3s] stage 5/8 s5_causal step 1950/4000: train bpb s5 0.0447; quiet KL s5 0.00038; 0.64 s/step; the job: 24.4 min spent, about 148.8 min of steps left; card peak 14.0 GB +[ 1497.7s] stage 5/8 s5_causal step 2000/4000: train bpb s5 0.0446; quiet KL s5 0.00036; 0.64 s/step; the job: 25.0 min spent, about 148.3 min of steps left; card peak 14.0 GB +[ 1529.2s] stage 5/8 s5_causal step 2050/4000: train bpb s5 0.0447; quiet KL s5 0.00034; 0.64 s/step; the job: 25.5 min spent, about 147.7 min of steps left; card peak 14.0 GB +[ 1560.7s] stage 5/8 s5_causal step 2100/4000: train bpb s5 0.0441; quiet KL s5 0.00033; 0.64 s/step; the job: 26.0 min spent, about 147.1 min of steps left; card peak 14.0 GB +[ 1592.3s] stage 5/8 s5_causal step 2150/4000: train bpb s5 0.0446; quiet KL s5 0.00035; 0.64 s/step; the job: 26.5 min spent, about 146.6 min of steps left; card peak 14.0 GB +[ 1624.2s] stage 5/8 s5_causal step 2200/4000: train bpb s5 0.0440; quiet KL s5 0.00033; 0.64 s/step; the job: 27.1 min spent, about 146.1 min of steps left; card peak 14.0 GB +[ 1655.7s] stage 5/8 s5_causal step 2250/4000: train bpb s5 0.0440; quiet KL s5 0.00032; 0.64 s/step; the job: 27.6 min spent, about 145.5 min of steps left; card peak 14.0 GB +[ 1687.3s] stage 5/8 s5_causal step 2300/4000: train bpb s5 0.0439; quiet KL s5 0.00031; 0.63 s/step; the job: 28.1 min spent, about 145.0 min of steps left; card peak 14.0 GB +[ 1718.9s] stage 5/8 s5_causal step 2350/4000: train bpb s5 0.0438; quiet KL s5 0.00034; 0.63 s/step; the job: 28.6 min spent, about 144.4 min of steps left; card peak 14.0 GB +[ 1750.3s] stage 5/8 s5_causal step 2400/4000: train bpb s5 0.0440; quiet KL s5 0.00035; 0.63 s/step; the job: 29.2 min spent, about 143.9 min of steps left; card peak 14.0 GB +[ 1781.8s] stage 5/8 s5_causal step 2450/4000: train bpb s5 0.0439; quiet KL s5 0.00035; 0.63 s/step; the job: 29.7 min spent, about 143.3 min of steps left; card peak 14.0 GB +[ 1813.4s] stage 5/8 s5_causal step 2500/4000: train bpb s5 0.0440; quiet KL s5 0.00035; 0.63 s/step; the job: 30.2 min spent, about 142.8 min of steps left; card peak 14.0 GB +[ 1845.1s] stage 5/8 s5_causal step 2550/4000: train bpb s5 0.0438; quiet KL s5 0.00034; 0.63 s/step; the job: 30.8 min spent, about 142.2 min of steps left; card peak 14.0 GB +[ 1876.6s] stage 5/8 s5_causal step 2600/4000: train bpb s5 0.0443; quiet KL s5 0.00037; 0.63 s/step; the job: 31.3 min spent, about 141.7 min of steps left; card peak 14.0 GB +[ 1908.3s] stage 5/8 s5_causal step 2650/4000: train bpb s5 0.0440; quiet KL s5 0.00032; 0.63 s/step; the job: 31.8 min spent, about 141.2 min of steps left; card peak 14.0 GB +[ 1939.8s] stage 5/8 s5_causal step 2700/4000: train bpb s5 0.0437; quiet KL s5 0.00033; 0.63 s/step; the job: 32.3 min spent, about 140.6 min of steps left; card peak 14.0 GB +[ 1971.3s] stage 5/8 s5_causal step 2750/4000: train bpb s5 0.0437; quiet KL s5 0.00032; 0.63 s/step; the job: 32.9 min spent, about 140.1 min of steps left; card peak 14.0 GB +[ 2002.9s] stage 5/8 s5_causal step 2800/4000: train bpb s5 0.0440; quiet KL s5 0.00035; 0.63 s/step; the job: 33.4 min spent, about 139.5 min of steps left; card peak 14.0 GB +[ 2034.4s] stage 5/8 s5_causal step 2850/4000: train bpb s5 0.0433; quiet KL s5 0.00031; 0.63 s/step; the job: 33.9 min spent, about 139.0 min of steps left; card peak 14.0 GB +[ 2065.8s] stage 5/8 s5_causal step 2900/4000: train bpb s5 0.0433; quiet KL s5 0.00030; 0.63 s/step; the job: 34.4 min spent, about 138.4 min of steps left; card peak 14.0 GB +[ 2097.5s] stage 5/8 s5_causal step 2950/4000: train bpb s5 0.0432; quiet KL s5 0.00030; 0.63 s/step; the job: 35.0 min spent, about 137.9 min of steps left; card peak 14.0 GB +[ 2129.0s] stage 5/8 s5_causal step 3000/4000: train bpb s5 0.0432; quiet KL s5 0.00031; 0.63 s/step; the job: 35.5 min spent, about 137.4 min of steps left; card peak 14.0 GB +[ 2160.5s] stage 5/8 s5_causal step 3050/4000: train bpb s5 0.0433; quiet KL s5 0.00031; 0.63 s/step; the job: 36.0 min spent, about 136.8 min of steps left; card peak 14.0 GB +[ 2192.2s] stage 5/8 s5_causal step 3100/4000: train bpb s5 0.0435; quiet KL s5 0.00031; 0.63 s/step; the job: 36.5 min spent, about 136.3 min of steps left; card peak 14.0 GB +[ 2223.7s] stage 5/8 s5_causal step 3150/4000: train bpb s5 0.0432; quiet KL s5 0.00031; 0.63 s/step; the job: 37.1 min spent, about 135.8 min of steps left; card peak 14.0 GB +[ 2255.2s] stage 5/8 s5_causal step 3200/4000: train bpb s5 0.0435; quiet KL s5 0.00029; 0.63 s/step; the job: 37.6 min spent, about 135.2 min of steps left; card peak 14.0 GB +[ 2287.0s] stage 5/8 s5_causal step 3250/4000: train bpb s5 0.0433; quiet KL s5 0.00032; 0.63 s/step; the job: 38.1 min spent, about 134.7 min of steps left; card peak 14.0 GB +[ 2318.5s] stage 5/8 s5_causal step 3300/4000: train bpb s5 0.0434; quiet KL s5 0.00034; 0.63 s/step; the job: 38.6 min spent, about 134.2 min of steps left; card peak 14.0 GB +[ 2350.0s] stage 5/8 s5_causal step 3350/4000: train bpb s5 0.0431; quiet KL s5 0.00031; 0.63 s/step; the job: 39.2 min spent, about 133.6 min of steps left; card peak 14.0 GB +[ 2381.6s] stage 5/8 s5_causal step 3400/4000: train bpb s5 0.0431; quiet KL s5 0.00032; 0.63 s/step; the job: 39.7 min spent, about 133.1 min of steps left; card peak 14.0 GB +[ 2413.1s] stage 5/8 s5_causal step 3450/4000: train bpb s5 0.0430; quiet KL s5 0.00037; 0.63 s/step; the job: 40.2 min spent, about 132.5 min of steps left; card peak 14.0 GB +[ 2444.6s] stage 5/8 s5_causal step 3500/4000: train bpb s5 0.0429; quiet KL s5 0.00030; 0.63 s/step; the job: 40.7 min spent, about 132.0 min of steps left; card peak 14.0 GB +[ 2476.4s] stage 5/8 s5_causal step 3550/4000: train bpb s5 0.0431; quiet KL s5 0.00033; 0.63 s/step; the job: 41.3 min spent, about 131.5 min of steps left; card peak 14.0 GB +[ 2507.9s] stage 5/8 s5_causal step 3600/4000: train bpb s5 0.0430; quiet KL s5 0.00034; 0.63 s/step; the job: 41.8 min spent, about 130.9 min of steps left; card peak 14.0 GB +[ 2539.4s] stage 5/8 s5_causal step 3650/4000: train bpb s5 0.0429; quiet KL s5 0.00029; 0.63 s/step; the job: 42.3 min spent, about 130.4 min of steps left; card peak 14.0 GB +[ 2571.2s] stage 5/8 s5_causal step 3700/4000: train bpb s5 0.0427; quiet KL s5 0.00031; 0.63 s/step; the job: 42.9 min spent, about 129.9 min of steps left; card peak 14.0 GB +[ 2602.9s] stage 5/8 s5_causal step 3750/4000: train bpb s5 0.0426; quiet KL s5 0.00031; 0.63 s/step; the job: 43.4 min spent, about 129.4 min of steps left; card peak 14.0 GB +[ 2634.6s] stage 5/8 s5_causal step 3800/4000: train bpb s5 0.0427; quiet KL s5 0.00030; 0.63 s/step; the job: 43.9 min spent, about 128.8 min of steps left; card peak 14.0 GB +[ 2666.5s] stage 5/8 s5_causal step 3850/4000: train bpb s5 0.0427; quiet KL s5 0.00031; 0.63 s/step; the job: 44.4 min spent, about 128.3 min of steps left; card peak 14.0 GB +[ 2698.3s] stage 5/8 s5_causal step 3900/4000: train bpb s5 0.0426; quiet KL s5 0.00030; 0.63 s/step; the job: 45.0 min spent, about 127.8 min of steps left; card peak 14.0 GB +[ 2729.9s] stage 5/8 s5_causal step 3950/4000: train bpb s5 0.0430; quiet KL s5 0.00041; 0.63 s/step; the job: 45.5 min spent, about 127.3 min of steps left; card peak 14.0 GB +[ 2761.7s] stage 5/8 s5_causal step 4000/4000: train bpb s5 0.0431; quiet KL s5 0.00035; 0.63 s/step; the job: 46.0 min spent, about 126.7 min of steps left; card peak 14.0 GB +[ 2762.5s] anchors saved: /root/refit/group/anchors/gXA_close5_.safetensors for ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal'] +[ 2917.3s] CLOSE 5/8: the group ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal'] all on, against arms off: +[ 2917.3s] s1_perspective rows: off 0.5538 -> all on 0.0489 (-0.5050 +- 0.0048); with s1_perspective masked 0.5074 (its share -0.4586); s1_perspective alone 0.0525 (-0.5013); gate mean 0.210 / widest 0.571 +[ 2917.3s] s2_concept rows: off 0.7212 -> all on 0.1116 (-0.6095 +- 0.0038); with s2_concept masked 0.2356 (its share -0.1239); s2_concept alone 0.3503 (-0.3709); gate mean 0.277 / widest 0.498 +[ 2917.3s] s3_rules rows: off 0.9296 -> all on 0.2241 (-0.7055 +- 0.0053); with s3_rules masked 0.3162 (its share -0.0920); s3_rules alone 0.8236 (-0.1060); gate mean 0.120 / widest 0.174 +[ 2917.3s] s4_arith rows: off 1.2339 -> all on 0.3385 (-0.8954 +- 0.0059); with s4_arith masked 0.4841 (its share -0.1456); s4_arith alone 1.1267 (-0.1072); gate mean 0.108 / widest 0.159 +[ 2917.3s] s5_causal rows: off 0.6604 -> all on 0.0432 (-0.6173 +- 0.0070); with s5_causal masked 0.4786 (its share -0.4354); s5_causal alone 0.0911 (-0.5693); gate mean 0.185 / widest 0.368 +[ 2917.3s] web holdout: off 0.9506 -> all on 0.9527 (+0.00211 +- 0.00042; the quiet bar +.012); each member's share: s1_perspective +0.00053, s2_concept -0.00026, s3_rules +0.00049, s4_arith +0.00022, s5_causal +0.00040; alone: s1_perspective +0.00143, s2_concept +0.00105, s3_rules +0.00053, s4_arith +0.00011, s5_causal +0.00028 +[ 2917.3s] a member's share on a PARTNER's rows (all on minus that member masked; near 0 = it stays out): s1_perspective on s2_concept -0.2330; s1_perspective on s3_rules -0.1819; s1_perspective on s4_arith -0.1539; s1_perspective on s5_causal -0.0159; s2_concept on s1_perspective -0.0057; s2_concept on s3_rules -0.1071; s2_concept on s4_arith -0.1193; s2_concept on s5_causal -0.0020; s3_rules on s1_perspective -0.0001; s3_rules on s2_concept +0.0008; s3_rules on s4_arith -0.0003; s3_rules on s5_causal +0.0001; s4_arith on s1_perspective +0.0001; s4_arith on s2_concept -0.0005; s4_arith on s3_rules +0.0002; s4_arith on s5_causal +0.0001; s5_causal on s1_perspective +0.0007; s5_causal on s2_concept -0.0006; s5_causal on s3_rules +0.0005; s5_causal on s4_arith +0.0001 +[ 2917.3s] a member ALONE on a partner's rows (alone minus off): s1_perspective on s2_concept -0.4817; s1_perspective on s3_rules -0.4209; s1_perspective on s4_arith -0.5113; s1_perspective on s5_causal -0.1978; s2_concept on s1_perspective -0.0273; s2_concept on s3_rules -0.3108; s2_concept on s4_arith -0.3537; s2_concept on s5_causal -0.0015; s3_rules on s1_perspective -0.0003; s3_rules on s2_concept +0.0031; s3_rules on s4_arith +0.0001; s3_rules on s5_causal +0.0071; s4_arith on s1_perspective -0.0009; s4_arith on s2_concept -0.0004; s4_arith on s3_rules -0.0016; s4_arith on s5_causal -0.0033; s5_causal on s1_perspective -0.0183; s5_causal on s2_concept -0.0109; s5_causal on s3_rules -0.0122; s5_causal on s4_arith -0.0091 +[ 2917.3s] the group on the rows of stages not yet attached: s6_tryfail -0.1896; s7_mixed -0.0081; s8_register -0.1126 +[ 2917.3s] s1_perspective battery accuracy, off -> all on: in 0.550 -> 0.912, transfer 0.463 -> 0.438; alone: in 0.838, transfer 0.463 +[ 2917.3s] s2_concept battery accuracy, off -> all on: in 0.298 -> 0.976, transfer 0.678 -> 0.746; alone: in 0.964, transfer 0.720 +[ 2917.3s] s3_rules battery accuracy, off -> all on: in 0.155 -> 0.880, transfer 0.070 -> 0.300, held-out 0.050 -> 0.508; alone: in 0.320, transfer 0.080, held-out 0.108 +[ 2917.3s] s4_arith battery accuracy, off -> all on: in 0.344 -> 0.606, transfer 0.242 -> 0.229, held-out 0.350 -> 0.338; alone: in 0.381, transfer 0.246, held-out 0.388 +[ 2917.3s] probes mean: off 0.411 -> all on 0.430 +[ 2917.3s] compiled chain: 13 cache entries (limit 256), 0 limit warnings; card peak 14.0 GB; host RSS 5.3 GB +[arms] attached 's6_tryfail' for phase 'curriculum_s6': 13.69M params x 32 blocks, lambda 2.0, spec {'n_slots': 16, 'K': 16, 'D': 8, 'hidden': 256}, recompute on, compile on +[ 2920.9s] STAGE 6/8: s6_tryfail attached over ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal'] (on, frozen); at birth the group reads exactly as before on its rows and on web text; 4000 steps of 1 task + 1 abstention chunks +[ 2922.4s] stage 6/8 s6_tryfail step 1/4000: train bpb s6 0.5003; quiet KL s6 0.00012; 1.53 s/step; the job: 48.7 min spent, about 126.8 min of steps left; card peak 14.0 GB +[ 2956.0s] stage 6/8 s6_tryfail step 50/4000: train bpb s6 0.4784; quiet KL s6 0.00018; 0.70 s/step; the job: 49.3 min spent, about 126.4 min of steps left; card peak 14.1 GB +[ 2989.8s] stage 6/8 s6_tryfail step 100/4000: train bpb s6 0.4475; quiet KL s6 0.00040; 0.69 s/step; the job: 49.8 min spent, about 125.9 min of steps left; card peak 14.1 GB +[ 3023.7s] stage 6/8 s6_tryfail step 150/4000: train bpb s6 0.4025; quiet KL s6 0.00051; 0.69 s/step; the job: 50.4 min spent, about 125.5 min of steps left; card peak 14.1 GB +[ 3057.7s] stage 6/8 s6_tryfail step 200/4000: train bpb s6 0.3530; quiet KL s6 0.00061; 0.68 s/step; the job: 51.0 min spent, about 125.1 min of steps left; card peak 14.1 GB +[ 3091.6s] stage 6/8 s6_tryfail step 250/4000: train bpb s6 0.3117; quiet KL s6 0.00074; 0.68 s/step; the job: 51.5 min spent, about 124.7 min of steps left; card peak 14.1 GB +[ 3125.3s] stage 6/8 s6_tryfail step 300/4000: train bpb s6 0.2728; quiet KL s6 0.00072; 0.68 s/step; the job: 52.1 min spent, about 124.2 min of steps left; card peak 14.1 GB +[ 3159.2s] stage 6/8 s6_tryfail step 350/4000: train bpb s6 0.2478; quiet KL s6 0.00068; 0.68 s/step; the job: 52.7 min spent, about 123.8 min of steps left; card peak 14.1 GB +[ 3193.1s] stage 6/8 s6_tryfail step 400/4000: train bpb s6 0.2154; quiet KL s6 0.00073; 0.68 s/step; the job: 53.2 min spent, about 123.3 min of steps left; card peak 14.1 GB +[ 3226.7s] stage 6/8 s6_tryfail step 450/4000: train bpb s6 0.1939; quiet KL s6 0.00070; 0.68 s/step; the job: 53.8 min spent, about 122.9 min of steps left; card peak 14.1 GB +[ 3260.7s] stage 6/8 s6_tryfail step 500/4000: train bpb s6 0.1731; quiet KL s6 0.00068; 0.68 s/step; the job: 54.3 min spent, about 122.4 min of steps left; card peak 14.1 GB +[ 3294.7s] stage 6/8 s6_tryfail step 550/4000: train bpb s6 0.1561; quiet KL s6 0.00075; 0.68 s/step; the job: 54.9 min spent, about 122.0 min of steps left; card peak 14.1 GB +[ 3328.4s] stage 6/8 s6_tryfail step 600/4000: train bpb s6 0.1364; quiet KL s6 0.00066; 0.68 s/step; the job: 55.5 min spent, about 121.5 min of steps left; card peak 14.1 GB +[ 3362.4s] stage 6/8 s6_tryfail step 650/4000: train bpb s6 0.1226; quiet KL s6 0.00068; 0.68 s/step; the job: 56.0 min spent, about 121.1 min of steps left; card peak 14.1 GB +[ 3396.2s] stage 6/8 s6_tryfail step 700/4000: train bpb s6 0.1139; quiet KL s6 0.00069; 0.68 s/step; the job: 56.6 min spent, about 120.6 min of steps left; card peak 14.1 GB +[ 3429.7s] stage 6/8 s6_tryfail step 750/4000: train bpb s6 0.1028; quiet KL s6 0.00070; 0.68 s/step; the job: 57.2 min spent, about 120.1 min of steps left; card peak 14.1 GB +[ 3463.4s] stage 6/8 s6_tryfail step 800/4000: train bpb s6 0.0953; quiet KL s6 0.00071; 0.68 s/step; the job: 57.7 min spent, about 119.7 min of steps left; card peak 14.1 GB +[ 3497.1s] stage 6/8 s6_tryfail step 850/4000: train bpb s6 0.0907; quiet KL s6 0.00060; 0.68 s/step; the job: 58.3 min spent, about 119.2 min of steps left; card peak 14.1 GB +[ 3530.9s] stage 6/8 s6_tryfail step 900/4000: train bpb s6 0.0918; quiet KL s6 0.00059; 0.68 s/step; the job: 58.8 min spent, about 118.7 min of steps left; card peak 14.1 GB +[ 3564.5s] stage 6/8 s6_tryfail step 950/4000: train bpb s6 0.0908; quiet KL s6 0.00059; 0.68 s/step; the job: 59.4 min spent, about 118.2 min of steps left; card peak 14.1 GB +[ 3598.4s] stage 6/8 s6_tryfail step 1000/4000: train bpb s6 0.0857; quiet KL s6 0.00064; 0.68 s/step; the job: 60.0 min spent, about 117.8 min of steps left; card peak 14.1 GB +[ 3632.1s] stage 6/8 s6_tryfail step 1050/4000: train bpb s6 0.0856; quiet KL s6 0.00057; 0.68 s/step; the job: 60.5 min spent, about 117.3 min of steps left; card peak 14.1 GB +[ 3665.7s] stage 6/8 s6_tryfail step 1100/4000: train bpb s6 0.0815; quiet KL s6 0.00056; 0.68 s/step; the job: 61.1 min spent, about 116.8 min of steps left; card peak 14.1 GB +[ 3699.4s] stage 6/8 s6_tryfail step 1150/4000: train bpb s6 0.0813; quiet KL s6 0.00060; 0.68 s/step; the job: 61.7 min spent, about 116.3 min of steps left; card peak 14.1 GB +[ 3733.1s] stage 6/8 s6_tryfail step 1200/4000: train bpb s6 0.0821; quiet KL s6 0.00064; 0.68 s/step; the job: 62.2 min spent, about 115.9 min of steps left; card peak 14.1 GB +[ 3766.7s] stage 6/8 s6_tryfail step 1250/4000: train bpb s6 0.0793; quiet KL s6 0.00057; 0.68 s/step; the job: 62.8 min spent, about 115.4 min of steps left; card peak 14.1 GB +[ 3800.4s] stage 6/8 s6_tryfail step 1300/4000: train bpb s6 0.0783; quiet KL s6 0.00056; 0.68 s/step; the job: 63.3 min spent, about 114.9 min of steps left; card peak 14.1 GB +[ 3834.2s] stage 6/8 s6_tryfail step 1350/4000: train bpb s6 0.0768; quiet KL s6 0.00059; 0.68 s/step; the job: 63.9 min spent, about 114.4 min of steps left; card peak 14.1 GB +[ 3867.7s] stage 6/8 s6_tryfail step 1400/4000: train bpb s6 0.0758; quiet KL s6 0.00056; 0.68 s/step; the job: 64.5 min spent, about 113.9 min of steps left; card peak 14.1 GB +[ 3901.5s] stage 6/8 s6_tryfail step 1450/4000: train bpb s6 0.0746; quiet KL s6 0.00055; 0.68 s/step; the job: 65.0 min spent, about 113.4 min of steps left; card peak 14.1 GB +[ 3935.3s] stage 6/8 s6_tryfail step 1500/4000: train bpb s6 0.0767; quiet KL s6 0.00057; 0.68 s/step; the job: 65.6 min spent, about 112.9 min of steps left; card peak 14.1 GB +[ 3969.0s] stage 6/8 s6_tryfail step 1550/4000: train bpb s6 0.0743; quiet KL s6 0.00053; 0.68 s/step; the job: 66.2 min spent, about 112.4 min of steps left; card peak 14.1 GB +[ 4002.6s] stage 6/8 s6_tryfail step 1600/4000: train bpb s6 0.0737; quiet KL s6 0.00049; 0.68 s/step; the job: 66.7 min spent, about 111.9 min of steps left; card peak 14.1 GB +[ 4036.4s] stage 6/8 s6_tryfail step 1650/4000: train bpb s6 0.0676; quiet KL s6 0.00053; 0.68 s/step; the job: 67.3 min spent, about 111.4 min of steps left; card peak 14.1 GB +[ 4070.1s] stage 6/8 s6_tryfail step 1700/4000: train bpb s6 0.0693; quiet KL s6 0.00053; 0.68 s/step; the job: 67.8 min spent, about 110.9 min of steps left; card peak 14.1 GB +[ 4103.7s] stage 6/8 s6_tryfail step 1750/4000: train bpb s6 0.0703; quiet KL s6 0.00054; 0.68 s/step; the job: 68.4 min spent, about 110.4 min of steps left; card peak 14.1 GB +[ 4137.5s] stage 6/8 s6_tryfail step 1800/4000: train bpb s6 0.0715; quiet KL s6 0.00053; 0.68 s/step; the job: 69.0 min spent, about 110.0 min of steps left; card peak 14.1 GB +[ 4171.3s] stage 6/8 s6_tryfail step 1850/4000: train bpb s6 0.0684; quiet KL s6 0.00053; 0.68 s/step; the job: 69.5 min spent, about 109.5 min of steps left; card peak 14.1 GB +[ 4204.8s] stage 6/8 s6_tryfail step 1900/4000: train bpb s6 0.0682; quiet KL s6 0.00055; 0.68 s/step; the job: 70.1 min spent, about 108.9 min of steps left; card peak 14.1 GB +[ 4238.7s] stage 6/8 s6_tryfail step 1950/4000: train bpb s6 0.0702; quiet KL s6 0.00054; 0.68 s/step; the job: 70.6 min spent, about 108.5 min of steps left; card peak 14.1 GB +[ 4272.8s] stage 6/8 s6_tryfail step 2000/4000: train bpb s6 0.0696; quiet KL s6 0.00049; 0.68 s/step; the job: 71.2 min spent, about 108.0 min of steps left; card peak 14.1 GB +[ 4306.5s] stage 6/8 s6_tryfail step 2050/4000: train bpb s6 0.0674; quiet KL s6 0.00048; 0.68 s/step; the job: 71.8 min spent, about 107.5 min of steps left; card peak 14.1 GB +[ 4340.4s] stage 6/8 s6_tryfail step 2100/4000: train bpb s6 0.0692; quiet KL s6 0.00048; 0.68 s/step; the job: 72.3 min spent, about 107.0 min of steps left; card peak 14.1 GB +[ 4373.9s] stage 6/8 s6_tryfail step 2150/4000: train bpb s6 0.0685; quiet KL s6 0.00054; 0.68 s/step; the job: 72.9 min spent, about 106.4 min of steps left; card peak 14.1 GB +[ 4407.4s] stage 6/8 s6_tryfail step 2200/4000: train bpb s6 0.0655; quiet KL s6 0.00045; 0.68 s/step; the job: 73.5 min spent, about 105.9 min of steps left; card peak 14.1 GB +[ 4440.7s] stage 6/8 s6_tryfail step 2250/4000: train bpb s6 0.0660; quiet KL s6 0.00047; 0.68 s/step; the job: 74.0 min spent, about 105.4 min of steps left; card peak 14.1 GB +[ 4474.3s] stage 6/8 s6_tryfail step 2300/4000: train bpb s6 0.0644; quiet KL s6 0.00051; 0.68 s/step; the job: 74.6 min spent, about 104.9 min of steps left; card peak 14.1 GB +[ 4508.1s] stage 6/8 s6_tryfail step 2350/4000: train bpb s6 0.0658; quiet KL s6 0.00047; 0.68 s/step; the job: 75.1 min spent, about 104.4 min of steps left; card peak 14.1 GB +[ 4541.6s] stage 6/8 s6_tryfail step 2400/4000: train bpb s6 0.0619; quiet KL s6 0.00042; 0.68 s/step; the job: 75.7 min spent, about 103.9 min of steps left; card peak 14.1 GB +[ 4575.4s] stage 6/8 s6_tryfail step 2450/4000: train bpb s6 0.0624; quiet KL s6 0.00047; 0.68 s/step; the job: 76.3 min spent, about 103.4 min of steps left; card peak 14.1 GB +[ 4609.1s] stage 6/8 s6_tryfail step 2500/4000: train bpb s6 0.0640; quiet KL s6 0.00049; 0.68 s/step; the job: 76.8 min spent, about 102.9 min of steps left; card peak 14.1 GB +[ 4642.6s] stage 6/8 s6_tryfail step 2550/4000: train bpb s6 0.0639; quiet KL s6 0.00041; 0.68 s/step; the job: 77.4 min spent, about 102.3 min of steps left; card peak 14.1 GB +[ 4676.4s] stage 6/8 s6_tryfail step 2600/4000: train bpb s6 0.0632; quiet KL s6 0.00045; 0.68 s/step; the job: 77.9 min spent, about 101.8 min of steps left; card peak 14.1 GB +[ 4710.2s] stage 6/8 s6_tryfail step 2650/4000: train bpb s6 0.0632; quiet KL s6 0.00042; 0.68 s/step; the job: 78.5 min spent, about 101.3 min of steps left; card peak 14.1 GB +[ 4744.0s] stage 6/8 s6_tryfail step 2700/4000: train bpb s6 0.0637; quiet KL s6 0.00045; 0.68 s/step; the job: 79.1 min spent, about 100.8 min of steps left; card peak 14.1 GB +[ 4777.7s] stage 6/8 s6_tryfail step 2750/4000: train bpb s6 0.0623; quiet KL s6 0.00041; 0.68 s/step; the job: 79.6 min spent, about 100.3 min of steps left; card peak 14.1 GB +[ 4811.5s] stage 6/8 s6_tryfail step 2800/4000: train bpb s6 0.0623; quiet KL s6 0.00044; 0.68 s/step; the job: 80.2 min spent, about 99.8 min of steps left; card peak 14.1 GB +[ 4845.3s] stage 6/8 s6_tryfail step 2850/4000: train bpb s6 0.0602; quiet KL s6 0.00043; 0.68 s/step; the job: 80.8 min spent, about 99.3 min of steps left; card peak 14.1 GB +[ 4878.9s] stage 6/8 s6_tryfail step 2900/4000: train bpb s6 0.0633; quiet KL s6 0.00043; 0.68 s/step; the job: 81.3 min spent, about 98.8 min of steps left; card peak 14.1 GB +[ 4912.7s] stage 6/8 s6_tryfail step 2950/4000: train bpb s6 0.0627; quiet KL s6 0.00048; 0.68 s/step; the job: 81.9 min spent, about 98.2 min of steps left; card peak 14.1 GB +[ 4946.5s] stage 6/8 s6_tryfail step 3000/4000: train bpb s6 0.0591; quiet KL s6 0.00041; 0.68 s/step; the job: 82.4 min spent, about 97.7 min of steps left; card peak 14.1 GB +[ 4980.0s] stage 6/8 s6_tryfail step 3050/4000: train bpb s6 0.0617; quiet KL s6 0.00047; 0.68 s/step; the job: 83.0 min spent, about 97.2 min of steps left; card peak 14.1 GB +[ 5013.5s] stage 6/8 s6_tryfail step 3100/4000: train bpb s6 0.0607; quiet KL s6 0.00043; 0.68 s/step; the job: 83.6 min spent, about 96.7 min of steps left; card peak 14.1 GB +[ 5047.0s] stage 6/8 s6_tryfail step 3150/4000: train bpb s6 0.0619; quiet KL s6 0.00047; 0.67 s/step; the job: 84.1 min spent, about 96.1 min of steps left; card peak 14.1 GB +[ 5080.2s] stage 6/8 s6_tryfail step 3200/4000: train bpb s6 0.0608; quiet KL s6 0.00043; 0.67 s/step; the job: 84.7 min spent, about 95.6 min of steps left; card peak 14.1 GB +[ 5113.7s] stage 6/8 s6_tryfail step 3250/4000: train bpb s6 0.0588; quiet KL s6 0.00043; 0.67 s/step; the job: 85.2 min spent, about 95.1 min of steps left; card peak 14.1 GB +[ 5147.2s] stage 6/8 s6_tryfail step 3300/4000: train bpb s6 0.0625; quiet KL s6 0.00041; 0.67 s/step; the job: 85.8 min spent, about 94.6 min of steps left; card peak 14.1 GB +[ 5180.5s] stage 6/8 s6_tryfail step 3350/4000: train bpb s6 0.0616; quiet KL s6 0.00044; 0.67 s/step; the job: 86.3 min spent, about 94.0 min of steps left; card peak 14.1 GB +[ 5214.0s] stage 6/8 s6_tryfail step 3400/4000: train bpb s6 0.0623; quiet KL s6 0.00039; 0.67 s/step; the job: 86.9 min spent, about 93.5 min of steps left; card peak 14.1 GB +[ 5247.6s] stage 6/8 s6_tryfail step 3450/4000: train bpb s6 0.0585; quiet KL s6 0.00043; 0.67 s/step; the job: 87.5 min spent, about 93.0 min of steps left; card peak 14.1 GB +[ 5281.2s] stage 6/8 s6_tryfail step 3500/4000: train bpb s6 0.0624; quiet KL s6 0.00044; 0.67 s/step; the job: 88.0 min spent, about 92.5 min of steps left; card peak 14.1 GB +[ 5314.6s] stage 6/8 s6_tryfail step 3550/4000: train bpb s6 0.0609; quiet KL s6 0.00043; 0.67 s/step; the job: 88.6 min spent, about 91.9 min of steps left; card peak 14.1 GB +[ 5348.1s] stage 6/8 s6_tryfail step 3600/4000: train bpb s6 0.0610; quiet KL s6 0.00039; 0.67 s/step; the job: 89.1 min spent, about 91.4 min of steps left; card peak 14.1 GB +[ 5381.6s] stage 6/8 s6_tryfail step 3650/4000: train bpb s6 0.0584; quiet KL s6 0.00039; 0.67 s/step; the job: 89.7 min spent, about 90.9 min of steps left; card peak 14.1 GB +[ 5414.8s] stage 6/8 s6_tryfail step 3700/4000: train bpb s6 0.0598; quiet KL s6 0.00046; 0.67 s/step; the job: 90.2 min spent, about 90.3 min of steps left; card peak 14.1 GB +[ 5448.3s] stage 6/8 s6_tryfail step 3750/4000: train bpb s6 0.0603; quiet KL s6 0.00038; 0.67 s/step; the job: 90.8 min spent, about 89.8 min of steps left; card peak 14.1 GB +[ 5482.0s] stage 6/8 s6_tryfail step 3800/4000: train bpb s6 0.0596; quiet KL s6 0.00048; 0.67 s/step; the job: 91.4 min spent, about 89.3 min of steps left; card peak 14.1 GB +[ 5515.6s] stage 6/8 s6_tryfail step 3850/4000: train bpb s6 0.0616; quiet KL s6 0.00040; 0.67 s/step; the job: 91.9 min spent, about 88.8 min of steps left; card peak 14.1 GB +[ 5549.4s] stage 6/8 s6_tryfail step 3900/4000: train bpb s6 0.0593; quiet KL s6 0.00047; 0.67 s/step; the job: 92.5 min spent, about 88.2 min of steps left; card peak 14.1 GB +[ 5583.2s] stage 6/8 s6_tryfail step 3950/4000: train bpb s6 0.0593; quiet KL s6 0.00044; 0.67 s/step; the job: 93.1 min spent, about 87.7 min of steps left; card peak 14.1 GB +[ 5616.8s] stage 6/8 s6_tryfail step 4000/4000: train bpb s6 0.0589; quiet KL s6 0.00040; 0.67 s/step; the job: 93.6 min spent, about 87.2 min of steps left; card peak 14.1 GB +[ 5617.8s] anchors saved: /root/refit/group/anchors/gXA_close6_.safetensors for ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail'] +[ 5798.1s] CLOSE 6/8: the group ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail'] all on, against arms off: +[ 5798.1s] s1_perspective rows: off 0.5538 -> all on 0.0488 (-0.5050 +- 0.0048); with s1_perspective masked 0.5048 (its share -0.4560); s1_perspective alone 0.0525 (-0.5013); gate mean 0.210 / widest 0.571 +[ 5798.1s] s2_concept rows: off 0.7212 -> all on 0.1119 (-0.6093 +- 0.0038); with s2_concept masked 0.2370 (its share -0.1252); s2_concept alone 0.3503 (-0.3709); gate mean 0.277 / widest 0.498 +[ 5798.1s] s3_rules rows: off 0.9296 -> all on 0.2241 (-0.7055 +- 0.0053); with s3_rules masked 0.3166 (its share -0.0925); s3_rules alone 0.8236 (-0.1060); gate mean 0.120 / widest 0.174 +[ 5798.1s] s4_arith rows: off 1.2339 -> all on 0.3379 (-0.8960 +- 0.0058); with s4_arith masked 0.4861 (its share -0.1482); s4_arith alone 1.1267 (-0.1072); gate mean 0.108 / widest 0.159 +[ 5798.1s] s5_causal rows: off 0.6604 -> all on 0.0430 (-0.6174 +- 0.0069); with s5_causal masked 0.4749 (its share -0.4319); s5_causal alone 0.0911 (-0.5693); gate mean 0.185 / widest 0.368 +[ 5798.1s] s6_tryfail rows: off 0.6728 -> all on 0.0549 (-0.6179 +- 0.0036); with s6_tryfail masked 0.4832 (its share -0.4283); s6_tryfail alone 0.1638 (-0.5090); gate mean 0.230 / widest 0.537 +[ 5798.1s] web holdout: off 0.9506 -> all on 0.9527 (+0.00214 +- 0.00043; the quiet bar +.012); each member's share: s1_perspective +0.00050, s2_concept -0.00014, s3_rules +0.00032, s4_arith +0.00033, s5_causal +0.00021, s6_tryfail +0.00003; alone: s1_perspective +0.00143, s2_concept +0.00105, s3_rules +0.00053, s4_arith +0.00011, s5_causal +0.00028, s6_tryfail +0.00013 +[ 5798.1s] a member's share on a PARTNER's rows (all on minus that member masked; near 0 = it stays out): s1_perspective on s2_concept -0.2334; s1_perspective on s3_rules -0.1806; s1_perspective on s4_arith -0.1488; s1_perspective on s5_causal -0.0177; s1_perspective on s6_tryfail -0.0385; s2_concept on s1_perspective -0.0055; s2_concept on s3_rules -0.1067; s2_concept on s4_arith -0.1187; s2_concept on s5_causal -0.0017; s2_concept on s6_tryfail -0.0033; s3_rules on s1_perspective +0.0000; s3_rules on s2_concept +0.0008; s3_rules on s4_arith -0.0001; s3_rules on s5_causal +0.0000; s3_rules on s6_tryfail -0.0002; s4_arith on s1_perspective +0.0001; s4_arith on s2_concept -0.0005; s4_arith on s3_rules +0.0001; s4_arith on s5_causal -0.0001; s4_arith on s6_tryfail -0.0044; s5_causal on s1_perspective +0.0006; s5_causal on s2_concept -0.0007; s5_causal on s3_rules +0.0003; s5_causal on s4_arith +0.0002; s5_causal on s6_tryfail -0.0008; s6_tryfail on s1_perspective -0.0001; s6_tryfail on s2_concept +0.0002; s6_tryfail on s3_rules -0.0001; s6_tryfail on s4_arith -0.0006; s6_tryfail on s5_causal -0.0002 +[ 5798.1s] a member ALONE on a partner's rows (alone minus off): s1_perspective on s2_concept -0.4817; s1_perspective on s3_rules -0.4209; s1_perspective on s4_arith -0.5113; s1_perspective on s5_causal -0.1978; s1_perspective on s6_tryfail -0.1782; s2_concept on s1_perspective -0.0273; s2_concept on s3_rules -0.3108; s2_concept on s4_arith -0.3537; s2_concept on s5_causal -0.0015; s2_concept on s6_tryfail +0.0004; s3_rules on s1_perspective -0.0003; s3_rules on s2_concept +0.0031; s3_rules on s4_arith +0.0001; s3_rules on s5_causal +0.0071; s3_rules on s6_tryfail +0.0010; s4_arith on s1_perspective -0.0009; s4_arith on s2_concept -0.0004; s4_arith on s3_rules -0.0016; s4_arith on s5_causal -0.0033; s4_arith on s6_tryfail -0.0135; s5_causal on s1_perspective -0.0183; s5_causal on s2_concept -0.0109; s5_causal on s3_rules -0.0122; s5_causal on s4_arith -0.0091; s5_causal on s6_tryfail -0.0065; s6_tryfail on s1_perspective -0.0020; s6_tryfail on s2_concept +0.0012; s6_tryfail on s3_rules +0.0006; s6_tryfail on s4_arith +0.0036; s6_tryfail on s5_causal -0.0098 +[ 5798.1s] the group on the rows of stages not yet attached: s7_mixed -0.0082; s8_register -0.1128 +[ 5798.1s] s1_perspective battery accuracy, off -> all on: in 0.550 -> 0.887, transfer 0.463 -> 0.425; alone: in 0.838, transfer 0.463 +[ 5798.1s] s2_concept battery accuracy, off -> all on: in 0.298 -> 0.976, transfer 0.678 -> 0.712; alone: in 0.964, transfer 0.720 +[ 5798.1s] s3_rules battery accuracy, off -> all on: in 0.155 -> 0.870, transfer 0.070 -> 0.320, held-out 0.050 -> 0.475; alone: in 0.320, transfer 0.080, held-out 0.108 +[ 5798.1s] s4_arith battery accuracy, off -> all on: in 0.344 -> 0.637, transfer 0.242 -> 0.242, held-out 0.350 -> 0.325; alone: in 0.381, transfer 0.246, held-out 0.388 +[ 5798.1s] probes mean: off 0.411 -> all on 0.433 +[ 5798.1s] compiled chain: 18 cache entries (limit 256), 0 limit warnings; card peak 14.1 GB; host RSS 5.4 GB +[arms] attached 's7_mixed' for phase 'curriculum_s7': 13.69M params x 32 blocks, lambda 2.0, spec {'n_slots': 16, 'K': 16, 'D': 8, 'hidden': 256}, recompute on, compile on +[ 5802.2s] STAGE 7/8: s7_mixed attached over ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail'] (on, frozen); at birth the group reads exactly as before on its rows and on web text; 4000 steps of 1 task + 1 abstention chunks +[ 5803.9s] stage 7/8 s7_mixed step 1/4000: train bpb s7 0.3642; quiet KL s7 0.00017; 1.69 s/step; the job: 96.7 min spent, about 87.2 min of steps left; card peak 14.2 GB +[stream] opening 'tinystories' (roneneldan/TinyStories) … +[stream] opening 'cosmo-young' (HuggingFaceTB/smollm-corpus) … +[stream] opening 'proofwriter-prose' (tasksource/proofwriter) … +[stream] opening 'atomic-flicker' (Estwld/atomic2020-origin) … +[stream] opening 'simple-wiki' (wikimedia/wikipedia) … +[stream] opening 'babi-prose' (Muennighoff/babi) … +Repo card metadata block was not found. Setting CardData to empty. +[stream] opening 'fineweb-good' (HuggingFaceFW/fineweb-edu) … +[ 5866.4s] stage 7/8 s7_mixed step 50/4000: train bpb s7 0.4690; quiet KL s7 0.00015; 1.28 s/step; the job: 97.8 min spent, about 87.2 min of steps left; card peak 14.3 GB +[ 5902.2s] stage 7/8 s7_mixed step 100/4000: train bpb s7 0.5390; quiet KL s7 0.00015; 1.00 s/step; the job: 98.4 min spent, about 86.7 min of steps left; card peak 14.3 GB +[ 5938.1s] stage 7/8 s7_mixed step 150/4000: train bpb s7 0.4871; quiet KL s7 0.00017; 0.91 s/step; the job: 99.0 min spent, about 86.1 min of steps left; card peak 14.3 GB +[ 5974.0s] stage 7/8 s7_mixed step 200/4000: train bpb s7 0.3962; quiet KL s7 0.00021; 0.86 s/step; the job: 99.6 min spent, about 85.6 min of steps left; card peak 14.3 GB +[ 6009.8s] stage 7/8 s7_mixed step 250/4000: train bpb s7 0.4321; quiet KL s7 0.00027; 0.83 s/step; the job: 100.2 min spent, about 85.1 min of steps left; card peak 14.3 GB +[ 6045.6s] stage 7/8 s7_mixed step 300/4000: train bpb s7 0.5527; quiet KL s7 0.00029; 0.81 s/step; the job: 100.8 min spent, about 84.6 min of steps left; card peak 14.3 GB +[ 6081.6s] stage 7/8 s7_mixed step 350/4000: train bpb s7 0.4444; quiet KL s7 0.00040; 0.80 s/step; the job: 101.4 min spent, about 84.1 min of steps left; card peak 14.3 GB +[ 6117.4s] stage 7/8 s7_mixed step 400/4000: train bpb s7 0.4818; quiet KL s7 0.00049; 0.79 s/step; the job: 102.0 min spent, about 83.6 min of steps left; card peak 14.3 GB +[ 6153.3s] stage 7/8 s7_mixed step 450/4000: train bpb s7 0.4726; quiet KL s7 0.00053; 0.78 s/step; the job: 102.6 min spent, about 83.1 min of steps left; card peak 14.3 GB +[ 6189.1s] stage 7/8 s7_mixed step 500/4000: train bpb s7 0.4682; quiet KL s7 0.00043; 0.77 s/step; the job: 103.2 min spent, about 82.6 min of steps left; card peak 14.3 GB +[ 6225.1s] stage 7/8 s7_mixed step 550/4000: train bpb s7 0.4382; quiet KL s7 0.00046; 0.77 s/step; the job: 103.8 min spent, about 82.1 min of steps left; card peak 14.3 GB +[ 6261.0s] stage 7/8 s7_mixed step 600/4000: train bpb s7 0.3918; quiet KL s7 0.00034; 0.76 s/step; the job: 104.4 min spent, about 81.6 min of steps left; card peak 14.3 GB +[ 6297.0s] stage 7/8 s7_mixed step 650/4000: train bpb s7 0.4019; quiet KL s7 0.00050; 0.76 s/step; the job: 104.9 min spent, about 81.1 min of steps left; card peak 14.3 GB +[ 6332.9s] stage 7/8 s7_mixed step 700/4000: train bpb s7 0.4465; quiet KL s7 0.00047; 0.76 s/step; the job: 105.5 min spent, about 80.6 min of steps left; card peak 14.3 GB +[ 6368.8s] stage 7/8 s7_mixed step 750/4000: train bpb s7 0.4083; quiet KL s7 0.00060; 0.76 s/step; the job: 106.1 min spent, about 80.1 min of steps left; card peak 14.3 GB +[ 6404.8s] stage 7/8 s7_mixed step 800/4000: train bpb s7 0.4267; quiet KL s7 0.00043; 0.75 s/step; the job: 106.7 min spent, about 79.5 min of steps left; card peak 14.3 GB +[ 6440.7s] stage 7/8 s7_mixed step 850/4000: train bpb s7 0.3861; quiet KL s7 0.00065; 0.75 s/step; the job: 107.3 min spent, about 79.0 min of steps left; card peak 14.3 GB +[ 6476.6s] stage 7/8 s7_mixed step 900/4000: train bpb s7 0.4337; quiet KL s7 0.00052; 0.75 s/step; the job: 107.9 min spent, about 78.5 min of steps left; card peak 14.3 GB +[ 6512.6s] stage 7/8 s7_mixed step 950/4000: train bpb s7 0.5111; quiet KL s7 0.00066; 0.75 s/step; the job: 108.5 min spent, about 78.0 min of steps left; card peak 14.3 GB +[ 6548.5s] stage 7/8 s7_mixed step 1000/4000: train bpb s7 0.4723; quiet KL s7 0.00064; 0.75 s/step; the job: 109.1 min spent, about 77.5 min of steps left; card peak 14.3 GB +[ 6584.5s] stage 7/8 s7_mixed step 1050/4000: train bpb s7 0.5203; quiet KL s7 0.00074; 0.75 s/step; the job: 109.7 min spent, about 77.0 min of steps left; card peak 14.3 GB +[ 6620.5s] stage 7/8 s7_mixed step 1100/4000: train bpb s7 0.3957; quiet KL s7 0.00068; 0.74 s/step; the job: 110.3 min spent, about 76.4 min of steps left; card peak 14.3 GB +[ 6656.4s] stage 7/8 s7_mixed step 1150/4000: train bpb s7 0.4222; quiet KL s7 0.00083; 0.74 s/step; the job: 110.9 min spent, about 75.9 min of steps left; card peak 14.3 GB +[ 6692.2s] stage 7/8 s7_mixed step 1200/4000: train bpb s7 0.4063; quiet KL s7 0.00053; 0.74 s/step; the job: 111.5 min spent, about 75.4 min of steps left; card peak 14.3 GB +[ 6727.9s] stage 7/8 s7_mixed step 1250/4000: train bpb s7 0.4492; quiet KL s7 0.00059; 0.74 s/step; the job: 112.1 min spent, about 74.9 min of steps left; card peak 14.3 GB +[ 6763.6s] stage 7/8 s7_mixed step 1300/4000: train bpb s7 0.5419; quiet KL s7 0.00079; 0.74 s/step; the job: 112.7 min spent, about 74.3 min of steps left; card peak 14.3 GB +[ 6799.4s] stage 7/8 s7_mixed step 1350/4000: train bpb s7 0.3960; quiet KL s7 0.00090; 0.74 s/step; the job: 113.3 min spent, about 73.8 min of steps left; card peak 14.3 GB +[ 6835.1s] stage 7/8 s7_mixed step 1400/4000: train bpb s7 0.3689; quiet KL s7 0.00061; 0.74 s/step; the job: 113.9 min spent, about 73.3 min of steps left; card peak 14.3 GB +[ 6870.9s] stage 7/8 s7_mixed step 1450/4000: train bpb s7 0.3995; quiet KL s7 0.00059; 0.74 s/step; the job: 114.5 min spent, about 72.8 min of steps left; card peak 14.3 GB +[ 6906.6s] stage 7/8 s7_mixed step 1500/4000: train bpb s7 0.4601; quiet KL s7 0.00056; 0.74 s/step; the job: 115.1 min spent, about 72.2 min of steps left; card peak 14.3 GB +[ 6942.4s] stage 7/8 s7_mixed step 1550/4000: train bpb s7 0.4147; quiet KL s7 0.00057; 0.74 s/step; the job: 115.7 min spent, about 71.7 min of steps left; card peak 14.3 GB +[ 6978.2s] stage 7/8 s7_mixed step 1600/4000: train bpb s7 0.3934; quiet KL s7 0.00057; 0.74 s/step; the job: 116.3 min spent, about 71.2 min of steps left; card peak 14.3 GB +[ 7013.9s] stage 7/8 s7_mixed step 1650/4000: train bpb s7 0.3674; quiet KL s7 0.00062; 0.73 s/step; the job: 116.9 min spent, about 70.7 min of steps left; card peak 14.3 GB +[ 7049.8s] stage 7/8 s7_mixed step 1700/4000: train bpb s7 0.4999; quiet KL s7 0.00057; 0.73 s/step; the job: 117.5 min spent, about 70.1 min of steps left; card peak 14.3 GB +[ 7085.6s] stage 7/8 s7_mixed step 1750/4000: train bpb s7 0.4460; quiet KL s7 0.00056; 0.73 s/step; the job: 118.1 min spent, about 69.6 min of steps left; card peak 14.3 GB +[ 7121.5s] stage 7/8 s7_mixed step 1800/4000: train bpb s7 0.5317; quiet KL s7 0.00065; 0.73 s/step; the job: 118.7 min spent, about 69.1 min of steps left; card peak 14.3 GB +[ 7157.5s] stage 7/8 s7_mixed step 1850/4000: train bpb s7 0.4875; quiet KL s7 0.00065; 0.73 s/step; the job: 119.3 min spent, about 68.5 min of steps left; card peak 14.3 GB +[ 7193.4s] stage 7/8 s7_mixed step 1900/4000: train bpb s7 0.4753; quiet KL s7 0.00065; 0.73 s/step; the job: 119.9 min spent, about 68.0 min of steps left; card peak 14.3 GB +[ 7229.1s] stage 7/8 s7_mixed step 1950/4000: train bpb s7 0.4237; quiet KL s7 0.00068; 0.73 s/step; the job: 120.5 min spent, about 67.5 min of steps left; card peak 14.3 GB +[ 7265.0s] stage 7/8 s7_mixed step 2000/4000: train bpb s7 0.3462; quiet KL s7 0.00062; 0.73 s/step; the job: 121.1 min spent, about 66.9 min of steps left; card peak 14.3 GB +[ 7300.8s] stage 7/8 s7_mixed step 2050/4000: train bpb s7 0.4922; quiet KL s7 0.00057; 0.73 s/step; the job: 121.7 min spent, about 66.4 min of steps left; card peak 14.3 GB +[ 7336.8s] stage 7/8 s7_mixed step 2100/4000: train bpb s7 0.3943; quiet KL s7 0.00059; 0.73 s/step; the job: 122.3 min spent, about 65.9 min of steps left; card peak 14.3 GB +[ 7373.1s] stage 7/8 s7_mixed step 2150/4000: train bpb s7 0.4072; quiet KL s7 0.00056; 0.73 s/step; the job: 122.9 min spent, about 65.3 min of steps left; card peak 14.3 GB +[ 7409.3s] stage 7/8 s7_mixed step 2200/4000: train bpb s7 0.3955; quiet KL s7 0.00061; 0.73 s/step; the job: 123.5 min spent, about 64.8 min of steps left; card peak 14.3 GB +[ 7445.6s] stage 7/8 s7_mixed step 2250/4000: train bpb s7 0.2781; quiet KL s7 0.00052; 0.73 s/step; the job: 124.1 min spent, about 64.3 min of steps left; card peak 14.3 GB +[ 7481.8s] stage 7/8 s7_mixed step 2300/4000: train bpb s7 0.3987; quiet KL s7 0.00059; 0.73 s/step; the job: 124.7 min spent, about 63.7 min of steps left; card peak 14.3 GB +[ 7518.0s] stage 7/8 s7_mixed step 2350/4000: train bpb s7 0.3842; quiet KL s7 0.00067; 0.73 s/step; the job: 125.3 min spent, about 63.2 min of steps left; card peak 14.3 GB +[ 7554.2s] stage 7/8 s7_mixed step 2400/4000: train bpb s7 0.4196; quiet KL s7 0.00062; 0.73 s/step; the job: 125.9 min spent, about 62.7 min of steps left; card peak 14.3 GB +[ 7590.3s] stage 7/8 s7_mixed step 2450/4000: train bpb s7 0.4184; quiet KL s7 0.00073; 0.73 s/step; the job: 126.5 min spent, about 62.1 min of steps left; card peak 14.3 GB +[ 7626.3s] stage 7/8 s7_mixed step 2500/4000: train bpb s7 0.4866; quiet KL s7 0.00063; 0.73 s/step; the job: 127.1 min spent, about 61.6 min of steps left; card peak 14.3 GB +[ 7662.2s] stage 7/8 s7_mixed step 2550/4000: train bpb s7 0.3877; quiet KL s7 0.00070; 0.73 s/step; the job: 127.7 min spent, about 61.0 min of steps left; card peak 14.3 GB +[ 7698.1s] stage 7/8 s7_mixed step 2600/4000: train bpb s7 0.4071; quiet KL s7 0.00067; 0.73 s/step; the job: 128.3 min spent, about 60.5 min of steps left; card peak 14.3 GB +[ 7734.0s] stage 7/8 s7_mixed step 2650/4000: train bpb s7 0.3705; quiet KL s7 0.00061; 0.73 s/step; the job: 128.9 min spent, about 60.0 min of steps left; card peak 14.3 GB +[ 7769.9s] stage 7/8 s7_mixed step 2700/4000: train bpb s7 0.4671; quiet KL s7 0.00066; 0.73 s/step; the job: 129.5 min spent, about 59.4 min of steps left; card peak 14.3 GB +[ 7805.8s] stage 7/8 s7_mixed step 2750/4000: train bpb s7 0.3768; quiet KL s7 0.00065; 0.73 s/step; the job: 130.1 min spent, about 58.9 min of steps left; card peak 14.3 GB +[ 7841.7s] stage 7/8 s7_mixed step 2800/4000: train bpb s7 0.4549; quiet KL s7 0.00063; 0.73 s/step; the job: 130.7 min spent, about 58.3 min of steps left; card peak 14.3 GB +[ 7877.6s] stage 7/8 s7_mixed step 2850/4000: train bpb s7 0.4400; quiet KL s7 0.00060; 0.73 s/step; the job: 131.3 min spent, about 57.8 min of steps left; card peak 14.3 GB +[ 7913.6s] stage 7/8 s7_mixed step 2900/4000: train bpb s7 0.4118; quiet KL s7 0.00063; 0.73 s/step; the job: 131.9 min spent, about 57.3 min of steps left; card peak 14.3 GB +[ 7949.5s] stage 7/8 s7_mixed step 2950/4000: train bpb s7 0.3883; quiet KL s7 0.00064; 0.73 s/step; the job: 132.5 min spent, about 56.7 min of steps left; card peak 14.3 GB +[ 7985.5s] stage 7/8 s7_mixed step 3000/4000: train bpb s7 0.4967; quiet KL s7 0.00061; 0.73 s/step; the job: 133.1 min spent, about 56.2 min of steps left; card peak 14.3 GB +[ 8021.5s] stage 7/8 s7_mixed step 3050/4000: train bpb s7 0.4448; quiet KL s7 0.00069; 0.73 s/step; the job: 133.7 min spent, about 55.6 min of steps left; card peak 14.3 GB +[ 8057.5s] stage 7/8 s7_mixed step 3100/4000: train bpb s7 0.3773; quiet KL s7 0.00074; 0.73 s/step; the job: 134.3 min spent, about 55.1 min of steps left; card peak 14.3 GB +[ 8093.4s] stage 7/8 s7_mixed step 3150/4000: train bpb s7 0.3895; quiet KL s7 0.00067; 0.73 s/step; the job: 134.9 min spent, about 54.5 min of steps left; card peak 14.3 GB +[ 8129.4s] stage 7/8 s7_mixed step 3200/4000: train bpb s7 0.3850; quiet KL s7 0.00074; 0.73 s/step; the job: 135.5 min spent, about 54.0 min of steps left; card peak 14.3 GB +[ 8165.3s] stage 7/8 s7_mixed step 3250/4000: train bpb s7 0.4035; quiet KL s7 0.00067; 0.73 s/step; the job: 136.1 min spent, about 53.4 min of steps left; card peak 14.3 GB +[ 8201.2s] stage 7/8 s7_mixed step 3300/4000: train bpb s7 0.4375; quiet KL s7 0.00062; 0.73 s/step; the job: 136.7 min spent, about 52.9 min of steps left; card peak 14.3 GB +[ 8237.1s] stage 7/8 s7_mixed step 3350/4000: train bpb s7 0.4425; quiet KL s7 0.00064; 0.73 s/step; the job: 137.3 min spent, about 52.3 min of steps left; card peak 14.3 GB +[ 8273.1s] stage 7/8 s7_mixed step 3400/4000: train bpb s7 0.3765; quiet KL s7 0.00062; 0.73 s/step; the job: 137.9 min spent, about 51.8 min of steps left; card peak 14.3 GB +[ 8309.0s] stage 7/8 s7_mixed step 3450/4000: train bpb s7 0.3871; quiet KL s7 0.00067; 0.73 s/step; the job: 138.5 min spent, about 51.2 min of steps left; card peak 14.3 GB +[ 8345.0s] stage 7/8 s7_mixed step 3500/4000: train bpb s7 0.4083; quiet KL s7 0.00075; 0.73 s/step; the job: 139.1 min spent, about 50.7 min of steps left; card peak 14.3 GB +[ 8381.0s] stage 7/8 s7_mixed step 3550/4000: train bpb s7 0.3569; quiet KL s7 0.00068; 0.73 s/step; the job: 139.7 min spent, about 50.1 min of steps left; card peak 14.3 GB +[ 8417.0s] stage 7/8 s7_mixed step 3600/4000: train bpb s7 0.3614; quiet KL s7 0.00076; 0.73 s/step; the job: 140.3 min spent, about 49.6 min of steps left; card peak 14.3 GB +[ 8452.9s] stage 7/8 s7_mixed step 3650/4000: train bpb s7 0.4565; quiet KL s7 0.00065; 0.73 s/step; the job: 140.9 min spent, about 49.0 min of steps left; card peak 14.3 GB +[ 8488.9s] stage 7/8 s7_mixed step 3700/4000: train bpb s7 0.3751; quiet KL s7 0.00064; 0.73 s/step; the job: 141.5 min spent, about 48.5 min of steps left; card peak 14.3 GB +[ 8524.9s] stage 7/8 s7_mixed step 3750/4000: train bpb s7 0.4692; quiet KL s7 0.00061; 0.73 s/step; the job: 142.1 min spent, about 47.9 min of steps left; card peak 14.3 GB +[ 8560.9s] stage 7/8 s7_mixed step 3800/4000: train bpb s7 0.4175; quiet KL s7 0.00115; 0.73 s/step; the job: 142.7 min spent, about 47.4 min of steps left; card peak 14.3 GB +[ 8596.9s] stage 7/8 s7_mixed step 3850/4000: train bpb s7 0.4314; quiet KL s7 0.00083; 0.73 s/step; the job: 143.3 min spent, about 46.8 min of steps left; card peak 14.3 GB +[ 8632.9s] stage 7/8 s7_mixed step 3900/4000: train bpb s7 0.3549; quiet KL s7 0.00071; 0.73 s/step; the job: 143.9 min spent, about 46.3 min of steps left; card peak 14.3 GB +[ 8668.6s] stage 7/8 s7_mixed step 3950/4000: train bpb s7 0.4113; quiet KL s7 0.00073; 0.73 s/step; the job: 144.5 min spent, about 45.7 min of steps left; card peak 14.3 GB +[ 8704.6s] stage 7/8 s7_mixed step 4000/4000: train bpb s7 0.4132; quiet KL s7 0.00065; 0.73 s/step; the job: 145.1 min spent, about 45.2 min of steps left; card peak 14.3 GB +[ 8706.0s] anchors saved: /root/refit/group/anchors/gXA_close7_.safetensors for ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail', 's7_mixed'] +[ 8915.9s] CLOSE 7/8: the group ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail', 's7_mixed'] all on, against arms off: +[ 8915.9s] s1_perspective rows: off 0.5538 -> all on 0.0485 (-0.5053 +- 0.0048); with s1_perspective masked 0.4463 (its share -0.3978); s1_perspective alone 0.0525 (-0.5013); gate mean 0.210 / widest 0.571 +[ 8915.9s] s2_concept rows: off 0.7212 -> all on 0.1112 (-0.6099 +- 0.0038); with s2_concept masked 0.2362 (its share -0.1250); s2_concept alone 0.3503 (-0.3709); gate mean 0.277 / widest 0.498 +[ 8915.9s] s3_rules rows: off 0.9296 -> all on 0.2232 (-0.7064 +- 0.0054); with s3_rules masked 0.3144 (its share -0.0912); s3_rules alone 0.8236 (-0.1060); gate mean 0.120 / widest 0.174 +[ 8915.9s] s4_arith rows: off 1.2339 -> all on 0.3279 (-0.9060 +- 0.0062); with s4_arith masked 0.4726 (its share -0.1447); s4_arith alone 1.1267 (-0.1072); gate mean 0.108 / widest 0.159 +[ 8915.9s] s5_causal rows: off 0.6604 -> all on 0.0429 (-0.6176 +- 0.0069); with s5_causal masked 0.4558 (its share -0.4130); s5_causal alone 0.0911 (-0.5693); gate mean 0.185 / widest 0.368 +[ 8915.9s] s6_tryfail rows: off 0.6728 -> all on 0.0546 (-0.6182 +- 0.0036); with s6_tryfail masked 0.4697 (its share -0.4151); s6_tryfail alone 0.1638 (-0.5090); gate mean 0.230 / widest 0.537 +[ 8915.9s] s7_mixed rows: off 0.7830 -> all on 0.7188 (-0.0642 +- 0.0392); with s7_mixed masked 0.7748 (its share -0.0560); s7_mixed alone 0.7242 (-0.0588); gate mean 0.251 / widest 0.464 +[ 8915.9s] web holdout: off 0.9506 -> all on 0.9520 (+0.00144 +- 0.00030; the quiet bar +.012); each member's share: s1_perspective +0.00030, s2_concept -0.00018, s3_rules +0.00036, s4_arith +0.00025, s5_causal +0.00031, s6_tryfail -0.00010, s7_mixed -0.00070; alone: s1_perspective +0.00143, s2_concept +0.00105, s3_rules +0.00053, s4_arith +0.00011, s5_causal +0.00028, s6_tryfail +0.00013, s7_mixed -0.00019 +[ 8915.9s] a member's share on a PARTNER's rows (all on minus that member masked; near 0 = it stays out): s1_perspective on s2_concept -0.2226; s1_perspective on s3_rules -0.1821; s1_perspective on s4_arith -0.1268; s1_perspective on s5_causal -0.0149; s1_perspective on s6_tryfail -0.0344; s1_perspective on s7_mixed -0.0038; s2_concept on s1_perspective -0.0049; s2_concept on s3_rules -0.1078; s2_concept on s4_arith -0.1064; s2_concept on s5_causal -0.0014; s2_concept on s6_tryfail -0.0035; s2_concept on s7_mixed -0.0009; s3_rules on s1_perspective +0.0000; s3_rules on s2_concept +0.0007; s3_rules on s4_arith -0.0001; s3_rules on s5_causal -0.0000; s3_rules on s6_tryfail -0.0002; s3_rules on s7_mixed +0.0001; s4_arith on s1_perspective -0.0000; s4_arith on s2_concept -0.0006; s4_arith on s3_rules -0.0000; s4_arith on s5_causal -0.0000; s4_arith on s6_tryfail -0.0040; s4_arith on s7_mixed +0.0000; s5_causal on s1_perspective +0.0002; s5_causal on s2_concept -0.0007; s5_causal on s3_rules +0.0002; s5_causal on s4_arith +0.0001; s5_causal on s6_tryfail -0.0009; s5_causal on s7_mixed -0.0001; s6_tryfail on s1_perspective +0.0000; s6_tryfail on s2_concept +0.0004; s6_tryfail on s3_rules +0.0000; s6_tryfail on s4_arith -0.0008; s6_tryfail on s5_causal -0.0001; s6_tryfail on s7_mixed -0.0001; s7_mixed on s1_perspective -0.0003; s7_mixed on s2_concept -0.0006; s7_mixed on s3_rules -0.0009; s7_mixed on s4_arith -0.0100; s7_mixed on s5_causal -0.0001; s7_mixed on s6_tryfail -0.0003 +[ 8915.9s] a member ALONE on a partner's rows (alone minus off): s1_perspective on s2_concept -0.4817; s1_perspective on s3_rules -0.4209; s1_perspective on s4_arith -0.5113; s1_perspective on s5_causal -0.1978; s1_perspective on s6_tryfail -0.1782; s1_perspective on s7_mixed -0.0101; s2_concept on s1_perspective -0.0273; s2_concept on s3_rules -0.3108; s2_concept on s4_arith -0.3537; s2_concept on s5_causal -0.0015; s2_concept on s6_tryfail +0.0004; s2_concept on s7_mixed -0.0000; s3_rules on s1_perspective -0.0003; s3_rules on s2_concept +0.0031; s3_rules on s4_arith +0.0001; s3_rules on s5_causal +0.0071; s3_rules on s6_tryfail +0.0010; s3_rules on s7_mixed -0.0005; s4_arith on s1_perspective -0.0009; s4_arith on s2_concept -0.0004; s4_arith on s3_rules -0.0016; s4_arith on s5_causal -0.0033; s4_arith on s6_tryfail -0.0135; s4_arith on s7_mixed -0.0002; s5_causal on s1_perspective -0.0183; s5_causal on s2_concept -0.0109; s5_causal on s3_rules -0.0122; s5_causal on s4_arith -0.0091; s5_causal on s6_tryfail -0.0065; s5_causal on s7_mixed -0.0011; s6_tryfail on s1_perspective -0.0020; s6_tryfail on s2_concept +0.0012; s6_tryfail on s3_rules +0.0006; s6_tryfail on s4_arith +0.0036; s6_tryfail on s5_causal -0.0098; s6_tryfail on s7_mixed -0.0000; s7_mixed on s1_perspective -0.0568; s7_mixed on s2_concept -0.0058; s7_mixed on s3_rules -0.0617; s7_mixed on s4_arith -0.0385; s7_mixed on s5_causal -0.1028; s7_mixed on s6_tryfail -0.0802 +[ 8915.9s] the group on the rows of stages not yet attached: s8_register -0.1149 +[ 8915.9s] s1_perspective battery accuracy, off -> all on: in 0.550 -> 0.900, transfer 0.463 -> 0.412; alone: in 0.838, transfer 0.463 +[ 8915.9s] s2_concept battery accuracy, off -> all on: in 0.298 -> 0.988, transfer 0.678 -> 0.686; alone: in 0.964, transfer 0.720 +[ 8915.9s] s3_rules battery accuracy, off -> all on: in 0.155 -> 0.870, transfer 0.070 -> 0.320, held-out 0.050 -> 0.442; alone: in 0.320, transfer 0.080, held-out 0.108 +[ 8915.9s] s4_arith battery accuracy, off -> all on: in 0.344 -> 0.644, transfer 0.242 -> 0.242, held-out 0.350 -> 0.325; alone: in 0.381, transfer 0.246, held-out 0.388 +[ 8915.9s] probes mean: off 0.411 -> all on 0.419 +[ 8915.9s] compiled chain: 23 cache entries (limit 256), 0 limit warnings; card peak 14.3 GB; host RSS 11.5 GB +[arms] attached 's8_register' for phase 'curriculum_s8': 13.69M params x 32 blocks, lambda 2.0, spec {'n_slots': 16, 'K': 16, 'D': 8, 'hidden': 256}, recompute on, compile on +[ 8919.9s] STAGE 8/8: s8_register attached over ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail', 's7_mixed'] (on, frozen); at birth the group reads exactly as before on its rows and on web text; 4000 steps of 1 task + 1 abstention chunks +[ 8921.9s] stage 8/8 s8_register step 1/4000: train bpb s8 0.3285; quiet KL s8 0.00012; 1.94 s/step; the job: 148.7 min spent, about 45.2 min of steps left; card peak 14.4 GB +[ 8959.0s] stage 8/8 s8_register step 50/4000: train bpb s8 0.3734; quiet KL s8 0.00015; 0.78 s/step; the job: 149.3 min spent, about 44.6 min of steps left; card peak 14.5 GB +[ 8997.1s] stage 8/8 s8_register step 100/4000: train bpb s8 0.3524; quiet KL s8 0.00029; 0.77 s/step; the job: 150.0 min spent, about 44.1 min of steps left; card peak 14.5 GB +[ 9035.0s] stage 8/8 s8_register step 150/4000: train bpb s8 0.3352; quiet KL s8 0.00045; 0.77 s/step; the job: 150.6 min spent, about 43.6 min of steps left; card peak 14.5 GB +[ 9073.4s] stage 8/8 s8_register step 200/4000: train bpb s8 0.3001; quiet KL s8 0.00057; 0.77 s/step; the job: 151.2 min spent, about 43.0 min of steps left; card peak 14.5 GB +[ 9111.4s] stage 8/8 s8_register step 250/4000: train bpb s8 0.2690; quiet KL s8 0.00075; 0.77 s/step; the job: 151.9 min spent, about 42.5 min of steps left; card peak 14.5 GB +[ 9149.8s] stage 8/8 s8_register step 300/4000: train bpb s8 0.2341; quiet KL s8 0.00092; 0.77 s/step; the job: 152.5 min spent, about 41.9 min of steps left; card peak 14.5 GB +[ 9187.8s] stage 8/8 s8_register step 350/4000: train bpb s8 0.1983; quiet KL s8 0.00088; 0.77 s/step; the job: 153.1 min spent, about 41.4 min of steps left; card peak 14.5 GB +[ 9226.0s] stage 8/8 s8_register step 400/4000: train bpb s8 0.1662; quiet KL s8 0.00080; 0.77 s/step; the job: 153.8 min spent, about 40.8 min of steps left; card peak 14.5 GB +[ 9263.8s] stage 8/8 s8_register step 450/4000: train bpb s8 0.1412; quiet KL s8 0.00079; 0.76 s/step; the job: 154.4 min spent, about 40.3 min of steps left; card peak 14.5 GB +[ 9301.9s] stage 8/8 s8_register step 500/4000: train bpb s8 0.1173; quiet KL s8 0.00079; 0.76 s/step; the job: 155.0 min spent, about 39.7 min of steps left; card peak 14.5 GB +[ 9339.8s] stage 8/8 s8_register step 550/4000: train bpb s8 0.0982; quiet KL s8 0.00075; 0.76 s/step; the job: 155.7 min spent, about 39.2 min of steps left; card peak 14.5 GB +[ 9378.1s] stage 8/8 s8_register step 600/4000: train bpb s8 0.0812; quiet KL s8 0.00078; 0.76 s/step; the job: 156.3 min spent, about 38.6 min of steps left; card peak 14.5 GB +[ 9416.2s] stage 8/8 s8_register step 650/4000: train bpb s8 0.0693; quiet KL s8 0.00076; 0.76 s/step; the job: 156.9 min spent, about 38.1 min of steps left; card peak 14.5 GB +[ 9454.6s] stage 8/8 s8_register step 700/4000: train bpb s8 0.0593; quiet KL s8 0.00077; 0.76 s/step; the job: 157.6 min spent, about 37.5 min of steps left; card peak 14.5 GB +[ 9492.7s] stage 8/8 s8_register step 750/4000: train bpb s8 0.0497; quiet KL s8 0.00076; 0.76 s/step; the job: 158.2 min spent, about 37.0 min of steps left; card peak 14.5 GB +[ 9531.2s] stage 8/8 s8_register step 800/4000: train bpb s8 0.0437; quiet KL s8 0.00075; 0.76 s/step; the job: 158.9 min spent, about 36.4 min of steps left; card peak 14.5 GB +[ 9569.4s] stage 8/8 s8_register step 850/4000: train bpb s8 0.0404; quiet KL s8 0.00069; 0.76 s/step; the job: 159.5 min spent, about 35.9 min of steps left; card peak 14.5 GB +[ 9608.0s] stage 8/8 s8_register step 900/4000: train bpb s8 0.0361; quiet KL s8 0.00074; 0.76 s/step; the job: 160.1 min spent, about 35.3 min of steps left; card peak 14.5 GB +[ 9646.2s] stage 8/8 s8_register step 950/4000: train bpb s8 0.0334; quiet KL s8 0.00060; 0.76 s/step; the job: 160.8 min spent, about 34.8 min of steps left; card peak 14.5 GB +[ 9684.8s] stage 8/8 s8_register step 1000/4000: train bpb s8 0.0326; quiet KL s8 0.00065; 0.76 s/step; the job: 161.4 min spent, about 34.2 min of steps left; card peak 14.5 GB +[ 9722.9s] stage 8/8 s8_register step 1050/4000: train bpb s8 0.0301; quiet KL s8 0.00058; 0.76 s/step; the job: 162.0 min spent, about 33.7 min of steps left; card peak 14.5 GB +[ 9761.4s] stage 8/8 s8_register step 1100/4000: train bpb s8 0.0295; quiet KL s8 0.00055; 0.76 s/step; the job: 162.7 min spent, about 33.1 min of steps left; card peak 14.5 GB +[ 9799.6s] stage 8/8 s8_register step 1150/4000: train bpb s8 0.0290; quiet KL s8 0.00057; 0.76 s/step; the job: 163.3 min spent, about 32.6 min of steps left; card peak 14.5 GB +[ 9838.1s] stage 8/8 s8_register step 1200/4000: train bpb s8 0.0286; quiet KL s8 0.00055; 0.77 s/step; the job: 164.0 min spent, about 32.0 min of steps left; card peak 14.5 GB +[ 9876.3s] stage 8/8 s8_register step 1250/4000: train bpb s8 0.0293; quiet KL s8 0.00068; 0.77 s/step; the job: 164.6 min spent, about 31.4 min of steps left; card peak 14.5 GB +[ 9914.7s] stage 8/8 s8_register step 1300/4000: train bpb s8 0.0272; quiet KL s8 0.00054; 0.77 s/step; the job: 165.2 min spent, about 30.9 min of steps left; card peak 14.5 GB +[ 9953.0s] stage 8/8 s8_register step 1350/4000: train bpb s8 0.0264; quiet KL s8 0.00050; 0.77 s/step; the job: 165.9 min spent, about 30.3 min of steps left; card peak 14.5 GB +[ 9991.5s] stage 8/8 s8_register step 1400/4000: train bpb s8 0.0263; quiet KL s8 0.00048; 0.77 s/step; the job: 166.5 min spent, about 29.8 min of steps left; card peak 14.5 GB +[10029.7s] stage 8/8 s8_register step 1450/4000: train bpb s8 0.0255; quiet KL s8 0.00047; 0.77 s/step; the job: 167.2 min spent, about 29.2 min of steps left; card peak 14.5 GB +[10068.1s] stage 8/8 s8_register step 1500/4000: train bpb s8 0.0251; quiet KL s8 0.00046; 0.77 s/step; the job: 167.8 min spent, about 28.6 min of steps left; card peak 14.5 GB +[10106.4s] stage 8/8 s8_register step 1550/4000: train bpb s8 0.0258; quiet KL s8 0.00045; 0.77 s/step; the job: 168.4 min spent, about 28.1 min of steps left; card peak 14.5 GB +[10144.6s] stage 8/8 s8_register step 1600/4000: train bpb s8 0.0254; quiet KL s8 0.00044; 0.77 s/step; the job: 169.1 min spent, about 27.5 min of steps left; card peak 14.5 GB +[10183.1s] stage 8/8 s8_register step 1650/4000: train bpb s8 0.0253; quiet KL s8 0.00042; 0.77 s/step; the job: 169.7 min spent, about 27.0 min of steps left; card peak 14.5 GB +[10221.3s] stage 8/8 s8_register step 1700/4000: train bpb s8 0.0252; quiet KL s8 0.00038; 0.77 s/step; the job: 170.4 min spent, about 26.4 min of steps left; card peak 14.5 GB +[10259.9s] stage 8/8 s8_register step 1750/4000: train bpb s8 0.0249; quiet KL s8 0.00040; 0.77 s/step; the job: 171.0 min spent, about 25.8 min of steps left; card peak 14.5 GB +[10298.1s] stage 8/8 s8_register step 1800/4000: train bpb s8 0.0243; quiet KL s8 0.00041; 0.77 s/step; the job: 171.6 min spent, about 25.3 min of steps left; card peak 14.5 GB +[10336.5s] stage 8/8 s8_register step 1850/4000: train bpb s8 0.0244; quiet KL s8 0.00037; 0.77 s/step; the job: 172.3 min spent, about 24.7 min of steps left; card peak 14.5 GB +[10374.6s] stage 8/8 s8_register step 1900/4000: train bpb s8 0.0246; quiet KL s8 0.00037; 0.77 s/step; the job: 172.9 min spent, about 24.1 min of steps left; card peak 14.5 GB +[10413.2s] stage 8/8 s8_register step 1950/4000: train bpb s8 0.0241; quiet KL s8 0.00041; 0.77 s/step; the job: 173.6 min spent, about 23.6 min of steps left; card peak 14.5 GB +[10451.5s] stage 8/8 s8_register step 2000/4000: train bpb s8 0.0244; quiet KL s8 0.00039; 0.77 s/step; the job: 174.2 min spent, about 23.0 min of steps left; card peak 14.5 GB +[10490.1s] stage 8/8 s8_register step 2050/4000: train bpb s8 0.0241; quiet KL s8 0.00044; 0.77 s/step; the job: 174.8 min spent, about 22.4 min of steps left; card peak 14.5 GB +[10528.3s] stage 8/8 s8_register step 2100/4000: train bpb s8 0.0240; quiet KL s8 0.00038; 0.77 s/step; the job: 175.5 min spent, about 21.9 min of steps left; card peak 14.5 GB +[10566.9s] stage 8/8 s8_register step 2150/4000: train bpb s8 0.0237; quiet KL s8 0.00038; 0.77 s/step; the job: 176.1 min spent, about 21.3 min of steps left; card peak 14.5 GB +[10605.2s] stage 8/8 s8_register step 2200/4000: train bpb s8 0.0244; quiet KL s8 0.00039; 0.77 s/step; the job: 176.8 min spent, about 20.7 min of steps left; card peak 14.5 GB +[10643.7s] stage 8/8 s8_register step 2250/4000: train bpb s8 0.0236; quiet KL s8 0.00035; 0.77 s/step; the job: 177.4 min spent, about 20.2 min of steps left; card peak 14.5 GB +[10682.0s] stage 8/8 s8_register step 2300/4000: train bpb s8 0.0239; quiet KL s8 0.00041; 0.77 s/step; the job: 178.0 min spent, about 19.6 min of steps left; card peak 14.5 GB +[10720.2s] stage 8/8 s8_register step 2350/4000: train bpb s8 0.0236; quiet KL s8 0.00038; 0.77 s/step; the job: 178.7 min spent, about 19.0 min of steps left; card peak 14.5 GB +[10758.1s] stage 8/8 s8_register step 2400/4000: train bpb s8 0.0236; quiet KL s8 0.00035; 0.77 s/step; the job: 179.3 min spent, about 18.5 min of steps left; card peak 14.5 GB +[10796.3s] stage 8/8 s8_register step 2450/4000: train bpb s8 0.0232; quiet KL s8 0.00044; 0.77 s/step; the job: 179.9 min spent, about 17.9 min of steps left; card peak 14.5 GB +[10834.3s] stage 8/8 s8_register step 2500/4000: train bpb s8 0.0238; quiet KL s8 0.00035; 0.77 s/step; the job: 180.6 min spent, about 17.3 min of steps left; card peak 14.5 GB +[10872.6s] stage 8/8 s8_register step 2550/4000: train bpb s8 0.0235; quiet KL s8 0.00039; 0.77 s/step; the job: 181.2 min spent, about 16.8 min of steps left; card peak 14.5 GB +[10910.5s] stage 8/8 s8_register step 2600/4000: train bpb s8 0.0236; quiet KL s8 0.00033; 0.77 s/step; the job: 181.8 min spent, about 16.2 min of steps left; card peak 14.5 GB +[10948.7s] stage 8/8 s8_register step 2650/4000: train bpb s8 0.0230; quiet KL s8 0.00037; 0.77 s/step; the job: 182.5 min spent, about 15.6 min of steps left; card peak 14.5 GB +[10986.6s] stage 8/8 s8_register step 2700/4000: train bpb s8 0.0229; quiet KL s8 0.00033; 0.77 s/step; the job: 183.1 min spent, about 15.0 min of steps left; card peak 14.5 GB +[11024.8s] stage 8/8 s8_register step 2750/4000: train bpb s8 0.0234; quiet KL s8 0.00033; 0.77 s/step; the job: 183.7 min spent, about 14.5 min of steps left; card peak 14.5 GB +[11062.8s] stage 8/8 s8_register step 2800/4000: train bpb s8 0.0231; quiet KL s8 0.00034; 0.77 s/step; the job: 184.4 min spent, about 13.9 min of steps left; card peak 14.5 GB +[11100.7s] stage 8/8 s8_register step 2850/4000: train bpb s8 0.0234; quiet KL s8 0.00035; 0.77 s/step; the job: 185.0 min spent, about 13.3 min of steps left; card peak 14.5 GB +[11138.9s] stage 8/8 s8_register step 2900/4000: train bpb s8 0.0230; quiet KL s8 0.00035; 0.77 s/step; the job: 185.6 min spent, about 12.7 min of steps left; card peak 14.5 GB +[11177.0s] stage 8/8 s8_register step 2950/4000: train bpb s8 0.0231; quiet KL s8 0.00034; 0.77 s/step; the job: 186.3 min spent, about 12.2 min of steps left; card peak 14.5 GB +[11215.2s] stage 8/8 s8_register step 3000/4000: train bpb s8 0.0230; quiet KL s8 0.00032; 0.77 s/step; the job: 186.9 min spent, about 11.6 min of steps left; card peak 14.5 GB +[11253.1s] stage 8/8 s8_register step 3050/4000: train bpb s8 0.0230; quiet KL s8 0.00035; 0.76 s/step; the job: 187.6 min spent, about 11.0 min of steps left; card peak 14.5 GB +[11291.6s] stage 8/8 s8_register step 3100/4000: train bpb s8 0.0228; quiet KL s8 0.00036; 0.77 s/step; the job: 188.2 min spent, about 10.4 min of steps left; card peak 14.5 GB +[11329.9s] stage 8/8 s8_register step 3150/4000: train bpb s8 0.0235; quiet KL s8 0.00035; 0.77 s/step; the job: 188.8 min spent, about 9.9 min of steps left; card peak 14.5 GB +[11368.0s] stage 8/8 s8_register step 3200/4000: train bpb s8 0.0232; quiet KL s8 0.00036; 0.77 s/step; the job: 189.5 min spent, about 9.3 min of steps left; card peak 14.5 GB +[11405.8s] stage 8/8 s8_register step 3250/4000: train bpb s8 0.0226; quiet KL s8 0.00036; 0.76 s/step; the job: 190.1 min spent, about 8.7 min of steps left; card peak 14.5 GB +[11443.8s] stage 8/8 s8_register step 3300/4000: train bpb s8 0.0227; quiet KL s8 0.00034; 0.76 s/step; the job: 190.7 min spent, about 8.1 min of steps left; card peak 14.5 GB +[11481.8s] stage 8/8 s8_register step 3350/4000: train bpb s8 0.0230; quiet KL s8 0.00034; 0.76 s/step; the job: 191.4 min spent, about 7.5 min of steps left; card peak 14.5 GB +[11520.0s] stage 8/8 s8_register step 3400/4000: train bpb s8 0.0231; quiet KL s8 0.00033; 0.76 s/step; the job: 192.0 min spent, about 7.0 min of steps left; card peak 14.5 GB +[11557.9s] stage 8/8 s8_register step 3450/4000: train bpb s8 0.0225; quiet KL s8 0.00037; 0.76 s/step; the job: 192.6 min spent, about 6.4 min of steps left; card peak 14.5 GB +[11596.1s] stage 8/8 s8_register step 3500/4000: train bpb s8 0.0229; quiet KL s8 0.00032; 0.76 s/step; the job: 193.3 min spent, about 5.8 min of steps left; card peak 14.5 GB +[11634.0s] stage 8/8 s8_register step 3550/4000: train bpb s8 0.0228; quiet KL s8 0.00032; 0.76 s/step; the job: 193.9 min spent, about 5.2 min of steps left; card peak 14.5 GB +[11672.3s] stage 8/8 s8_register step 3600/4000: train bpb s8 0.0223; quiet KL s8 0.00032; 0.76 s/step; the job: 194.5 min spent, about 4.7 min of steps left; card peak 14.5 GB +[11710.5s] stage 8/8 s8_register step 3650/4000: train bpb s8 0.0220; quiet KL s8 0.00032; 0.76 s/step; the job: 195.2 min spent, about 4.1 min of steps left; card peak 14.5 GB +[11749.0s] stage 8/8 s8_register step 3700/4000: train bpb s8 0.0222; quiet KL s8 0.00033; 0.76 s/step; the job: 195.8 min spent, about 3.5 min of steps left; card peak 14.5 GB +[11787.0s] stage 8/8 s8_register step 3750/4000: train bpb s8 0.0226; quiet KL s8 0.00031; 0.76 s/step; the job: 196.5 min spent, about 2.9 min of steps left; card peak 14.5 GB +[11825.0s] stage 8/8 s8_register step 3800/4000: train bpb s8 0.0221; quiet KL s8 0.00035; 0.76 s/step; the job: 197.1 min spent, about 2.3 min of steps left; card peak 14.5 GB +[11863.2s] stage 8/8 s8_register step 3850/4000: train bpb s8 0.0221; quiet KL s8 0.00031; 0.76 s/step; the job: 197.7 min spent, about 1.7 min of steps left; card peak 14.5 GB +[11901.2s] stage 8/8 s8_register step 3900/4000: train bpb s8 0.0225; quiet KL s8 0.00030; 0.76 s/step; the job: 198.4 min spent, about 1.2 min of steps left; card peak 14.5 GB +[11939.4s] stage 8/8 s8_register step 3950/4000: train bpb s8 0.0228; quiet KL s8 0.00029; 0.76 s/step; the job: 199.0 min spent, about 0.6 min of steps left; card peak 14.5 GB +[11977.4s] stage 8/8 s8_register step 4000/4000: train bpb s8 0.0226; quiet KL s8 0.00031; 0.76 s/step; the job: 199.6 min spent, about 0.0 min of steps left; card peak 14.5 GB +[11978.7s] anchors saved: /root/refit/group/anchors/gXA_close8_.safetensors for ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail', 's7_mixed', 's8_register'] +[12220.2s] CLOSE 8/8: the group ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail', 's7_mixed', 's8_register'] all on, against arms off: +[12220.2s] s1_perspective rows: off 0.5538 -> all on 0.0492 (-0.5046 +- 0.0048); with s1_perspective masked 0.4324 (its share -0.3832); s1_perspective alone 0.0525 (-0.5013); gate mean 0.210 / widest 0.571 +[12220.2s] s2_concept rows: off 0.7212 -> all on 0.1105 (-0.6107 +- 0.0038); with s2_concept masked 0.2307 (its share -0.1202); s2_concept alone 0.3503 (-0.3709); gate mean 0.277 / widest 0.498 +[12220.2s] s3_rules rows: off 0.9296 -> all on 0.2237 (-0.7059 +- 0.0053); with s3_rules masked 0.3137 (its share -0.0900); s3_rules alone 0.8236 (-0.1060); gate mean 0.120 / widest 0.174 +[12220.2s] s4_arith rows: off 1.2339 -> all on 0.3279 (-0.9059 +- 0.0062); with s4_arith masked 0.4711 (its share -0.1431); s4_arith alone 1.1267 (-0.1072); gate mean 0.108 / widest 0.159 +[12220.2s] s5_causal rows: off 0.6604 -> all on 0.0431 (-0.6174 +- 0.0069); with s5_causal masked 0.4581 (its share -0.4150); s5_causal alone 0.0911 (-0.5693); gate mean 0.185 / widest 0.368 +[12220.2s] s6_tryfail rows: off 0.6728 -> all on 0.0548 (-0.6180 +- 0.0037); with s6_tryfail masked 0.4686 (its share -0.4138); s6_tryfail alone 0.1638 (-0.5090); gate mean 0.230 / widest 0.537 +[12220.2s] s7_mixed rows: off 0.7830 -> all on 0.7200 (-0.0630 +- 0.0390); with s7_mixed masked 0.7768 (its share -0.0568); s7_mixed alone 0.7242 (-0.0588); gate mean 0.251 / widest 0.464 +[12220.2s] s8_register rows: off 0.4907 -> all on 0.0212 (-0.4695 +- 0.0085); with s8_register masked 0.3757 (its share -0.3546); s8_register alone 0.0678 (-0.4228); gate mean 0.203 / widest 0.452 +[12220.2s] web holdout: off 0.9506 -> all on 0.9525 (+0.00189 +- 0.00032; the quiet bar +.012); each member's share: s1_perspective +0.00025, s2_concept -0.00043, s3_rules +0.00009, s4_arith +0.00020, s5_causal +0.00035, s6_tryfail -0.00025, s7_mixed -0.00096, s8_register +0.00045; alone: s1_perspective +0.00143, s2_concept +0.00105, s3_rules +0.00053, s4_arith +0.00011, s5_causal +0.00028, s6_tryfail +0.00013, s7_mixed -0.00019, s8_register +0.00048 +[12220.2s] a member's share on a PARTNER's rows (all on minus that member masked; near 0 = it stays out): s1_perspective on s2_concept -0.2174; s1_perspective on s3_rules -0.1832; s1_perspective on s4_arith -0.1280; s1_perspective on s5_causal -0.0128; s1_perspective on s6_tryfail -0.0280; s1_perspective on s7_mixed -0.0029; s1_perspective on s8_register -0.0272; s2_concept on s1_perspective -0.0057; s2_concept on s3_rules -0.1086; s2_concept on s4_arith -0.1060; s2_concept on s5_causal -0.0013; s2_concept on s6_tryfail -0.0035; s2_concept on s7_mixed -0.0008; s2_concept on s8_register -0.0007; s3_rules on s1_perspective +0.0001; s3_rules on s2_concept +0.0005; s3_rules on s4_arith -0.0002; s3_rules on s5_causal -0.0001; s3_rules on s6_tryfail -0.0003; s3_rules on s7_mixed +0.0001; s3_rules on s8_register -0.0001; s4_arith on s1_perspective +0.0000; s4_arith on s2_concept -0.0006; s4_arith on s3_rules -0.0001; s4_arith on s5_causal -0.0000; s4_arith on s6_tryfail -0.0036; s4_arith on s7_mixed +0.0001; s4_arith on s8_register -0.0000; s5_causal on s1_perspective +0.0007; s5_causal on s2_concept -0.0005; s5_causal on s3_rules +0.0005; s5_causal on s4_arith +0.0002; s5_causal on s6_tryfail -0.0005; s5_causal on s7_mixed +0.0004; s5_causal on s8_register -0.0003; s6_tryfail on s1_perspective -0.0001; s6_tryfail on s2_concept +0.0002; s6_tryfail on s3_rules -0.0002; s6_tryfail on s4_arith -0.0009; s6_tryfail on s5_causal -0.0001; s6_tryfail on s7_mixed +0.0001; s6_tryfail on s8_register -0.0006; s7_mixed on s1_perspective -0.0004; s7_mixed on s2_concept -0.0005; s7_mixed on s3_rules -0.0010; s7_mixed on s4_arith -0.0103; s7_mixed on s5_causal -0.0002; s7_mixed on s6_tryfail -0.0004; s7_mixed on s8_register -0.0001; s8_register on s1_perspective +0.0007; s8_register on s2_concept -0.0008; s8_register on s3_rules +0.0005; s8_register on s4_arith +0.0001; s8_register on s5_causal +0.0002; s8_register on s6_tryfail +0.0002; s8_register on s7_mixed +0.0012 +[12220.2s] a member ALONE on a partner's rows (alone minus off): s1_perspective on s2_concept -0.4817; s1_perspective on s3_rules -0.4209; s1_perspective on s4_arith -0.5113; s1_perspective on s5_causal -0.1978; s1_perspective on s6_tryfail -0.1782; s1_perspective on s7_mixed -0.0101; s1_perspective on s8_register -0.1119; s2_concept on s1_perspective -0.0273; s2_concept on s3_rules -0.3108; s2_concept on s4_arith -0.3537; s2_concept on s5_causal -0.0015; s2_concept on s6_tryfail +0.0004; s2_concept on s7_mixed -0.0000; s2_concept on s8_register -0.0215; s3_rules on s1_perspective -0.0003; s3_rules on s2_concept +0.0031; s3_rules on s4_arith +0.0001; s3_rules on s5_causal +0.0071; s3_rules on s6_tryfail +0.0010; s3_rules on s7_mixed -0.0005; s3_rules on s8_register +0.0002; s4_arith on s1_perspective -0.0009; s4_arith on s2_concept -0.0004; s4_arith on s3_rules -0.0016; s4_arith on s5_causal -0.0033; s4_arith on s6_tryfail -0.0135; s4_arith on s7_mixed -0.0002; s4_arith on s8_register -0.0015; s5_causal on s1_perspective -0.0183; s5_causal on s2_concept -0.0109; s5_causal on s3_rules -0.0122; s5_causal on s4_arith -0.0091; s5_causal on s6_tryfail -0.0065; s5_causal on s7_mixed -0.0011; s5_causal on s8_register -0.0064; s6_tryfail on s1_perspective -0.0020; s6_tryfail on s2_concept +0.0012; s6_tryfail on s3_rules +0.0006; s6_tryfail on s4_arith +0.0036; s6_tryfail on s5_causal -0.0098; s6_tryfail on s7_mixed -0.0000; s6_tryfail on s8_register -0.0003; s7_mixed on s1_perspective -0.0568; s7_mixed on s2_concept -0.0058; s7_mixed on s3_rules -0.0617; s7_mixed on s4_arith -0.0385; s7_mixed on s5_causal -0.1028; s7_mixed on s6_tryfail -0.0802; s7_mixed on s8_register -0.0080; s8_register on s1_perspective -0.0172; s8_register on s2_concept -0.0079; s8_register on s3_rules -0.0048; s8_register on s4_arith -0.0065; s8_register on s5_causal -0.0189; s8_register on s6_tryfail -0.0220; s8_register on s7_mixed -0.0002 +[12220.2s] s1_perspective battery accuracy, off -> all on: in 0.550 -> 0.887, transfer 0.463 -> 0.438; alone: in 0.838, transfer 0.463 +[12220.2s] s2_concept battery accuracy, off -> all on: in 0.298 -> 0.976, transfer 0.678 -> 0.720; alone: in 0.964, transfer 0.720 +[12220.2s] s3_rules battery accuracy, off -> all on: in 0.155 -> 0.855, transfer 0.070 -> 0.310, held-out 0.050 -> 0.433; alone: in 0.320, transfer 0.080, held-out 0.108 +[12220.2s] s4_arith battery accuracy, off -> all on: in 0.344 -> 0.644, transfer 0.242 -> 0.242, held-out 0.350 -> 0.425; alone: in 0.381, transfer 0.246, held-out 0.388 +[12220.2s] probes mean: off 0.411 -> all on 0.422 +[12220.2s] compiled chain: 28 cache entries (limit 256), 0 limit warnings; card peak 14.5 GB; host RSS 11.6 GB +[arms] adapter chain compile OFF on 256 wrapped blocks (8 arms) +[12223.7s] the write of s1_perspective (mean |delta| / |x|, all members live): own rows 1.63e-02; web 6.65e-03 (own over web 2.44x); partners' rows s2_concept 1.27e-02, s3_rules 1.43e-02, s4_arith 1.51e-02, s5_causal 1.39e-02, s6_tryfail 1.39e-02, s7_mixed 6.62e-03, s8_register 1.28e-02 +[12223.7s] the write of s2_concept (mean |delta| / |x|, all members live): own rows 1.14e-02; web 4.36e-03 (own over web 2.61x); partners' rows s1_perspective 8.64e-03, s3_rules 1.11e-02, s4_arith 1.17e-02, s5_causal 8.99e-03, s6_tryfail 9.33e-03, s7_mixed 4.44e-03, s8_register 8.74e-03 +[12223.7s] the write of s3_rules (mean |delta| / |x|, all members live): own rows 8.02e-03; web 1.41e-03 (own over web 5.69x); partners' rows s1_perspective 1.77e-03, s2_concept 1.77e-03, s4_arith 1.33e-03, s5_causal 1.98e-03, s6_tryfail 1.81e-03, s7_mixed 1.40e-03, s8_register 1.72e-03 +[12223.7s] the write of s4_arith (mean |delta| / |x|, all members live): own rows 7.78e-03; web 1.06e-03 (own over web 7.37x); partners' rows s1_perspective 1.50e-03, s2_concept 1.30e-03, s3_rules 1.41e-03, s5_causal 1.95e-03, s6_tryfail 3.49e-03, s7_mixed 1.07e-03, s8_register 1.60e-03 +[12223.7s] the write of s5_causal (mean |delta| / |x|, all members live): own rows 1.62e-02; web 3.10e-03 (own over web 5.23x); partners' rows s1_perspective 5.62e-03, s2_concept 3.61e-03, s3_rules 4.00e-03, s4_arith 3.83e-03, s6_tryfail 4.88e-03, s7_mixed 3.13e-03, s8_register 5.08e-03 +[12223.7s] the write of s6_tryfail (mean |delta| / |x|, all members live): own rows 1.91e-02; web 4.97e-03 (own over web 3.84x); partners' rows s1_perspective 6.86e-03, s2_concept 5.46e-03, s3_rules 6.16e-03, s4_arith 7.01e-03, s5_causal 7.47e-03, s7_mixed 4.95e-03, s8_register 7.23e-03 +[12223.7s] the write of s7_mixed (mean |delta| / |x|, all members live): own rows 4.20e-03; web 4.29e-03 (own over web 0.98x); partners' rows s1_perspective 7.05e-03, s2_concept 5.80e-03, s3_rules 8.15e-03, s4_arith 6.69e-03, s5_causal 6.33e-03, s6_tryfail 6.51e-03, s8_register 5.64e-03 +[12223.7s] the write of s8_register (mean |delta| / |x|, all members live): own rows 1.58e-02; web 3.65e-03 (own over web 4.35x); partners' rows s1_perspective 5.72e-03, s2_concept 4.29e-03, s3_rules 4.27e-03, s4_arith 4.54e-03, s5_causal 6.14e-03, s6_tryfail 5.95e-03, s7_mixed 3.60e-03 +[12223.8s] DONE gXA: 5 closes; 12224 s; card peak 14.5 GB; host RSS 11.6 GB; compiled chain: 28 cache entries (limit 256), 0 limit warnings diff --git a/mini-beatrix-3/arm_refit/logs/group_XB.log b/mini-beatrix-3/arm_refit/logs/group_XB.log new file mode 100644 index 0000000000000000000000000000000000000000..9313262dd87644cc1dce96fb8c00f5664cdf6af5 --- /dev/null +++ b/mini-beatrix-3/arm_refit/logs/group_XB.log @@ -0,0 +1,454 @@ +[ 0.2s] THE JOB: gXB: the group ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail', 's7_mixed', 's8_register'], staged, EARLIER ARMS FROZEN (on, not training); 4000 steps per stage; each step = one task chunk + one abstention chunk of 2 x 4096 bytes per newest member; lambda 2.0; pure Adam lr 0.001; the compiled chain with recompute; trunk step 245,674; card 31.4 GB with 16.4 GB free; host RSS 0.6 GB; the first 4 members loaded from gRB:4 (the later members attach fresh); a member's partner rows come from every other member of the roster, attached or not +Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads. +[ 15.8s] mini-beatrix-3 @ 245,674: trunk loaded from /root/v3/runs/mission/mini-beatrix-3/checkpoints/step_00245674.safetensors and frozen (376.0M); amoe-lora 0.2.11, torch 2.8.0+cu128 +[stream] opening 'fineweb-good' (HuggingFaceFW/fineweb-edu) … +[stream] opening 'simple-wiki' (wikimedia/wikipedia) … +[stream] opening 'babi-prose' (Muennighoff/babi) … +Repo card metadata block was not found. Setting CardData to empty. +[stream] opening 'tinystories' (roneneldan/TinyStories) … +[stream] opening 'proofwriter-prose' (tasksource/proofwriter) … +[ 79.5s] ARMS OFF (the bare trunk): s1_perspective rows 0.5538; s2_concept rows 0.7212; s3_rules rows 0.9296; s4_arith rows 1.2339; s5_causal rows 0.6604; s6_tryfail rows 0.6728; s7_mixed rows 0.7830; s8_register rows 0.4907; web rows 0.9506; s1_perspective battery in 0.550, transfer 0.463; s2_concept battery in 0.298, transfer 0.678; s3_rules battery in 0.155, transfer 0.070, held-out 0.050; s4_arith battery in 0.344, transfer 0.242, held-out 0.350; probes mean 0.411 +[arms] attached 's1_perspective' for phase 'curriculum_s1': 13.69M params x 32 blocks, lambda 2.0, spec {'n_slots': 16, 'K': 16, 'D': 8, 'hidden': 256}, recompute on, compile on +[arms] attached 's2_concept' for phase 'curriculum_s2': 13.69M params x 32 blocks, lambda 2.0, spec {'n_slots': 16, 'K': 16, 'D': 8, 'hidden': 256}, recompute on, compile on +[arms] attached 's3_rules' for phase 'curriculum_s3': 13.69M params x 32 blocks, lambda 2.0, spec {'n_slots': 16, 'K': 16, 'D': 8, 'hidden': 256}, recompute on, compile on +[arms] attached 's4_arith' for phase 'curriculum_s4': 13.69M params x 32 blocks, lambda 2.0, spec {'n_slots': 16, 'K': 16, 'D': 8, 'hidden': 256}, recompute on, compile on +[ 84.2s] RESUMED from gRB at its close 4: ['s1_perspective', 's2_concept', 's3_rules', 's4_arith'] loaded (on; frozen while the later arms train); the next stage attaches over them +[ 221.9s] CLOSE 4/8: the group ['s1_perspective', 's2_concept', 's3_rules', 's4_arith'] all on, against arms off: +[ 221.9s] s1_perspective rows: off 0.5538 -> all on 0.0479 (-0.5060 +- 0.0048); with s1_perspective masked 0.5211 (its share -0.4733); s1_perspective alone 0.0522 (-0.5017); gate mean 0.212 / widest 0.555 +[ 221.9s] s2_concept rows: off 0.7212 -> all on 0.1212 (-0.6000 +- 0.0037); with s2_concept masked 0.2424 (its share -0.1212); s2_concept alone 0.3603 (-0.3608); gate mean 0.273 / widest 0.487 +[ 221.9s] s3_rules rows: off 0.9296 -> all on 0.2222 (-0.7074 +- 0.0054); with s3_rules masked 0.3181 (its share -0.0958); s3_rules alone 0.8702 (-0.0594); gate mean 0.124 / widest 0.175 +[ 221.9s] s4_arith rows: off 1.2339 -> all on 0.3379 (-0.8960 +- 0.0059); with s4_arith masked 0.5116 (its share -0.1737); s4_arith alone 1.1550 (-0.0789); gate mean 0.108 / widest 0.167 +[ 221.9s] web holdout: off 0.9506 -> all on 0.9517 (+0.00110 +- 0.00039; the quiet bar +.012); each member's share: s1_perspective +0.00004, s2_concept -0.00055, s3_rules +0.00038, s4_arith -0.00001; alone: s1_perspective +0.00134, s2_concept +0.00086, s3_rules +0.00034, s4_arith +0.00006 +[ 221.9s] a member's share on a PARTNER's rows (all on minus that member masked; near 0 = it stays out): s1_perspective on s2_concept -0.2380; s1_perspective on s3_rules -0.1951; s1_perspective on s4_arith -0.1591; s2_concept on s1_perspective -0.0041; s2_concept on s3_rules -0.1005; s2_concept on s4_arith -0.1138; s3_rules on s1_perspective +0.0001; s3_rules on s2_concept -0.0006; s3_rules on s4_arith -0.0003; s4_arith on s1_perspective +0.0000; s4_arith on s2_concept -0.0000; s4_arith on s3_rules +0.0000 +[ 221.9s] a member ALONE on a partner's rows (alone minus off): s1_perspective on s2_concept -0.4772; s1_perspective on s3_rules -0.4262; s1_perspective on s4_arith -0.4807; s2_concept on s1_perspective -0.0300; s2_concept on s3_rules -0.3138; s2_concept on s4_arith -0.3577; s3_rules on s1_perspective -0.0003; s3_rules on s2_concept +0.0016; s3_rules on s4_arith -0.0047; s4_arith on s1_perspective -0.0013; s4_arith on s2_concept -0.0015; s4_arith on s3_rules +0.0005 +[ 221.9s] the group on the rows of stages not yet attached: s5_causal -0.1748; s6_tryfail -0.1850; s7_mixed -0.0060; s8_register -0.1104 +[ 221.9s] s1_perspective battery accuracy, off -> all on: in 0.550 -> 1.000, transfer 0.463 -> 0.463; alone: in 0.925, transfer 0.463 +[ 221.9s] s2_concept battery accuracy, off -> all on: in 0.298 -> 0.952, transfer 0.678 -> 0.754; alone: in 0.905, transfer 0.703 +[ 221.9s] s3_rules battery accuracy, off -> all on: in 0.155 -> 0.885, transfer 0.070 -> 0.330, held-out 0.050 -> 0.483; alone: in 0.335, transfer 0.100, held-out 0.142 +[ 221.9s] s4_arith battery accuracy, off -> all on: in 0.344 -> 0.594, transfer 0.242 -> 0.271, held-out 0.350 -> 0.287; alone: in 0.412, transfer 0.237, held-out 0.400 +[ 221.9s] probes mean: off 0.411 -> all on 0.415 +[ 221.9s] compiled chain: 8 cache entries (limit 256), 0 limit warnings; card peak 3.0 GB; host RSS 4.6 GB +[arms] attached 's5_causal' for phase 'curriculum_s5': 13.69M params x 32 blocks, lambda 2.0, spec {'n_slots': 16, 'K': 16, 'D': 8, 'hidden': 256}, recompute on, compile on +[ 225.4s] STAGE 5/8: s5_causal attached over ['s1_perspective', 's2_concept', 's3_rules', 's4_arith'] (on, frozen); at birth the group reads exactly as before on its rows and on web text; 4000 steps of 1 task + 1 abstention chunks +[stream] opening 'fineweb-edu' (HuggingFaceFW/fineweb-edu) … +[ 235.3s] stage 5/8 s5_causal step 1/4000: train bpb s5 0.5003; quiet KL s5 0.00014; 9.95 s/step; the job: 3.9 min spent, about 2653.7 min of steps left; card peak 13.9 GB +[ 268.5s] stage 5/8 s5_causal step 50/4000: train bpb s5 0.4748; quiet KL s5 0.00021; 0.86 s/step; the job: 4.5 min spent, about 229.4 min of steps left; card peak 14.0 GB +[ 300.7s] stage 5/8 s5_causal step 100/4000: train bpb s5 0.4433; quiet KL s5 0.00058; 0.75 s/step; the job: 5.0 min spent, about 199.7 min of steps left; card peak 14.0 GB +[ 332.8s] stage 5/8 s5_causal step 150/4000: train bpb s5 0.4122; quiet KL s5 0.00060; 0.72 s/step; the job: 5.5 min spent, about 189.2 min of steps left; card peak 14.0 GB +[ 365.2s] stage 5/8 s5_causal step 200/4000: train bpb s5 0.3751; quiet KL s5 0.00061; 0.70 s/step; the job: 6.1 min spent, about 184.1 min of steps left; card peak 14.0 GB +[ 397.3s] stage 5/8 s5_causal step 250/4000: train bpb s5 0.3346; quiet KL s5 0.00069; 0.69 s/step; the job: 6.6 min spent, about 180.5 min of steps left; card peak 14.0 GB +[ 429.3s] stage 5/8 s5_causal step 300/4000: train bpb s5 0.2950; quiet KL s5 0.00084; 0.68 s/step; the job: 7.2 min spent, about 177.9 min of steps left; card peak 14.0 GB +[ 461.5s] stage 5/8 s5_causal step 350/4000: train bpb s5 0.2538; quiet KL s5 0.00090; 0.67 s/step; the job: 7.7 min spent, about 176.0 min of steps left; card peak 14.0 GB +[ 493.5s] stage 5/8 s5_causal step 400/4000: train bpb s5 0.2116; quiet KL s5 0.00085; 0.67 s/step; the job: 8.2 min spent, about 174.3 min of steps left; card peak 14.0 GB +[ 525.5s] stage 5/8 s5_causal step 450/4000: train bpb s5 0.1766; quiet KL s5 0.00090; 0.67 s/step; the job: 8.8 min spent, about 172.9 min of steps left; card peak 14.0 GB +[ 557.6s] stage 5/8 s5_causal step 500/4000: train bpb s5 0.1449; quiet KL s5 0.00088; 0.66 s/step; the job: 9.3 min spent, about 171.6 min of steps left; card peak 14.0 GB +[ 589.6s] stage 5/8 s5_causal step 550/4000: train bpb s5 0.1175; quiet KL s5 0.00088; 0.66 s/step; the job: 9.8 min spent, about 170.5 min of steps left; card peak 14.0 GB +[ 621.6s] stage 5/8 s5_causal step 600/4000: train bpb s5 0.0999; quiet KL s5 0.00083; 0.66 s/step; the job: 10.4 min spent, about 169.5 min of steps left; card peak 14.0 GB +[ 653.8s] stage 5/8 s5_causal step 650/4000: train bpb s5 0.0838; quiet KL s5 0.00082; 0.66 s/step; the job: 10.9 min spent, about 168.6 min of steps left; card peak 14.0 GB +[ 685.8s] stage 5/8 s5_causal step 700/4000: train bpb s5 0.0732; quiet KL s5 0.00075; 0.66 s/step; the job: 11.4 min spent, about 167.7 min of steps left; card peak 14.0 GB +[ 717.9s] stage 5/8 s5_causal step 750/4000: train bpb s5 0.0662; quiet KL s5 0.00073; 0.66 s/step; the job: 12.0 min spent, about 166.9 min of steps left; card peak 14.0 GB +[ 750.2s] stage 5/8 s5_causal step 800/4000: train bpb s5 0.0610; quiet KL s5 0.00070; 0.66 s/step; the job: 12.5 min spent, about 166.2 min of steps left; card peak 14.0 GB +[ 782.4s] stage 5/8 s5_causal step 850/4000: train bpb s5 0.0570; quiet KL s5 0.00063; 0.66 s/step; the job: 13.0 min spent, about 165.5 min of steps left; card peak 14.0 GB +[ 814.2s] stage 5/8 s5_causal step 900/4000: train bpb s5 0.0549; quiet KL s5 0.00067; 0.65 s/step; the job: 13.6 min spent, about 164.7 min of steps left; card peak 14.0 GB +[ 846.1s] stage 5/8 s5_causal step 950/4000: train bpb s5 0.0533; quiet KL s5 0.00061; 0.65 s/step; the job: 14.1 min spent, about 163.9 min of steps left; card peak 14.0 GB +[ 878.2s] stage 5/8 s5_causal step 1000/4000: train bpb s5 0.0513; quiet KL s5 0.00056; 0.65 s/step; the job: 14.6 min spent, about 163.2 min of steps left; card peak 14.0 GB +[ 910.1s] stage 5/8 s5_causal step 1050/4000: train bpb s5 0.0503; quiet KL s5 0.00057; 0.65 s/step; the job: 15.2 min spent, about 162.5 min of steps left; card peak 14.0 GB +[ 942.2s] stage 5/8 s5_causal step 1100/4000: train bpb s5 0.0491; quiet KL s5 0.00051; 0.65 s/step; the job: 15.7 min spent, about 161.8 min of steps left; card peak 14.0 GB +[ 974.6s] stage 5/8 s5_causal step 1150/4000: train bpb s5 0.0489; quiet KL s5 0.00049; 0.65 s/step; the job: 16.2 min spent, about 161.2 min of steps left; card peak 14.0 GB +[ 1006.7s] stage 5/8 s5_causal step 1200/4000: train bpb s5 0.0481; quiet KL s5 0.00047; 0.65 s/step; the job: 16.8 min spent, about 160.6 min of steps left; card peak 14.0 GB +[ 1039.1s] stage 5/8 s5_causal step 1250/4000: train bpb s5 0.0479; quiet KL s5 0.00048; 0.65 s/step; the job: 17.3 min spent, about 160.0 min of steps left; card peak 14.0 GB +[ 1071.2s] stage 5/8 s5_causal step 1300/4000: train bpb s5 0.0477; quiet KL s5 0.00050; 0.65 s/step; the job: 17.9 min spent, about 159.4 min of steps left; card peak 14.0 GB +[ 1103.3s] stage 5/8 s5_causal step 1350/4000: train bpb s5 0.0472; quiet KL s5 0.00046; 0.65 s/step; the job: 18.4 min spent, about 158.8 min of steps left; card peak 14.0 GB +[ 1135.4s] stage 5/8 s5_causal step 1400/4000: train bpb s5 0.0468; quiet KL s5 0.00044; 0.65 s/step; the job: 18.9 min spent, about 158.2 min of steps left; card peak 14.0 GB +[ 1167.5s] stage 5/8 s5_causal step 1450/4000: train bpb s5 0.0464; quiet KL s5 0.00043; 0.65 s/step; the job: 19.5 min spent, about 157.6 min of steps left; card peak 14.0 GB +[ 1199.4s] stage 5/8 s5_causal step 1500/4000: train bpb s5 0.0459; quiet KL s5 0.00041; 0.65 s/step; the job: 20.0 min spent, about 156.9 min of steps left; card peak 14.0 GB +[ 1231.6s] stage 5/8 s5_causal step 1550/4000: train bpb s5 0.0457; quiet KL s5 0.00039; 0.65 s/step; the job: 20.5 min spent, about 156.3 min of steps left; card peak 14.0 GB +[ 1264.0s] stage 5/8 s5_causal step 1600/4000: train bpb s5 0.0456; quiet KL s5 0.00038; 0.65 s/step; the job: 21.1 min spent, about 155.8 min of steps left; card peak 14.0 GB +[ 1296.0s] stage 5/8 s5_causal step 1650/4000: train bpb s5 0.0453; quiet KL s5 0.00039; 0.65 s/step; the job: 21.6 min spent, about 155.2 min of steps left; card peak 14.0 GB +[ 1328.1s] stage 5/8 s5_causal step 1700/4000: train bpb s5 0.0453; quiet KL s5 0.00041; 0.65 s/step; the job: 22.1 min spent, about 154.6 min of steps left; card peak 14.0 GB +[ 1360.1s] stage 5/8 s5_causal step 1750/4000: train bpb s5 0.0453; quiet KL s5 0.00038; 0.65 s/step; the job: 22.7 min spent, about 154.0 min of steps left; card peak 14.0 GB +[ 1392.1s] stage 5/8 s5_causal step 1800/4000: train bpb s5 0.0449; quiet KL s5 0.00036; 0.65 s/step; the job: 23.2 min spent, about 153.4 min of steps left; card peak 14.0 GB +[ 1424.2s] stage 5/8 s5_causal step 1850/4000: train bpb s5 0.0452; quiet KL s5 0.00048; 0.65 s/step; the job: 23.7 min spent, about 152.8 min of steps left; card peak 14.0 GB +[ 1456.5s] stage 5/8 s5_causal step 1900/4000: train bpb s5 0.0447; quiet KL s5 0.00040; 0.65 s/step; the job: 24.3 min spent, about 152.3 min of steps left; card peak 14.0 GB +[ 1488.7s] stage 5/8 s5_causal step 1950/4000: train bpb s5 0.0444; quiet KL s5 0.00038; 0.65 s/step; the job: 24.8 min spent, about 151.7 min of steps left; card peak 14.0 GB +[ 1520.8s] stage 5/8 s5_causal step 2000/4000: train bpb s5 0.0448; quiet KL s5 0.00034; 0.65 s/step; the job: 25.3 min spent, about 151.1 min of steps left; card peak 14.0 GB +[ 1553.0s] stage 5/8 s5_causal step 2050/4000: train bpb s5 0.0442; quiet KL s5 0.00034; 0.65 s/step; the job: 25.9 min spent, about 150.6 min of steps left; card peak 14.0 GB +[ 1585.1s] stage 5/8 s5_causal step 2100/4000: train bpb s5 0.0445; quiet KL s5 0.00037; 0.65 s/step; the job: 26.4 min spent, about 150.0 min of steps left; card peak 14.0 GB +[ 1617.3s] stage 5/8 s5_causal step 2150/4000: train bpb s5 0.0441; quiet KL s5 0.00035; 0.65 s/step; the job: 27.0 min spent, about 149.4 min of steps left; card peak 14.0 GB +[ 1650.3s] stage 5/8 s5_causal step 2200/4000: train bpb s5 0.0439; quiet KL s5 0.00035; 0.65 s/step; the job: 27.5 min spent, about 149.0 min of steps left; card peak 14.0 GB +[ 1682.7s] stage 5/8 s5_causal step 2250/4000: train bpb s5 0.0449; quiet KL s5 0.00036; 0.65 s/step; the job: 28.0 min spent, about 148.4 min of steps left; card peak 14.0 GB +[ 1714.6s] stage 5/8 s5_causal step 2300/4000: train bpb s5 0.0445; quiet KL s5 0.00036; 0.65 s/step; the job: 28.6 min spent, about 147.8 min of steps left; card peak 14.0 GB +[ 1746.7s] stage 5/8 s5_causal step 2350/4000: train bpb s5 0.0440; quiet KL s5 0.00035; 0.65 s/step; the job: 29.1 min spent, about 147.3 min of steps left; card peak 14.0 GB +[ 1778.8s] stage 5/8 s5_causal step 2400/4000: train bpb s5 0.0441; quiet KL s5 0.00037; 0.65 s/step; the job: 29.6 min spent, about 146.7 min of steps left; card peak 14.0 GB +[ 1811.0s] stage 5/8 s5_causal step 2450/4000: train bpb s5 0.0438; quiet KL s5 0.00032; 0.65 s/step; the job: 30.2 min spent, about 146.2 min of steps left; card peak 14.0 GB +[ 1843.4s] stage 5/8 s5_causal step 2500/4000: train bpb s5 0.0439; quiet KL s5 0.00033; 0.65 s/step; the job: 30.7 min spent, about 145.6 min of steps left; card peak 14.0 GB +[ 1875.5s] stage 5/8 s5_causal step 2550/4000: train bpb s5 0.0437; quiet KL s5 0.00036; 0.65 s/step; the job: 31.3 min spent, about 145.1 min of steps left; card peak 14.0 GB +[ 1907.7s] stage 5/8 s5_causal step 2600/4000: train bpb s5 0.0438; quiet KL s5 0.00033; 0.65 s/step; the job: 31.8 min spent, about 144.5 min of steps left; card peak 14.0 GB +[ 1940.0s] stage 5/8 s5_causal step 2650/4000: train bpb s5 0.0440; quiet KL s5 0.00035; 0.65 s/step; the job: 32.3 min spent, about 144.0 min of steps left; card peak 14.0 GB +[ 1972.2s] stage 5/8 s5_causal step 2700/4000: train bpb s5 0.0438; quiet KL s5 0.00037; 0.65 s/step; the job: 32.9 min spent, about 143.4 min of steps left; card peak 14.0 GB +[ 2004.5s] stage 5/8 s5_causal step 2750/4000: train bpb s5 0.0435; quiet KL s5 0.00033; 0.65 s/step; the job: 33.4 min spent, about 142.9 min of steps left; card peak 14.0 GB +[ 2036.9s] stage 5/8 s5_causal step 2800/4000: train bpb s5 0.0435; quiet KL s5 0.00033; 0.65 s/step; the job: 33.9 min spent, about 142.3 min of steps left; card peak 14.0 GB +[ 2069.1s] stage 5/8 s5_causal step 2850/4000: train bpb s5 0.0436; quiet KL s5 0.00032; 0.65 s/step; the job: 34.5 min spent, about 141.8 min of steps left; card peak 14.0 GB +[ 2101.3s] stage 5/8 s5_causal step 2900/4000: train bpb s5 0.0433; quiet KL s5 0.00031; 0.65 s/step; the job: 35.0 min spent, about 141.2 min of steps left; card peak 14.0 GB +[ 2133.7s] stage 5/8 s5_causal step 2950/4000: train bpb s5 0.0433; quiet KL s5 0.00031; 0.65 s/step; the job: 35.6 min spent, about 140.7 min of steps left; card peak 14.0 GB +[ 2165.9s] stage 5/8 s5_causal step 3000/4000: train bpb s5 0.0433; quiet KL s5 0.00032; 0.65 s/step; the job: 36.1 min spent, about 140.2 min of steps left; card peak 14.0 GB +[ 2198.0s] stage 5/8 s5_causal step 3050/4000: train bpb s5 0.0432; quiet KL s5 0.00033; 0.65 s/step; the job: 36.6 min spent, about 139.6 min of steps left; card peak 14.0 GB +[ 2230.3s] stage 5/8 s5_causal step 3100/4000: train bpb s5 0.0437; quiet KL s5 0.00040; 0.65 s/step; the job: 37.2 min spent, about 139.1 min of steps left; card peak 14.0 GB +[ 2262.5s] stage 5/8 s5_causal step 3150/4000: train bpb s5 0.0436; quiet KL s5 0.00035; 0.65 s/step; the job: 37.7 min spent, about 138.5 min of steps left; card peak 14.0 GB +[ 2294.8s] stage 5/8 s5_causal step 3200/4000: train bpb s5 0.0433; quiet KL s5 0.00034; 0.65 s/step; the job: 38.2 min spent, about 138.0 min of steps left; card peak 14.0 GB +[ 2327.2s] stage 5/8 s5_causal step 3250/4000: train bpb s5 0.0430; quiet KL s5 0.00032; 0.65 s/step; the job: 38.8 min spent, about 137.4 min of steps left; card peak 14.0 GB +[ 2359.5s] stage 5/8 s5_causal step 3300/4000: train bpb s5 0.0429; quiet KL s5 0.00032; 0.65 s/step; the job: 39.3 min spent, about 136.9 min of steps left; card peak 14.0 GB +[ 2391.9s] stage 5/8 s5_causal step 3350/4000: train bpb s5 0.0428; quiet KL s5 0.00030; 0.65 s/step; the job: 39.9 min spent, about 136.4 min of steps left; card peak 14.0 GB +[ 2424.3s] stage 5/8 s5_causal step 3400/4000: train bpb s5 0.0432; quiet KL s5 0.00031; 0.65 s/step; the job: 40.4 min spent, about 135.8 min of steps left; card peak 14.0 GB +[ 2456.6s] stage 5/8 s5_causal step 3450/4000: train bpb s5 0.0428; quiet KL s5 0.00028; 0.65 s/step; the job: 40.9 min spent, about 135.3 min of steps left; card peak 14.0 GB +[ 2488.8s] stage 5/8 s5_causal step 3500/4000: train bpb s5 0.0428; quiet KL s5 0.00029; 0.65 s/step; the job: 41.5 min spent, about 134.7 min of steps left; card peak 14.0 GB +[ 2521.1s] stage 5/8 s5_causal step 3550/4000: train bpb s5 0.0429; quiet KL s5 0.00033; 0.65 s/step; the job: 42.0 min spent, about 134.2 min of steps left; card peak 14.0 GB +[ 2553.1s] stage 5/8 s5_causal step 3600/4000: train bpb s5 0.0430; quiet KL s5 0.00033; 0.65 s/step; the job: 42.6 min spent, about 133.6 min of steps left; card peak 14.0 GB +[ 2585.4s] stage 5/8 s5_causal step 3650/4000: train bpb s5 0.0430; quiet KL s5 0.00032; 0.65 s/step; the job: 43.1 min spent, about 133.1 min of steps left; card peak 14.0 GB +[ 2617.8s] stage 5/8 s5_causal step 3700/4000: train bpb s5 0.0429; quiet KL s5 0.00033; 0.65 s/step; the job: 43.6 min spent, about 132.6 min of steps left; card peak 14.0 GB +[ 2649.8s] stage 5/8 s5_causal step 3750/4000: train bpb s5 0.0428; quiet KL s5 0.00030; 0.65 s/step; the job: 44.2 min spent, about 132.0 min of steps left; card peak 14.0 GB +[ 2682.0s] stage 5/8 s5_causal step 3800/4000: train bpb s5 0.0428; quiet KL s5 0.00030; 0.65 s/step; the job: 44.7 min spent, about 131.4 min of steps left; card peak 14.0 GB +[ 2714.3s] stage 5/8 s5_causal step 3850/4000: train bpb s5 0.0425; quiet KL s5 0.00030; 0.65 s/step; the job: 45.2 min spent, about 130.9 min of steps left; card peak 14.0 GB +[ 2746.4s] stage 5/8 s5_causal step 3900/4000: train bpb s5 0.0426; quiet KL s5 0.00032; 0.65 s/step; the job: 45.8 min spent, about 130.4 min of steps left; card peak 14.0 GB +[ 2778.7s] stage 5/8 s5_causal step 3950/4000: train bpb s5 0.0426; quiet KL s5 0.00031; 0.65 s/step; the job: 46.3 min spent, about 129.8 min of steps left; card peak 14.0 GB +[ 2811.1s] stage 5/8 s5_causal step 4000/4000: train bpb s5 0.0425; quiet KL s5 0.00030; 0.65 s/step; the job: 46.9 min spent, about 129.3 min of steps left; card peak 14.0 GB +[ 2811.9s] anchors saved: /root/refit/group/anchors/gXB_close5_.safetensors for ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal'] +[ 2974.8s] CLOSE 5/8: the group ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal'] all on, against arms off: +[ 2974.8s] s1_perspective rows: off 0.5538 -> all on 0.0479 (-0.5059 +- 0.0048); with s1_perspective masked 0.5187 (its share -0.4708); s1_perspective alone 0.0522 (-0.5017); gate mean 0.212 / widest 0.555 +[ 2974.8s] s2_concept rows: off 0.7212 -> all on 0.1212 (-0.6000 +- 0.0037); with s2_concept masked 0.2421 (its share -0.1209); s2_concept alone 0.3603 (-0.3608); gate mean 0.273 / widest 0.487 +[ 2974.8s] s3_rules rows: off 0.9296 -> all on 0.2226 (-0.7070 +- 0.0054); with s3_rules masked 0.3178 (its share -0.0952); s3_rules alone 0.8702 (-0.0594); gate mean 0.124 / widest 0.175 +[ 2974.8s] s4_arith rows: off 1.2339 -> all on 0.3379 (-0.8959 +- 0.0058); with s4_arith masked 0.5107 (its share -0.1728); s4_arith alone 1.1550 (-0.0789); gate mean 0.108 / widest 0.167 +[ 2974.8s] s5_causal rows: off 0.6604 -> all on 0.0426 (-0.6179 +- 0.0069); with s5_causal masked 0.4856 (its share -0.4430); s5_causal alone 0.1418 (-0.5187); gate mean 0.186 / widest 0.360 +[ 2974.8s] web holdout: off 0.9506 -> all on 0.9520 (+0.00138 +- 0.00042; the quiet bar +.012); each member's share: s1_perspective +0.00024, s2_concept -0.00036, s3_rules +0.00040, s4_arith +0.00021, s5_causal +0.00028; alone: s1_perspective +0.00134, s2_concept +0.00086, s3_rules +0.00034, s4_arith +0.00006, s5_causal +0.00000 +[ 2974.8s] a member's share on a PARTNER's rows (all on minus that member masked; near 0 = it stays out): s1_perspective on s2_concept -0.2370; s1_perspective on s3_rules -0.1968; s1_perspective on s4_arith -0.1597; s1_perspective on s5_causal -0.0428; s2_concept on s1_perspective -0.0042; s2_concept on s3_rules -0.1007; s2_concept on s4_arith -0.1135; s2_concept on s5_causal -0.0008; s3_rules on s1_perspective +0.0001; s3_rules on s2_concept -0.0006; s3_rules on s4_arith -0.0002; s3_rules on s5_causal -0.0001; s4_arith on s1_perspective +0.0001; s4_arith on s2_concept -0.0001; s4_arith on s3_rules +0.0002; s4_arith on s5_causal -0.0001; s5_causal on s1_perspective +0.0001; s5_causal on s2_concept +0.0000; s5_causal on s3_rules +0.0004; s5_causal on s4_arith +0.0001 +[ 2974.8s] a member ALONE on a partner's rows (alone minus off): s1_perspective on s2_concept -0.4772; s1_perspective on s3_rules -0.4262; s1_perspective on s4_arith -0.4807; s1_perspective on s5_causal -0.1919; s2_concept on s1_perspective -0.0300; s2_concept on s3_rules -0.3138; s2_concept on s4_arith -0.3577; s2_concept on s5_causal -0.0127; s3_rules on s1_perspective -0.0003; s3_rules on s2_concept +0.0016; s3_rules on s4_arith -0.0047; s3_rules on s5_causal +0.0060; s4_arith on s1_perspective -0.0013; s4_arith on s2_concept -0.0015; s4_arith on s3_rules +0.0005; s4_arith on s5_causal -0.0011; s5_causal on s1_perspective -0.0022; s5_causal on s2_concept -0.0010; s5_causal on s3_rules -0.0018; s5_causal on s4_arith -0.0022 +[ 2974.8s] the group on the rows of stages not yet attached: s6_tryfail -0.1851; s7_mixed -0.0062; s8_register -0.1113 +[ 2974.8s] s1_perspective battery accuracy, off -> all on: in 0.550 -> 1.000, transfer 0.463 -> 0.463; alone: in 0.925, transfer 0.463 +[ 2974.8s] s2_concept battery accuracy, off -> all on: in 0.298 -> 0.952, transfer 0.678 -> 0.780; alone: in 0.905, transfer 0.703 +[ 2974.8s] s3_rules battery accuracy, off -> all on: in 0.155 -> 0.880, transfer 0.070 -> 0.330, held-out 0.050 -> 0.475; alone: in 0.335, transfer 0.100, held-out 0.142 +[ 2974.8s] s4_arith battery accuracy, off -> all on: in 0.344 -> 0.594, transfer 0.242 -> 0.275, held-out 0.350 -> 0.287; alone: in 0.412, transfer 0.237, held-out 0.400 +[ 2974.8s] probes mean: off 0.411 -> all on 0.415 +[ 2974.8s] compiled chain: 13 cache entries (limit 256), 0 limit warnings; card peak 14.0 GB; host RSS 5.4 GB +[arms] attached 's6_tryfail' for phase 'curriculum_s6': 13.69M params x 32 blocks, lambda 2.0, spec {'n_slots': 16, 'K': 16, 'D': 8, 'hidden': 256}, recompute on, compile on +[ 2978.5s] STAGE 6/8: s6_tryfail attached over ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal'] (on, frozen); at birth the group reads exactly as before on its rows and on web text; 4000 steps of 1 task + 1 abstention chunks +[ 2980.1s] stage 6/8 s6_tryfail step 1/4000: train bpb s6 0.4916; quiet KL s6 0.00017; 1.60 s/step; the job: 49.7 min spent, about 129.3 min of steps left; card peak 14.0 GB +[ 3014.0s] stage 6/8 s6_tryfail step 50/4000: train bpb s6 0.4851; quiet KL s6 0.00020; 0.71 s/step; the job: 50.2 min spent, about 128.9 min of steps left; card peak 14.1 GB +[ 3048.4s] stage 6/8 s6_tryfail step 100/4000: train bpb s6 0.4561; quiet KL s6 0.00042; 0.70 s/step; the job: 50.8 min spent, about 128.5 min of steps left; card peak 14.1 GB +[ 3082.7s] stage 6/8 s6_tryfail step 150/4000: train bpb s6 0.4148; quiet KL s6 0.00052; 0.70 s/step; the job: 51.4 min spent, about 128.0 min of steps left; card peak 14.1 GB +[ 3117.4s] stage 6/8 s6_tryfail step 200/4000: train bpb s6 0.3716; quiet KL s6 0.00069; 0.69 s/step; the job: 52.0 min spent, about 127.6 min of steps left; card peak 14.1 GB +[ 3152.0s] stage 6/8 s6_tryfail step 250/4000: train bpb s6 0.3276; quiet KL s6 0.00079; 0.69 s/step; the job: 52.5 min spent, about 127.1 min of steps left; card peak 14.1 GB +[ 3186.3s] stage 6/8 s6_tryfail step 300/4000: train bpb s6 0.2863; quiet KL s6 0.00081; 0.69 s/step; the job: 53.1 min spent, about 126.7 min of steps left; card peak 14.1 GB +[ 3220.9s] stage 6/8 s6_tryfail step 350/4000: train bpb s6 0.2533; quiet KL s6 0.00088; 0.69 s/step; the job: 53.7 min spent, about 126.2 min of steps left; card peak 14.1 GB +[ 3255.6s] stage 6/8 s6_tryfail step 400/4000: train bpb s6 0.2208; quiet KL s6 0.00072; 0.69 s/step; the job: 54.3 min spent, about 125.8 min of steps left; card peak 14.1 GB +[ 3290.0s] stage 6/8 s6_tryfail step 450/4000: train bpb s6 0.1977; quiet KL s6 0.00064; 0.69 s/step; the job: 54.8 min spent, about 125.3 min of steps left; card peak 14.1 GB +[ 3324.5s] stage 6/8 s6_tryfail step 500/4000: train bpb s6 0.1729; quiet KL s6 0.00067; 0.69 s/step; the job: 55.4 min spent, about 124.9 min of steps left; card peak 14.1 GB +[ 3358.8s] stage 6/8 s6_tryfail step 550/4000: train bpb s6 0.1565; quiet KL s6 0.00076; 0.69 s/step; the job: 56.0 min spent, about 124.4 min of steps left; card peak 14.1 GB +[ 3393.1s] stage 6/8 s6_tryfail step 600/4000: train bpb s6 0.1378; quiet KL s6 0.00071; 0.69 s/step; the job: 56.6 min spent, about 123.9 min of steps left; card peak 14.1 GB +[ 3427.5s] stage 6/8 s6_tryfail step 650/4000: train bpb s6 0.1256; quiet KL s6 0.00067; 0.69 s/step; the job: 57.1 min spent, about 123.5 min of steps left; card peak 14.1 GB +[ 3462.1s] stage 6/8 s6_tryfail step 700/4000: train bpb s6 0.1160; quiet KL s6 0.00073; 0.69 s/step; the job: 57.7 min spent, about 123.0 min of steps left; card peak 14.1 GB +[ 3496.5s] stage 6/8 s6_tryfail step 750/4000: train bpb s6 0.1070; quiet KL s6 0.00074; 0.69 s/step; the job: 58.3 min spent, about 122.5 min of steps left; card peak 14.1 GB +[ 3531.0s] stage 6/8 s6_tryfail step 800/4000: train bpb s6 0.1000; quiet KL s6 0.00080; 0.69 s/step; the job: 58.8 min spent, about 122.0 min of steps left; card peak 14.1 GB +[ 3565.3s] stage 6/8 s6_tryfail step 850/4000: train bpb s6 0.0933; quiet KL s6 0.00076; 0.69 s/step; the job: 59.4 min spent, about 121.6 min of steps left; card peak 14.1 GB +[ 3599.9s] stage 6/8 s6_tryfail step 900/4000: train bpb s6 0.0891; quiet KL s6 0.00073; 0.69 s/step; the job: 60.0 min spent, about 121.1 min of steps left; card peak 14.1 GB +[ 3634.3s] stage 6/8 s6_tryfail step 950/4000: train bpb s6 0.0876; quiet KL s6 0.00060; 0.69 s/step; the job: 60.6 min spent, about 120.6 min of steps left; card peak 14.1 GB +[ 3669.0s] stage 6/8 s6_tryfail step 1000/4000: train bpb s6 0.0855; quiet KL s6 0.00060; 0.69 s/step; the job: 61.1 min spent, about 120.1 min of steps left; card peak 14.1 GB +[ 3703.5s] stage 6/8 s6_tryfail step 1050/4000: train bpb s6 0.0852; quiet KL s6 0.00064; 0.69 s/step; the job: 61.7 min spent, about 119.6 min of steps left; card peak 14.1 GB +[ 3738.0s] stage 6/8 s6_tryfail step 1100/4000: train bpb s6 0.0812; quiet KL s6 0.00059; 0.69 s/step; the job: 62.3 min spent, about 119.2 min of steps left; card peak 14.1 GB +[ 3772.5s] stage 6/8 s6_tryfail step 1150/4000: train bpb s6 0.0794; quiet KL s6 0.00053; 0.69 s/step; the job: 62.9 min spent, about 118.7 min of steps left; card peak 14.1 GB +[ 3807.1s] stage 6/8 s6_tryfail step 1200/4000: train bpb s6 0.0786; quiet KL s6 0.00060; 0.69 s/step; the job: 63.5 min spent, about 118.2 min of steps left; card peak 14.1 GB +[ 3841.5s] stage 6/8 s6_tryfail step 1250/4000: train bpb s6 0.0742; quiet KL s6 0.00050; 0.69 s/step; the job: 64.0 min spent, about 117.7 min of steps left; card peak 14.1 GB +[ 3876.0s] stage 6/8 s6_tryfail step 1300/4000: train bpb s6 0.0750; quiet KL s6 0.00054; 0.69 s/step; the job: 64.6 min spent, about 117.2 min of steps left; card peak 14.1 GB +[ 3910.6s] stage 6/8 s6_tryfail step 1350/4000: train bpb s6 0.0781; quiet KL s6 0.00061; 0.69 s/step; the job: 65.2 min spent, about 116.7 min of steps left; card peak 14.1 GB +[ 3945.0s] stage 6/8 s6_tryfail step 1400/4000: train bpb s6 0.0748; quiet KL s6 0.00054; 0.69 s/step; the job: 65.7 min spent, about 116.2 min of steps left; card peak 14.1 GB +[ 3979.5s] stage 6/8 s6_tryfail step 1450/4000: train bpb s6 0.0718; quiet KL s6 0.00052; 0.69 s/step; the job: 66.3 min spent, about 115.7 min of steps left; card peak 14.1 GB +[ 4014.0s] stage 6/8 s6_tryfail step 1500/4000: train bpb s6 0.0745; quiet KL s6 0.00051; 0.69 s/step; the job: 66.9 min spent, about 115.2 min of steps left; card peak 14.1 GB +[ 4048.5s] stage 6/8 s6_tryfail step 1550/4000: train bpb s6 0.0728; quiet KL s6 0.00051; 0.69 s/step; the job: 67.5 min spent, about 114.7 min of steps left; card peak 14.1 GB +[ 4082.9s] stage 6/8 s6_tryfail step 1600/4000: train bpb s6 0.0721; quiet KL s6 0.00050; 0.69 s/step; the job: 68.0 min spent, about 114.2 min of steps left; card peak 14.1 GB +[ 4117.9s] stage 6/8 s6_tryfail step 1650/4000: train bpb s6 0.0734; quiet KL s6 0.00045; 0.69 s/step; the job: 68.6 min spent, about 113.7 min of steps left; card peak 14.1 GB +[ 4152.6s] stage 6/8 s6_tryfail step 1700/4000: train bpb s6 0.0684; quiet KL s6 0.00048; 0.69 s/step; the job: 69.2 min spent, about 113.2 min of steps left; card peak 14.1 GB +[ 4187.0s] stage 6/8 s6_tryfail step 1750/4000: train bpb s6 0.0703; quiet KL s6 0.00050; 0.69 s/step; the job: 69.8 min spent, about 112.7 min of steps left; card peak 14.1 GB +[ 4221.9s] stage 6/8 s6_tryfail step 1800/4000: train bpb s6 0.0681; quiet KL s6 0.00051; 0.69 s/step; the job: 70.4 min spent, about 112.2 min of steps left; card peak 14.1 GB +[ 4257.4s] stage 6/8 s6_tryfail step 1850/4000: train bpb s6 0.0702; quiet KL s6 0.00048; 0.69 s/step; the job: 71.0 min spent, about 111.8 min of steps left; card peak 14.1 GB +[ 4294.3s] stage 6/8 s6_tryfail step 1900/4000: train bpb s6 0.0676; quiet KL s6 0.00046; 0.69 s/step; the job: 71.6 min spent, about 111.3 min of steps left; card peak 14.1 GB +[ 4329.6s] stage 6/8 s6_tryfail step 1950/4000: train bpb s6 0.0682; quiet KL s6 0.00049; 0.69 s/step; the job: 72.2 min spent, about 110.8 min of steps left; card peak 14.1 GB +[ 4364.6s] stage 6/8 s6_tryfail step 2000/4000: train bpb s6 0.0688; quiet KL s6 0.00049; 0.69 s/step; the job: 72.7 min spent, about 110.3 min of steps left; card peak 14.1 GB +[ 4399.0s] stage 6/8 s6_tryfail step 2050/4000: train bpb s6 0.0677; quiet KL s6 0.00045; 0.69 s/step; the job: 73.3 min spent, about 109.8 min of steps left; card peak 14.1 GB +[ 4433.7s] stage 6/8 s6_tryfail step 2100/4000: train bpb s6 0.0646; quiet KL s6 0.00051; 0.69 s/step; the job: 73.9 min spent, about 109.3 min of steps left; card peak 14.1 GB +[ 4468.3s] stage 6/8 s6_tryfail step 2150/4000: train bpb s6 0.0658; quiet KL s6 0.00046; 0.69 s/step; the job: 74.5 min spent, about 108.8 min of steps left; card peak 14.1 GB +[ 4504.2s] stage 6/8 s6_tryfail step 2200/4000: train bpb s6 0.0663; quiet KL s6 0.00051; 0.69 s/step; the job: 75.1 min spent, about 108.3 min of steps left; card peak 14.1 GB +[ 4540.6s] stage 6/8 s6_tryfail step 2250/4000: train bpb s6 0.0638; quiet KL s6 0.00042; 0.69 s/step; the job: 75.7 min spent, about 107.8 min of steps left; card peak 14.1 GB +[ 4577.2s] stage 6/8 s6_tryfail step 2300/4000: train bpb s6 0.0629; quiet KL s6 0.00047; 0.70 s/step; the job: 76.3 min spent, about 107.4 min of steps left; card peak 14.1 GB +[ 4613.7s] stage 6/8 s6_tryfail step 2350/4000: train bpb s6 0.0646; quiet KL s6 0.00044; 0.70 s/step; the job: 76.9 min spent, about 106.9 min of steps left; card peak 14.1 GB +[ 4649.8s] stage 6/8 s6_tryfail step 2400/4000: train bpb s6 0.0642; quiet KL s6 0.00045; 0.70 s/step; the job: 77.5 min spent, about 106.4 min of steps left; card peak 14.1 GB +[ 4686.1s] stage 6/8 s6_tryfail step 2450/4000: train bpb s6 0.0607; quiet KL s6 0.00044; 0.70 s/step; the job: 78.1 min spent, about 105.9 min of steps left; card peak 14.1 GB +[ 4722.1s] stage 6/8 s6_tryfail step 2500/4000: train bpb s6 0.0641; quiet KL s6 0.00043; 0.70 s/step; the job: 78.7 min spent, about 105.5 min of steps left; card peak 14.1 GB +[ 4756.2s] stage 6/8 s6_tryfail step 2550/4000: train bpb s6 0.0633; quiet KL s6 0.00039; 0.70 s/step; the job: 79.3 min spent, about 104.9 min of steps left; card peak 14.1 GB +[ 4790.6s] stage 6/8 s6_tryfail step 2600/4000: train bpb s6 0.0645; quiet KL s6 0.00040; 0.70 s/step; the job: 79.8 min spent, about 104.4 min of steps left; card peak 14.1 GB +[ 4824.9s] stage 6/8 s6_tryfail step 2650/4000: train bpb s6 0.0638; quiet KL s6 0.00041; 0.70 s/step; the job: 80.4 min spent, about 103.9 min of steps left; card peak 14.1 GB +[ 4858.9s] stage 6/8 s6_tryfail step 2700/4000: train bpb s6 0.0614; quiet KL s6 0.00041; 0.70 s/step; the job: 81.0 min spent, about 103.3 min of steps left; card peak 14.1 GB +[ 4893.3s] stage 6/8 s6_tryfail step 2750/4000: train bpb s6 0.0629; quiet KL s6 0.00044; 0.70 s/step; the job: 81.6 min spent, about 102.8 min of steps left; card peak 14.1 GB +[ 4927.5s] stage 6/8 s6_tryfail step 2800/4000: train bpb s6 0.0624; quiet KL s6 0.00042; 0.70 s/step; the job: 82.1 min spent, about 102.3 min of steps left; card peak 14.1 GB +[ 4962.0s] stage 6/8 s6_tryfail step 2850/4000: train bpb s6 0.0639; quiet KL s6 0.00041; 0.70 s/step; the job: 82.7 min spent, about 101.7 min of steps left; card peak 14.1 GB +[ 4996.1s] stage 6/8 s6_tryfail step 2900/4000: train bpb s6 0.0613; quiet KL s6 0.00043; 0.70 s/step; the job: 83.3 min spent, about 101.2 min of steps left; card peak 14.1 GB +[ 5030.4s] stage 6/8 s6_tryfail step 2950/4000: train bpb s6 0.0599; quiet KL s6 0.00040; 0.70 s/step; the job: 83.8 min spent, about 100.6 min of steps left; card peak 14.1 GB +[ 5064.7s] stage 6/8 s6_tryfail step 3000/4000: train bpb s6 0.0638; quiet KL s6 0.00040; 0.70 s/step; the job: 84.4 min spent, about 100.1 min of steps left; card peak 14.1 GB +[ 5098.7s] stage 6/8 s6_tryfail step 3050/4000: train bpb s6 0.0600; quiet KL s6 0.00041; 0.70 s/step; the job: 85.0 min spent, about 99.6 min of steps left; card peak 14.1 GB +[ 5133.0s] stage 6/8 s6_tryfail step 3100/4000: train bpb s6 0.0609; quiet KL s6 0.00041; 0.70 s/step; the job: 85.6 min spent, about 99.0 min of steps left; card peak 14.1 GB +[ 5167.2s] stage 6/8 s6_tryfail step 3150/4000: train bpb s6 0.0604; quiet KL s6 0.00042; 0.69 s/step; the job: 86.1 min spent, about 98.5 min of steps left; card peak 14.1 GB +[ 5200.9s] stage 6/8 s6_tryfail step 3200/4000: train bpb s6 0.0620; quiet KL s6 0.00039; 0.69 s/step; the job: 86.7 min spent, about 97.9 min of steps left; card peak 14.1 GB +[ 5234.8s] stage 6/8 s6_tryfail step 3250/4000: train bpb s6 0.0595; quiet KL s6 0.00041; 0.69 s/step; the job: 87.2 min spent, about 97.4 min of steps left; card peak 14.1 GB +[ 5268.7s] stage 6/8 s6_tryfail step 3300/4000: train bpb s6 0.0595; quiet KL s6 0.00044; 0.69 s/step; the job: 87.8 min spent, about 96.9 min of steps left; card peak 14.1 GB +[ 5302.4s] stage 6/8 s6_tryfail step 3350/4000: train bpb s6 0.0603; quiet KL s6 0.00044; 0.69 s/step; the job: 88.4 min spent, about 96.3 min of steps left; card peak 14.1 GB +[ 5336.2s] stage 6/8 s6_tryfail step 3400/4000: train bpb s6 0.0622; quiet KL s6 0.00045; 0.69 s/step; the job: 88.9 min spent, about 95.8 min of steps left; card peak 14.1 GB +[ 5370.1s] stage 6/8 s6_tryfail step 3450/4000: train bpb s6 0.0580; quiet KL s6 0.00042; 0.69 s/step; the job: 89.5 min spent, about 95.2 min of steps left; card peak 14.1 GB +[ 5404.0s] stage 6/8 s6_tryfail step 3500/4000: train bpb s6 0.0603; quiet KL s6 0.00040; 0.69 s/step; the job: 90.1 min spent, about 94.7 min of steps left; card peak 14.1 GB +[ 5437.7s] stage 6/8 s6_tryfail step 3550/4000: train bpb s6 0.0610; quiet KL s6 0.00040; 0.69 s/step; the job: 90.6 min spent, about 94.1 min of steps left; card peak 14.1 GB +[ 5471.6s] stage 6/8 s6_tryfail step 3600/4000: train bpb s6 0.0593; quiet KL s6 0.00038; 0.69 s/step; the job: 91.2 min spent, about 93.6 min of steps left; card peak 14.1 GB +[ 5505.8s] stage 6/8 s6_tryfail step 3650/4000: train bpb s6 0.0591; quiet KL s6 0.00043; 0.69 s/step; the job: 91.8 min spent, about 93.0 min of steps left; card peak 14.1 GB +[ 5539.9s] stage 6/8 s6_tryfail step 3700/4000: train bpb s6 0.0571; quiet KL s6 0.00041; 0.69 s/step; the job: 92.3 min spent, about 92.5 min of steps left; card peak 14.1 GB +[ 5574.1s] stage 6/8 s6_tryfail step 3750/4000: train bpb s6 0.0578; quiet KL s6 0.00041; 0.69 s/step; the job: 92.9 min spent, about 91.9 min of steps left; card peak 14.1 GB +[ 5608.5s] stage 6/8 s6_tryfail step 3800/4000: train bpb s6 0.0572; quiet KL s6 0.00041; 0.69 s/step; the job: 93.5 min spent, about 91.4 min of steps left; card peak 14.1 GB +[ 5642.5s] stage 6/8 s6_tryfail step 3850/4000: train bpb s6 0.0593; quiet KL s6 0.00040; 0.69 s/step; the job: 94.0 min spent, about 90.8 min of steps left; card peak 14.1 GB +[ 5676.7s] stage 6/8 s6_tryfail step 3900/4000: train bpb s6 0.0614; quiet KL s6 0.00040; 0.69 s/step; the job: 94.6 min spent, about 90.3 min of steps left; card peak 14.1 GB +[ 5710.9s] stage 6/8 s6_tryfail step 3950/4000: train bpb s6 0.0590; quiet KL s6 0.00037; 0.69 s/step; the job: 95.2 min spent, about 89.7 min of steps left; card peak 14.1 GB +[ 5744.9s] stage 6/8 s6_tryfail step 4000/4000: train bpb s6 0.0607; quiet KL s6 0.00042; 0.69 s/step; the job: 95.7 min spent, about 89.2 min of steps left; card peak 14.1 GB +[ 5746.0s] anchors saved: /root/refit/group/anchors/gXB_close6_.safetensors for ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail'] +[ 5930.9s] CLOSE 6/8: the group ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail'] all on, against arms off: +[ 5930.9s] s1_perspective rows: off 0.5538 -> all on 0.0481 (-0.5057 +- 0.0048); with s1_perspective masked 0.5211 (its share -0.4730); s1_perspective alone 0.0522 (-0.5017); gate mean 0.212 / widest 0.555 +[ 5930.9s] s2_concept rows: off 0.7212 -> all on 0.1215 (-0.5997 +- 0.0037); with s2_concept masked 0.2431 (its share -0.1216); s2_concept alone 0.3603 (-0.3608); gate mean 0.273 / widest 0.487 +[ 5930.9s] s3_rules rows: off 0.9296 -> all on 0.2225 (-0.7071 +- 0.0055); with s3_rules masked 0.3181 (its share -0.0956); s3_rules alone 0.8702 (-0.0594); gate mean 0.124 / widest 0.175 +[ 5930.9s] s4_arith rows: off 1.2339 -> all on 0.3364 (-0.8975 +- 0.0058); with s4_arith masked 0.5119 (its share -0.1755); s4_arith alone 1.1550 (-0.0789); gate mean 0.108 / widest 0.167 +[ 5930.9s] s5_causal rows: off 0.6604 -> all on 0.0426 (-0.6178 +- 0.0069); with s5_causal masked 0.4881 (its share -0.4455); s5_causal alone 0.1418 (-0.5187); gate mean 0.186 / widest 0.360 +[ 5930.9s] s6_tryfail rows: off 0.6728 -> all on 0.0546 (-0.6182 +- 0.0036); with s6_tryfail masked 0.4878 (its share -0.4332); s6_tryfail alone 0.1437 (-0.5291); gate mean 0.229 / widest 0.581 +[ 5930.9s] web holdout: off 0.9506 -> all on 0.9521 (+0.00147 +- 0.00034; the quiet bar +.012); each member's share: s1_perspective +0.00006, s2_concept -0.00053, s3_rules +0.00037, s4_arith +0.00016, s5_causal +0.00001, s6_tryfail +0.00009; alone: s1_perspective +0.00134, s2_concept +0.00086, s3_rules +0.00034, s4_arith +0.00006, s5_causal +0.00000, s6_tryfail +0.00011 +[ 5930.9s] a member's share on a PARTNER's rows (all on minus that member masked; near 0 = it stays out): s1_perspective on s2_concept -0.2369; s1_perspective on s3_rules -0.1977; s1_perspective on s4_arith -0.1493; s1_perspective on s5_causal -0.0459; s1_perspective on s6_tryfail -0.0241; s2_concept on s1_perspective -0.0043; s2_concept on s3_rules -0.1012; s2_concept on s4_arith -0.1118; s2_concept on s5_causal -0.0008; s2_concept on s6_tryfail -0.0025; s3_rules on s1_perspective +0.0002; s3_rules on s2_concept -0.0005; s3_rules on s4_arith -0.0005; s3_rules on s5_causal -0.0001; s3_rules on s6_tryfail -0.0002; s4_arith on s1_perspective +0.0001; s4_arith on s2_concept -0.0002; s4_arith on s3_rules -0.0001; s4_arith on s5_causal -0.0001; s4_arith on s6_tryfail -0.0041; s5_causal on s1_perspective +0.0001; s5_causal on s2_concept +0.0000; s5_causal on s3_rules +0.0003; s5_causal on s4_arith -0.0001; s5_causal on s6_tryfail -0.0003; s6_tryfail on s1_perspective +0.0002; s6_tryfail on s2_concept +0.0003; s6_tryfail on s3_rules -0.0001; s6_tryfail on s4_arith -0.0016; s6_tryfail on s5_causal +0.0000 +[ 5930.9s] a member ALONE on a partner's rows (alone minus off): s1_perspective on s2_concept -0.4772; s1_perspective on s3_rules -0.4262; s1_perspective on s4_arith -0.4807; s1_perspective on s5_causal -0.1919; s1_perspective on s6_tryfail -0.1690; s2_concept on s1_perspective -0.0300; s2_concept on s3_rules -0.3138; s2_concept on s4_arith -0.3577; s2_concept on s5_causal -0.0127; s2_concept on s6_tryfail -0.0062; s3_rules on s1_perspective -0.0003; s3_rules on s2_concept +0.0016; s3_rules on s4_arith -0.0047; s3_rules on s5_causal +0.0060; s3_rules on s6_tryfail -0.0011; s4_arith on s1_perspective -0.0013; s4_arith on s2_concept -0.0015; s4_arith on s3_rules +0.0005; s4_arith on s5_causal -0.0011; s4_arith on s6_tryfail -0.0134; s5_causal on s1_perspective -0.0022; s5_causal on s2_concept -0.0010; s5_causal on s3_rules -0.0018; s5_causal on s4_arith -0.0022; s5_causal on s6_tryfail -0.0023; s6_tryfail on s1_perspective +0.0029; s6_tryfail on s2_concept +0.0013; s6_tryfail on s3_rules -0.0003; s6_tryfail on s4_arith -0.0010; s6_tryfail on s5_causal -0.0026 +[ 5930.9s] the group on the rows of stages not yet attached: s7_mixed -0.0066; s8_register -0.1112 +[ 5930.9s] s1_perspective battery accuracy, off -> all on: in 0.550 -> 1.000, transfer 0.463 -> 0.475; alone: in 0.925, transfer 0.463 +[ 5930.9s] s2_concept battery accuracy, off -> all on: in 0.298 -> 0.964, transfer 0.678 -> 0.737; alone: in 0.905, transfer 0.703 +[ 5930.9s] s3_rules battery accuracy, off -> all on: in 0.155 -> 0.890, transfer 0.070 -> 0.330, held-out 0.050 -> 0.467; alone: in 0.335, transfer 0.100, held-out 0.142 +[ 5930.9s] s4_arith battery accuracy, off -> all on: in 0.344 -> 0.619, transfer 0.242 -> 0.283, held-out 0.350 -> 0.325; alone: in 0.412, transfer 0.237, held-out 0.400 +[ 5930.9s] probes mean: off 0.411 -> all on 0.426 +[ 5930.9s] compiled chain: 18 cache entries (limit 256), 0 limit warnings; card peak 14.1 GB; host RSS 5.4 GB +[arms] attached 's7_mixed' for phase 'curriculum_s7': 13.69M params x 32 blocks, lambda 2.0, spec {'n_slots': 16, 'K': 16, 'D': 8, 'hidden': 256}, recompute on, compile on +[ 5935.0s] STAGE 7/8: s7_mixed attached over ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail'] (on, frozen); at birth the group reads exactly as before on its rows and on web text; 4000 steps of 1 task + 1 abstention chunks +[ 5936.7s] stage 7/8 s7_mixed step 1/4000: train bpb s7 0.1839; quiet KL s7 0.00011; 1.77 s/step; the job: 98.9 min spent, about 89.2 min of steps left; card peak 14.2 GB +[stream] opening 'cosmo-young' (HuggingFaceTB/smollm-corpus) … +[stream] opening 'tinystories' (roneneldan/TinyStories) … +[stream] opening 'proofwriter-prose' (tasksource/proofwriter) … +[stream] opening 'fineweb-good' (HuggingFaceFW/fineweb-edu) … +[stream] opening 'simple-wiki' (wikimedia/wikipedia) … +[stream] opening 'atomic-flicker' (Estwld/atomic2020-origin) … +[ 6007.3s] stage 7/8 s7_mixed step 50/4000: train bpb s7 0.4395; quiet KL s7 0.00014; 1.45 s/step; the job: 100.1 min spent, about 89.3 min of steps left; card peak 14.3 GB +[stream] opening 'babi-prose' (Muennighoff/babi) … +Repo card metadata block was not found. Setting CardData to empty. +[ 6045.4s] stage 7/8 s7_mixed step 100/4000: train bpb s7 0.5081; quiet KL s7 0.00016; 1.10 s/step; the job: 100.8 min spent, about 88.8 min of steps left; card peak 14.3 GB +[ 6081.5s] stage 7/8 s7_mixed step 150/4000: train bpb s7 0.4529; quiet KL s7 0.00020; 0.98 s/step; the job: 101.4 min spent, about 88.3 min of steps left; card peak 14.3 GB +[ 6117.7s] stage 7/8 s7_mixed step 200/4000: train bpb s7 0.4662; quiet KL s7 0.00022; 0.91 s/step; the job: 102.0 min spent, about 87.7 min of steps left; card peak 14.3 GB +[ 6154.0s] stage 7/8 s7_mixed step 250/4000: train bpb s7 0.4931; quiet KL s7 0.00031; 0.88 s/step; the job: 102.6 min spent, about 87.2 min of steps left; card peak 14.3 GB +[ 6190.2s] stage 7/8 s7_mixed step 300/4000: train bpb s7 0.5269; quiet KL s7 0.00033; 0.85 s/step; the job: 103.2 min spent, about 86.7 min of steps left; card peak 14.3 GB +[ 6226.6s] stage 7/8 s7_mixed step 350/4000: train bpb s7 0.3769; quiet KL s7 0.00038; 0.83 s/step; the job: 103.8 min spent, about 86.2 min of steps left; card peak 14.3 GB +[ 6262.9s] stage 7/8 s7_mixed step 400/4000: train bpb s7 0.4406; quiet KL s7 0.00036; 0.82 s/step; the job: 104.4 min spent, about 85.7 min of steps left; card peak 14.3 GB +[ 6299.2s] stage 7/8 s7_mixed step 450/4000: train bpb s7 0.4244; quiet KL s7 0.00035; 0.81 s/step; the job: 105.0 min spent, about 85.1 min of steps left; card peak 14.3 GB +[ 6335.5s] stage 7/8 s7_mixed step 500/4000: train bpb s7 0.4225; quiet KL s7 0.00045; 0.80 s/step; the job: 105.6 min spent, about 84.6 min of steps left; card peak 14.3 GB +[ 6371.6s] stage 7/8 s7_mixed step 550/4000: train bpb s7 0.5462; quiet KL s7 0.00045; 0.79 s/step; the job: 106.2 min spent, about 84.1 min of steps left; card peak 14.3 GB +[ 6407.8s] stage 7/8 s7_mixed step 600/4000: train bpb s7 0.3887; quiet KL s7 0.00040; 0.79 s/step; the job: 106.8 min spent, about 83.5 min of steps left; card peak 14.3 GB +[ 6444.2s] stage 7/8 s7_mixed step 650/4000: train bpb s7 0.4559; quiet KL s7 0.00041; 0.78 s/step; the job: 107.4 min spent, about 83.0 min of steps left; card peak 14.3 GB +[ 6480.7s] stage 7/8 s7_mixed step 700/4000: train bpb s7 0.5294; quiet KL s7 0.00056; 0.78 s/step; the job: 108.0 min spent, about 82.5 min of steps left; card peak 14.3 GB +[ 6517.2s] stage 7/8 s7_mixed step 750/4000: train bpb s7 0.4394; quiet KL s7 0.00057; 0.78 s/step; the job: 108.6 min spent, about 82.0 min of steps left; card peak 14.3 GB +[ 6553.7s] stage 7/8 s7_mixed step 800/4000: train bpb s7 0.4267; quiet KL s7 0.00070; 0.77 s/step; the job: 109.2 min spent, about 81.4 min of steps left; card peak 14.3 GB +[ 6590.1s] stage 7/8 s7_mixed step 850/4000: train bpb s7 0.4690; quiet KL s7 0.00063; 0.77 s/step; the job: 109.8 min spent, about 80.9 min of steps left; card peak 14.3 GB +[ 6626.4s] stage 7/8 s7_mixed step 900/4000: train bpb s7 0.5084; quiet KL s7 0.00079; 0.77 s/step; the job: 110.4 min spent, about 80.4 min of steps left; card peak 14.3 GB +[ 6662.7s] stage 7/8 s7_mixed step 950/4000: train bpb s7 0.4823; quiet KL s7 0.00078; 0.77 s/step; the job: 111.0 min spent, about 79.8 min of steps left; card peak 14.3 GB +[ 6699.2s] stage 7/8 s7_mixed step 1000/4000: train bpb s7 0.4056; quiet KL s7 0.00066; 0.76 s/step; the job: 111.7 min spent, about 79.3 min of steps left; card peak 14.3 GB +[ 6735.7s] stage 7/8 s7_mixed step 1050/4000: train bpb s7 0.4017; quiet KL s7 0.00067; 0.76 s/step; the job: 112.3 min spent, about 78.8 min of steps left; card peak 14.3 GB +[ 6772.2s] stage 7/8 s7_mixed step 1100/4000: train bpb s7 0.3963; quiet KL s7 0.00068; 0.76 s/step; the job: 112.9 min spent, about 78.2 min of steps left; card peak 14.3 GB +[ 6808.8s] stage 7/8 s7_mixed step 1150/4000: train bpb s7 0.4763; quiet KL s7 0.00067; 0.76 s/step; the job: 113.5 min spent, about 77.7 min of steps left; card peak 14.3 GB +[ 6845.2s] stage 7/8 s7_mixed step 1200/4000: train bpb s7 0.4271; quiet KL s7 0.00056; 0.76 s/step; the job: 114.1 min spent, about 77.1 min of steps left; card peak 14.3 GB +[ 6881.7s] stage 7/8 s7_mixed step 1250/4000: train bpb s7 0.4475; quiet KL s7 0.00055; 0.76 s/step; the job: 114.7 min spent, about 76.6 min of steps left; card peak 14.3 GB +[ 6918.2s] stage 7/8 s7_mixed step 1300/4000: train bpb s7 0.4104; quiet KL s7 0.00071; 0.76 s/step; the job: 115.3 min spent, about 76.1 min of steps left; card peak 14.3 GB +[ 6954.7s] stage 7/8 s7_mixed step 1350/4000: train bpb s7 0.5277; quiet KL s7 0.00085; 0.76 s/step; the job: 115.9 min spent, about 75.5 min of steps left; card peak 14.3 GB +[ 6990.9s] stage 7/8 s7_mixed step 1400/4000: train bpb s7 0.4393; quiet KL s7 0.00078; 0.75 s/step; the job: 116.5 min spent, about 75.0 min of steps left; card peak 14.3 GB +[ 7027.1s] stage 7/8 s7_mixed step 1450/4000: train bpb s7 0.3759; quiet KL s7 0.00058; 0.75 s/step; the job: 117.1 min spent, about 74.4 min of steps left; card peak 14.3 GB +[ 7063.3s] stage 7/8 s7_mixed step 1500/4000: train bpb s7 0.4307; quiet KL s7 0.00066; 0.75 s/step; the job: 117.7 min spent, about 73.9 min of steps left; card peak 14.3 GB +[ 7099.6s] stage 7/8 s7_mixed step 1550/4000: train bpb s7 0.3765; quiet KL s7 0.00056; 0.75 s/step; the job: 118.3 min spent, about 73.4 min of steps left; card peak 14.3 GB +[ 7135.8s] stage 7/8 s7_mixed step 1600/4000: train bpb s7 0.4728; quiet KL s7 0.00054; 0.75 s/step; the job: 118.9 min spent, about 72.8 min of steps left; card peak 14.3 GB +[ 7172.1s] stage 7/8 s7_mixed step 1650/4000: train bpb s7 0.4451; quiet KL s7 0.00060; 0.75 s/step; the job: 119.5 min spent, about 72.3 min of steps left; card peak 14.3 GB +[ 7208.4s] stage 7/8 s7_mixed step 1700/4000: train bpb s7 0.3963; quiet KL s7 0.00060; 0.75 s/step; the job: 120.1 min spent, about 71.7 min of steps left; card peak 14.3 GB +[ 7244.7s] stage 7/8 s7_mixed step 1750/4000: train bpb s7 0.4506; quiet KL s7 0.00063; 0.75 s/step; the job: 120.7 min spent, about 71.2 min of steps left; card peak 14.3 GB +[ 7281.4s] stage 7/8 s7_mixed step 1800/4000: train bpb s7 0.4440; quiet KL s7 0.00060; 0.75 s/step; the job: 121.4 min spent, about 70.6 min of steps left; card peak 14.3 GB +[ 7317.9s] stage 7/8 s7_mixed step 1850/4000: train bpb s7 0.4507; quiet KL s7 0.00061; 0.75 s/step; the job: 122.0 min spent, about 70.1 min of steps left; card peak 14.3 GB +[ 7354.5s] stage 7/8 s7_mixed step 1900/4000: train bpb s7 0.4530; quiet KL s7 0.00064; 0.75 s/step; the job: 122.6 min spent, about 69.5 min of steps left; card peak 14.3 GB +[ 7391.5s] stage 7/8 s7_mixed step 1950/4000: train bpb s7 0.4510; quiet KL s7 0.00067; 0.75 s/step; the job: 123.2 min spent, about 69.0 min of steps left; card peak 14.3 GB +[ 7428.3s] stage 7/8 s7_mixed step 2000/4000: train bpb s7 0.4289; quiet KL s7 0.00069; 0.75 s/step; the job: 123.8 min spent, about 68.5 min of steps left; card peak 14.3 GB +[ 7465.1s] stage 7/8 s7_mixed step 2050/4000: train bpb s7 0.3810; quiet KL s7 0.00064; 0.75 s/step; the job: 124.4 min spent, about 67.9 min of steps left; card peak 14.3 GB +[ 7501.9s] stage 7/8 s7_mixed step 2100/4000: train bpb s7 0.3504; quiet KL s7 0.00061; 0.75 s/step; the job: 125.0 min spent, about 67.4 min of steps left; card peak 14.3 GB +[ 7538.7s] stage 7/8 s7_mixed step 2150/4000: train bpb s7 0.4724; quiet KL s7 0.00057; 0.75 s/step; the job: 125.6 min spent, about 66.8 min of steps left; card peak 14.3 GB +[ 7575.4s] stage 7/8 s7_mixed step 2200/4000: train bpb s7 0.4324; quiet KL s7 0.00060; 0.75 s/step; the job: 126.3 min spent, about 66.3 min of steps left; card peak 14.3 GB +[ 7612.2s] stage 7/8 s7_mixed step 2250/4000: train bpb s7 0.3504; quiet KL s7 0.00059; 0.75 s/step; the job: 126.9 min spent, about 65.7 min of steps left; card peak 14.3 GB +[ 7649.0s] stage 7/8 s7_mixed step 2300/4000: train bpb s7 0.4314; quiet KL s7 0.00064; 0.75 s/step; the job: 127.5 min spent, about 65.2 min of steps left; card peak 14.3 GB +[ 7685.7s] stage 7/8 s7_mixed step 2350/4000: train bpb s7 0.4042; quiet KL s7 0.00068; 0.75 s/step; the job: 128.1 min spent, about 64.6 min of steps left; card peak 14.3 GB +[ 7722.4s] stage 7/8 s7_mixed step 2400/4000: train bpb s7 0.4466; quiet KL s7 0.00062; 0.74 s/step; the job: 128.7 min spent, about 64.1 min of steps left; card peak 14.3 GB +[ 7759.1s] stage 7/8 s7_mixed step 2450/4000: train bpb s7 0.4690; quiet KL s7 0.00071; 0.74 s/step; the job: 129.3 min spent, about 63.5 min of steps left; card peak 14.3 GB +[ 7795.7s] stage 7/8 s7_mixed step 2500/4000: train bpb s7 0.4238; quiet KL s7 0.00065; 0.74 s/step; the job: 129.9 min spent, about 63.0 min of steps left; card peak 14.3 GB +[ 7832.4s] stage 7/8 s7_mixed step 2550/4000: train bpb s7 0.3321; quiet KL s7 0.00055; 0.74 s/step; the job: 130.5 min spent, about 62.4 min of steps left; card peak 14.3 GB +[ 7869.1s] stage 7/8 s7_mixed step 2600/4000: train bpb s7 0.4020; quiet KL s7 0.00068; 0.74 s/step; the job: 131.2 min spent, about 61.9 min of steps left; card peak 14.3 GB +[ 7905.8s] stage 7/8 s7_mixed step 2650/4000: train bpb s7 0.3683; quiet KL s7 0.00064; 0.74 s/step; the job: 131.8 min spent, about 61.3 min of steps left; card peak 14.3 GB +[ 7942.5s] stage 7/8 s7_mixed step 2700/4000: train bpb s7 0.4454; quiet KL s7 0.00065; 0.74 s/step; the job: 132.4 min spent, about 60.8 min of steps left; card peak 14.3 GB +[ 7979.2s] stage 7/8 s7_mixed step 2750/4000: train bpb s7 0.3968; quiet KL s7 0.00063; 0.74 s/step; the job: 133.0 min spent, about 60.2 min of steps left; card peak 14.3 GB +[ 8015.9s] stage 7/8 s7_mixed step 2800/4000: train bpb s7 0.4423; quiet KL s7 0.00063; 0.74 s/step; the job: 133.6 min spent, about 59.6 min of steps left; card peak 14.3 GB +[ 8052.7s] stage 7/8 s7_mixed step 2850/4000: train bpb s7 0.3865; quiet KL s7 0.00067; 0.74 s/step; the job: 134.2 min spent, about 59.1 min of steps left; card peak 14.3 GB +[ 8089.4s] stage 7/8 s7_mixed step 2900/4000: train bpb s7 0.3749; quiet KL s7 0.00087; 0.74 s/step; the job: 134.8 min spent, about 58.5 min of steps left; card peak 14.3 GB +[ 8126.1s] stage 7/8 s7_mixed step 2950/4000: train bpb s7 0.3731; quiet KL s7 0.00071; 0.74 s/step; the job: 135.4 min spent, about 58.0 min of steps left; card peak 14.3 GB +[ 8163.1s] stage 7/8 s7_mixed step 3000/4000: train bpb s7 0.4007; quiet KL s7 0.00063; 0.74 s/step; the job: 136.1 min spent, about 57.4 min of steps left; card peak 14.3 GB +[ 8200.1s] stage 7/8 s7_mixed step 3050/4000: train bpb s7 0.3594; quiet KL s7 0.00063; 0.74 s/step; the job: 136.7 min spent, about 56.9 min of steps left; card peak 14.3 GB +[ 8236.9s] stage 7/8 s7_mixed step 3100/4000: train bpb s7 0.4813; quiet KL s7 0.00066; 0.74 s/step; the job: 137.3 min spent, about 56.3 min of steps left; card peak 14.3 GB +[ 8273.8s] stage 7/8 s7_mixed step 3150/4000: train bpb s7 0.4742; quiet KL s7 0.00060; 0.74 s/step; the job: 137.9 min spent, about 55.8 min of steps left; card peak 14.3 GB +[ 8310.8s] stage 7/8 s7_mixed step 3200/4000: train bpb s7 0.4433; quiet KL s7 0.00059; 0.74 s/step; the job: 138.5 min spent, about 55.2 min of steps left; card peak 14.3 GB +[ 8347.7s] stage 7/8 s7_mixed step 3250/4000: train bpb s7 0.4342; quiet KL s7 0.00061; 0.74 s/step; the job: 139.1 min spent, about 54.6 min of steps left; card peak 14.3 GB +[ 8384.7s] stage 7/8 s7_mixed step 3300/4000: train bpb s7 0.3662; quiet KL s7 0.00068; 0.74 s/step; the job: 139.7 min spent, about 54.1 min of steps left; card peak 14.3 GB +[ 8421.6s] stage 7/8 s7_mixed step 3350/4000: train bpb s7 0.3878; quiet KL s7 0.00063; 0.74 s/step; the job: 140.4 min spent, about 53.5 min of steps left; card peak 14.3 GB +[ 8458.4s] stage 7/8 s7_mixed step 3400/4000: train bpb s7 0.4314; quiet KL s7 0.00070; 0.74 s/step; the job: 141.0 min spent, about 53.0 min of steps left; card peak 14.3 GB +[ 8495.3s] stage 7/8 s7_mixed step 3450/4000: train bpb s7 0.3234; quiet KL s7 0.00072; 0.74 s/step; the job: 141.6 min spent, about 52.4 min of steps left; card peak 14.3 GB +[ 8532.1s] stage 7/8 s7_mixed step 3500/4000: train bpb s7 0.3902; quiet KL s7 0.00062; 0.74 s/step; the job: 142.2 min spent, about 51.8 min of steps left; card peak 14.3 GB +[ 8568.9s] stage 7/8 s7_mixed step 3550/4000: train bpb s7 0.4182; quiet KL s7 0.00058; 0.74 s/step; the job: 142.8 min spent, about 51.3 min of steps left; card peak 14.3 GB +[ 8605.7s] stage 7/8 s7_mixed step 3600/4000: train bpb s7 0.3988; quiet KL s7 0.00064; 0.74 s/step; the job: 143.4 min spent, about 50.7 min of steps left; card peak 14.3 GB +[ 8642.5s] stage 7/8 s7_mixed step 3650/4000: train bpb s7 0.4450; quiet KL s7 0.00072; 0.74 s/step; the job: 144.0 min spent, about 50.2 min of steps left; card peak 14.3 GB +[ 8679.3s] stage 7/8 s7_mixed step 3700/4000: train bpb s7 0.4566; quiet KL s7 0.00069; 0.74 s/step; the job: 144.7 min spent, about 49.6 min of steps left; card peak 14.3 GB +[ 8716.2s] stage 7/8 s7_mixed step 3750/4000: train bpb s7 0.5194; quiet KL s7 0.00071; 0.74 s/step; the job: 145.3 min spent, about 49.0 min of steps left; card peak 14.3 GB +[ 8753.0s] stage 7/8 s7_mixed step 3800/4000: train bpb s7 0.3606; quiet KL s7 0.00060; 0.74 s/step; the job: 145.9 min spent, about 48.5 min of steps left; card peak 14.3 GB +[ 8789.8s] stage 7/8 s7_mixed step 3850/4000: train bpb s7 0.3989; quiet KL s7 0.00065; 0.74 s/step; the job: 146.5 min spent, about 47.9 min of steps left; card peak 14.3 GB +[ 8826.6s] stage 7/8 s7_mixed step 3900/4000: train bpb s7 0.4600; quiet KL s7 0.00092; 0.74 s/step; the job: 147.1 min spent, about 47.3 min of steps left; card peak 14.3 GB +[ 8863.2s] stage 7/8 s7_mixed step 3950/4000: train bpb s7 0.3999; quiet KL s7 0.00064; 0.74 s/step; the job: 147.7 min spent, about 46.8 min of steps left; card peak 14.3 GB +[ 8900.1s] stage 7/8 s7_mixed step 4000/4000: train bpb s7 0.4241; quiet KL s7 0.00065; 0.74 s/step; the job: 148.3 min spent, about 46.2 min of steps left; card peak 14.3 GB +[ 8901.2s] anchors saved: /root/refit/group/anchors/gXB_close7_.safetensors for ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail', 's7_mixed'] +[ 9115.8s] CLOSE 7/8: the group ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail', 's7_mixed'] all on, against arms off: +[ 9115.8s] s1_perspective rows: off 0.5538 -> all on 0.0482 (-0.5056 +- 0.0048); with s1_perspective masked 0.4921 (its share -0.4439); s1_perspective alone 0.0522 (-0.5017); gate mean 0.212 / widest 0.555 +[ 9115.8s] s2_concept rows: off 0.7212 -> all on 0.1212 (-0.6000 +- 0.0038); with s2_concept masked 0.2420 (its share -0.1208); s2_concept alone 0.3603 (-0.3608); gate mean 0.273 / widest 0.487 +[ 9115.8s] s3_rules rows: off 0.9296 -> all on 0.2217 (-0.7079 +- 0.0055); with s3_rules masked 0.3142 (its share -0.0925); s3_rules alone 0.8702 (-0.0594); gate mean 0.124 / widest 0.175 +[ 9115.8s] s4_arith rows: off 1.2339 -> all on 0.3265 (-0.9074 +- 0.0063); with s4_arith masked 0.4916 (its share -0.1651); s4_arith alone 1.1550 (-0.0789); gate mean 0.108 / widest 0.167 +[ 9115.8s] s5_causal rows: off 0.6604 -> all on 0.0426 (-0.6178 +- 0.0070); with s5_causal masked 0.4702 (its share -0.4276); s5_causal alone 0.1418 (-0.5187); gate mean 0.186 / widest 0.360 +[ 9115.8s] s6_tryfail rows: off 0.6728 -> all on 0.0542 (-0.6186 +- 0.0036); with s6_tryfail masked 0.4688 (its share -0.4146); s6_tryfail alone 0.1437 (-0.5291); gate mean 0.229 / widest 0.581 +[ 9115.8s] s7_mixed rows: off 0.7830 -> all on 0.7196 (-0.0634 +- 0.0390); with s7_mixed masked 0.7764 (its share -0.0568); s7_mixed alone 0.7254 (-0.0576); gate mean 0.253 / widest 0.467 +[ 9115.8s] web holdout: off 0.9506 -> all on 0.9512 (+0.00057 +- 0.00032; the quiet bar +.012); each member's share: s1_perspective +0.00001, s2_concept -0.00103, s3_rules +0.00014, s4_arith -0.00015, s5_causal +0.00011, s6_tryfail -0.00008, s7_mixed -0.00090; alone: s1_perspective +0.00134, s2_concept +0.00086, s3_rules +0.00034, s4_arith +0.00006, s5_causal +0.00000, s6_tryfail +0.00011, s7_mixed -0.00032 +[ 9115.8s] a member's share on a PARTNER's rows (all on minus that member masked; near 0 = it stays out): s1_perspective on s2_concept -0.2305; s1_perspective on s3_rules -0.1953; s1_perspective on s4_arith -0.1345; s1_perspective on s5_causal -0.0308; s1_perspective on s6_tryfail -0.0212; s1_perspective on s7_mixed -0.0043; s2_concept on s1_perspective -0.0041; s2_concept on s3_rules -0.1009; s2_concept on s4_arith -0.0990; s2_concept on s5_causal -0.0009; s2_concept on s6_tryfail -0.0024; s2_concept on s7_mixed -0.0002; s3_rules on s1_perspective +0.0001; s3_rules on s2_concept -0.0005; s3_rules on s4_arith -0.0003; s3_rules on s5_causal -0.0001; s3_rules on s6_tryfail -0.0003; s3_rules on s7_mixed -0.0000; s4_arith on s1_perspective +0.0001; s4_arith on s2_concept -0.0000; s4_arith on s3_rules +0.0001; s4_arith on s5_causal -0.0002; s4_arith on s6_tryfail -0.0038; s4_arith on s7_mixed +0.0002; s5_causal on s1_perspective +0.0000; s5_causal on s2_concept +0.0002; s5_causal on s3_rules +0.0003; s5_causal on s4_arith +0.0001; s5_causal on s6_tryfail -0.0004; s5_causal on s7_mixed +0.0001; s6_tryfail on s1_perspective +0.0002; s6_tryfail on s2_concept +0.0004; s6_tryfail on s3_rules +0.0001; s6_tryfail on s4_arith -0.0015; s6_tryfail on s5_causal +0.0001; s6_tryfail on s7_mixed -0.0000; s7_mixed on s1_perspective +0.0001; s7_mixed on s2_concept -0.0004; s7_mixed on s3_rules -0.0008; s7_mixed on s4_arith -0.0099; s7_mixed on s5_causal +0.0000; s7_mixed on s6_tryfail -0.0004 +[ 9115.8s] a member ALONE on a partner's rows (alone minus off): s1_perspective on s2_concept -0.4772; s1_perspective on s3_rules -0.4262; s1_perspective on s4_arith -0.4807; s1_perspective on s5_causal -0.1919; s1_perspective on s6_tryfail -0.1690; s1_perspective on s7_mixed -0.0088; s2_concept on s1_perspective -0.0300; s2_concept on s3_rules -0.3138; s2_concept on s4_arith -0.3577; s2_concept on s5_causal -0.0127; s2_concept on s6_tryfail -0.0062; s2_concept on s7_mixed +0.0002; s3_rules on s1_perspective -0.0003; s3_rules on s2_concept +0.0016; s3_rules on s4_arith -0.0047; s3_rules on s5_causal +0.0060; s3_rules on s6_tryfail -0.0011; s3_rules on s7_mixed -0.0007; s4_arith on s1_perspective -0.0013; s4_arith on s2_concept -0.0015; s4_arith on s3_rules +0.0005; s4_arith on s5_causal -0.0011; s4_arith on s6_tryfail -0.0134; s4_arith on s7_mixed -0.0003; s5_causal on s1_perspective -0.0022; s5_causal on s2_concept -0.0010; s5_causal on s3_rules -0.0018; s5_causal on s4_arith -0.0022; s5_causal on s6_tryfail -0.0023; s5_causal on s7_mixed -0.0001; s6_tryfail on s1_perspective +0.0029; s6_tryfail on s2_concept +0.0013; s6_tryfail on s3_rules -0.0003; s6_tryfail on s4_arith -0.0010; s6_tryfail on s5_causal -0.0026; s6_tryfail on s7_mixed +0.0001; s7_mixed on s1_perspective -0.0237; s7_mixed on s2_concept -0.0022; s7_mixed on s3_rules -0.0529; s7_mixed on s4_arith -0.0414; s7_mixed on s5_causal -0.0976; s7_mixed on s6_tryfail -0.0774 +[ 9115.8s] the group on the rows of stages not yet attached: s8_register -0.1125 +[ 9115.8s] s1_perspective battery accuracy, off -> all on: in 0.550 -> 1.000, transfer 0.463 -> 0.425; alone: in 0.925, transfer 0.463 +[ 9115.8s] s2_concept battery accuracy, off -> all on: in 0.298 -> 0.952, transfer 0.678 -> 0.737; alone: in 0.905, transfer 0.703 +[ 9115.8s] s3_rules battery accuracy, off -> all on: in 0.155 -> 0.925, transfer 0.070 -> 0.350, held-out 0.050 -> 0.483; alone: in 0.335, transfer 0.100, held-out 0.142 +[ 9115.8s] s4_arith battery accuracy, off -> all on: in 0.344 -> 0.625, transfer 0.242 -> 0.258, held-out 0.350 -> 0.287; alone: in 0.412, transfer 0.237, held-out 0.400 +[ 9115.8s] probes mean: off 0.411 -> all on 0.415 +[ 9115.8s] compiled chain: 23 cache entries (limit 256), 0 limit warnings; card peak 14.3 GB; host RSS 11.9 GB +[arms] attached 's8_register' for phase 'curriculum_s8': 13.69M params x 32 blocks, lambda 2.0, spec {'n_slots': 16, 'K': 16, 'D': 8, 'hidden': 256}, recompute on, compile on +[ 9120.1s] STAGE 8/8: s8_register attached over ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail', 's7_mixed'] (on, frozen); at birth the group reads exactly as before on its rows and on web text; 4000 steps of 1 task + 1 abstention chunks +[ 9122.0s] stage 8/8 s8_register step 1/4000: train bpb s8 0.3816; quiet KL s8 0.00016; 1.89 s/step; the job: 152.0 min spent, about 46.2 min of steps left; card peak 14.4 GB +[ 9159.8s] stage 8/8 s8_register step 50/4000: train bpb s8 0.3760; quiet KL s8 0.00015; 0.79 s/step; the job: 152.7 min spent, about 45.7 min of steps left; card peak 14.5 GB +[ 9198.6s] stage 8/8 s8_register step 100/4000: train bpb s8 0.3565; quiet KL s8 0.00026; 0.79 s/step; the job: 153.3 min spent, about 45.1 min of steps left; card peak 14.5 GB +[ 9237.2s] stage 8/8 s8_register step 150/4000: train bpb s8 0.3341; quiet KL s8 0.00044; 0.78 s/step; the job: 154.0 min spent, about 44.5 min of steps left; card peak 14.5 GB +[ 9276.1s] stage 8/8 s8_register step 200/4000: train bpb s8 0.3108; quiet KL s8 0.00055; 0.78 s/step; the job: 154.6 min spent, about 44.0 min of steps left; card peak 14.5 GB +[ 9314.7s] stage 8/8 s8_register step 250/4000: train bpb s8 0.2774; quiet KL s8 0.00069; 0.78 s/step; the job: 155.2 min spent, about 43.4 min of steps left; card peak 14.5 GB +[ 9353.6s] stage 8/8 s8_register step 300/4000: train bpb s8 0.2435; quiet KL s8 0.00078; 0.78 s/step; the job: 155.9 min spent, about 42.9 min of steps left; card peak 14.5 GB +[ 9392.2s] stage 8/8 s8_register step 350/4000: train bpb s8 0.2065; quiet KL s8 0.00084; 0.78 s/step; the job: 156.5 min spent, about 42.3 min of steps left; card peak 14.5 GB +[ 9431.1s] stage 8/8 s8_register step 400/4000: train bpb s8 0.1763; quiet KL s8 0.00078; 0.78 s/step; the job: 157.2 min spent, about 41.7 min of steps left; card peak 14.5 GB +[ 9469.8s] stage 8/8 s8_register step 450/4000: train bpb s8 0.1446; quiet KL s8 0.00080; 0.78 s/step; the job: 157.8 min spent, about 41.2 min of steps left; card peak 14.5 GB +[ 9508.8s] stage 8/8 s8_register step 500/4000: train bpb s8 0.1194; quiet KL s8 0.00067; 0.78 s/step; the job: 158.5 min spent, about 40.6 min of steps left; card peak 14.5 GB +[ 9547.5s] stage 8/8 s8_register step 550/4000: train bpb s8 0.0959; quiet KL s8 0.00073; 0.78 s/step; the job: 159.1 min spent, about 40.1 min of steps left; card peak 14.5 GB +[ 9586.4s] stage 8/8 s8_register step 600/4000: train bpb s8 0.0803; quiet KL s8 0.00074; 0.78 s/step; the job: 159.8 min spent, about 39.5 min of steps left; card peak 14.5 GB +[ 9625.1s] stage 8/8 s8_register step 650/4000: train bpb s8 0.0669; quiet KL s8 0.00072; 0.78 s/step; the job: 160.4 min spent, about 38.9 min of steps left; card peak 14.5 GB +[ 9664.1s] stage 8/8 s8_register step 700/4000: train bpb s8 0.0574; quiet KL s8 0.00070; 0.78 s/step; the job: 161.1 min spent, about 38.4 min of steps left; card peak 14.5 GB +[ 9702.9s] stage 8/8 s8_register step 750/4000: train bpb s8 0.0489; quiet KL s8 0.00068; 0.78 s/step; the job: 161.7 min spent, about 37.8 min of steps left; card peak 14.5 GB +[ 9741.8s] stage 8/8 s8_register step 800/4000: train bpb s8 0.0437; quiet KL s8 0.00072; 0.78 s/step; the job: 162.4 min spent, about 37.2 min of steps left; card peak 14.5 GB +[ 9780.5s] stage 8/8 s8_register step 850/4000: train bpb s8 0.0395; quiet KL s8 0.00064; 0.78 s/step; the job: 163.0 min spent, about 36.7 min of steps left; card peak 14.5 GB +[ 9819.5s] stage 8/8 s8_register step 900/4000: train bpb s8 0.0366; quiet KL s8 0.00060; 0.78 s/step; the job: 163.7 min spent, about 36.1 min of steps left; card peak 14.5 GB +[ 9858.2s] stage 8/8 s8_register step 950/4000: train bpb s8 0.0337; quiet KL s8 0.00062; 0.78 s/step; the job: 164.3 min spent, about 35.5 min of steps left; card peak 14.5 GB +[ 9896.9s] stage 8/8 s8_register step 1000/4000: train bpb s8 0.0324; quiet KL s8 0.00056; 0.78 s/step; the job: 164.9 min spent, about 35.0 min of steps left; card peak 14.5 GB +[ 9935.9s] stage 8/8 s8_register step 1050/4000: train bpb s8 0.0302; quiet KL s8 0.00054; 0.78 s/step; the job: 165.6 min spent, about 34.4 min of steps left; card peak 14.5 GB +[ 9974.6s] stage 8/8 s8_register step 1100/4000: train bpb s8 0.0336; quiet KL s8 0.00090; 0.78 s/step; the job: 166.2 min spent, about 33.8 min of steps left; card peak 14.5 GB +[10013.6s] stage 8/8 s8_register step 1150/4000: train bpb s8 0.0288; quiet KL s8 0.00059; 0.78 s/step; the job: 166.9 min spent, about 33.3 min of steps left; card peak 14.5 GB +[10052.4s] stage 8/8 s8_register step 1200/4000: train bpb s8 0.0281; quiet KL s8 0.00051; 0.78 s/step; the job: 167.5 min spent, about 32.7 min of steps left; card peak 14.5 GB +[10091.4s] stage 8/8 s8_register step 1250/4000: train bpb s8 0.0271; quiet KL s8 0.00047; 0.78 s/step; the job: 168.2 min spent, about 32.1 min of steps left; card peak 14.5 GB +[10130.1s] stage 8/8 s8_register step 1300/4000: train bpb s8 0.0274; quiet KL s8 0.00045; 0.78 s/step; the job: 168.8 min spent, about 31.6 min of steps left; card peak 14.5 GB +[10169.0s] stage 8/8 s8_register step 1350/4000: train bpb s8 0.0261; quiet KL s8 0.00045; 0.78 s/step; the job: 169.5 min spent, about 31.0 min of steps left; card peak 14.5 GB +[10207.8s] stage 8/8 s8_register step 1400/4000: train bpb s8 0.0258; quiet KL s8 0.00045; 0.78 s/step; the job: 170.1 min spent, about 30.4 min of steps left; card peak 14.5 GB +[10246.8s] stage 8/8 s8_register step 1450/4000: train bpb s8 0.0258; quiet KL s8 0.00046; 0.78 s/step; the job: 170.8 min spent, about 29.8 min of steps left; card peak 14.5 GB +[10285.2s] stage 8/8 s8_register step 1500/4000: train bpb s8 0.0254; quiet KL s8 0.00045; 0.78 s/step; the job: 171.4 min spent, about 29.3 min of steps left; card peak 14.5 GB +[10323.8s] stage 8/8 s8_register step 1550/4000: train bpb s8 0.0246; quiet KL s8 0.00040; 0.78 s/step; the job: 172.1 min spent, about 28.7 min of steps left; card peak 14.5 GB +[10362.2s] stage 8/8 s8_register step 1600/4000: train bpb s8 0.0251; quiet KL s8 0.00041; 0.78 s/step; the job: 172.7 min spent, about 28.1 min of steps left; card peak 14.5 GB +[10400.7s] stage 8/8 s8_register step 1650/4000: train bpb s8 0.0250; quiet KL s8 0.00043; 0.78 s/step; the job: 173.3 min spent, about 27.5 min of steps left; card peak 14.5 GB +[10439.1s] stage 8/8 s8_register step 1700/4000: train bpb s8 0.0250; quiet KL s8 0.00040; 0.78 s/step; the job: 174.0 min spent, about 27.0 min of steps left; card peak 14.5 GB +[10477.9s] stage 8/8 s8_register step 1750/4000: train bpb s8 0.0247; quiet KL s8 0.00041; 0.78 s/step; the job: 174.6 min spent, about 26.4 min of steps left; card peak 14.5 GB +[10516.3s] stage 8/8 s8_register step 1800/4000: train bpb s8 0.0245; quiet KL s8 0.00043; 0.78 s/step; the job: 175.3 min spent, about 25.8 min of steps left; card peak 14.5 GB +[10555.2s] stage 8/8 s8_register step 1850/4000: train bpb s8 0.0246; quiet KL s8 0.00041; 0.78 s/step; the job: 175.9 min spent, about 25.2 min of steps left; card peak 14.5 GB +[10593.9s] stage 8/8 s8_register step 1900/4000: train bpb s8 0.0243; quiet KL s8 0.00036; 0.78 s/step; the job: 176.6 min spent, about 24.7 min of steps left; card peak 14.5 GB +[10632.8s] stage 8/8 s8_register step 1950/4000: train bpb s8 0.0243; quiet KL s8 0.00036; 0.78 s/step; the job: 177.2 min spent, about 24.1 min of steps left; card peak 14.5 GB +[10671.5s] stage 8/8 s8_register step 2000/4000: train bpb s8 0.0240; quiet KL s8 0.00040; 0.78 s/step; the job: 177.9 min spent, about 23.5 min of steps left; card peak 14.5 GB +[10710.6s] stage 8/8 s8_register step 2050/4000: train bpb s8 0.0244; quiet KL s8 0.00036; 0.78 s/step; the job: 178.5 min spent, about 22.9 min of steps left; card peak 14.5 GB +[10749.2s] stage 8/8 s8_register step 2100/4000: train bpb s8 0.0240; quiet KL s8 0.00039; 0.78 s/step; the job: 179.2 min spent, about 22.3 min of steps left; card peak 14.5 GB +[10788.3s] stage 8/8 s8_register step 2150/4000: train bpb s8 0.0238; quiet KL s8 0.00037; 0.78 s/step; the job: 179.8 min spent, about 21.8 min of steps left; card peak 14.5 GB +[10827.0s] stage 8/8 s8_register step 2200/4000: train bpb s8 0.0235; quiet KL s8 0.00038; 0.78 s/step; the job: 180.5 min spent, about 21.2 min of steps left; card peak 14.5 GB +[10865.8s] stage 8/8 s8_register step 2250/4000: train bpb s8 0.0233; quiet KL s8 0.00036; 0.78 s/step; the job: 181.1 min spent, about 20.6 min of steps left; card peak 14.5 GB +[10905.1s] stage 8/8 s8_register step 2300/4000: train bpb s8 0.0232; quiet KL s8 0.00037; 0.78 s/step; the job: 181.8 min spent, about 20.0 min of steps left; card peak 14.5 GB +[10943.8s] stage 8/8 s8_register step 2350/4000: train bpb s8 0.0233; quiet KL s8 0.00036; 0.78 s/step; the job: 182.4 min spent, about 19.4 min of steps left; card peak 14.5 GB +[10983.1s] stage 8/8 s8_register step 2400/4000: train bpb s8 0.0236; quiet KL s8 0.00035; 0.78 s/step; the job: 183.1 min spent, about 18.9 min of steps left; card peak 14.5 GB +[11022.0s] stage 8/8 s8_register step 2450/4000: train bpb s8 0.0238; quiet KL s8 0.00036; 0.78 s/step; the job: 183.7 min spent, about 18.3 min of steps left; card peak 14.5 GB +[11061.2s] stage 8/8 s8_register step 2500/4000: train bpb s8 0.0230; quiet KL s8 0.00036; 0.78 s/step; the job: 184.4 min spent, about 17.7 min of steps left; card peak 14.5 GB +[11100.0s] stage 8/8 s8_register step 2550/4000: train bpb s8 0.0230; quiet KL s8 0.00033; 0.78 s/step; the job: 185.0 min spent, about 17.1 min of steps left; card peak 14.5 GB +[11138.8s] stage 8/8 s8_register step 2600/4000: train bpb s8 0.0235; quiet KL s8 0.00032; 0.78 s/step; the job: 185.6 min spent, about 16.5 min of steps left; card peak 14.5 GB +[11177.2s] stage 8/8 s8_register step 2650/4000: train bpb s8 0.0233; quiet KL s8 0.00038; 0.78 s/step; the job: 186.3 min spent, about 15.9 min of steps left; card peak 14.5 GB +[11215.9s] stage 8/8 s8_register step 2700/4000: train bpb s8 0.0227; quiet KL s8 0.00035; 0.78 s/step; the job: 186.9 min spent, about 15.3 min of steps left; card peak 14.5 GB +[11254.3s] stage 8/8 s8_register step 2750/4000: train bpb s8 0.0233; quiet KL s8 0.00035; 0.78 s/step; the job: 187.6 min spent, about 14.8 min of steps left; card peak 14.5 GB +[11293.0s] stage 8/8 s8_register step 2800/4000: train bpb s8 0.0223; quiet KL s8 0.00031; 0.78 s/step; the job: 188.2 min spent, about 14.2 min of steps left; card peak 14.5 GB +[11331.4s] stage 8/8 s8_register step 2850/4000: train bpb s8 0.0227; quiet KL s8 0.00035; 0.78 s/step; the job: 188.9 min spent, about 13.6 min of steps left; card peak 14.5 GB +[11370.0s] stage 8/8 s8_register step 2900/4000: train bpb s8 0.0228; quiet KL s8 0.00037; 0.78 s/step; the job: 189.5 min spent, about 13.0 min of steps left; card peak 14.5 GB +[11408.4s] stage 8/8 s8_register step 2950/4000: train bpb s8 0.0245; quiet KL s8 0.00072; 0.78 s/step; the job: 190.1 min spent, about 12.4 min of steps left; card peak 14.5 GB +[11447.1s] stage 8/8 s8_register step 3000/4000: train bpb s8 0.0237; quiet KL s8 0.00042; 0.78 s/step; the job: 190.8 min spent, about 11.8 min of steps left; card peak 14.5 GB +[11485.5s] stage 8/8 s8_register step 3050/4000: train bpb s8 0.0232; quiet KL s8 0.00040; 0.78 s/step; the job: 191.4 min spent, about 11.2 min of steps left; card peak 14.5 GB +[11524.2s] stage 8/8 s8_register step 3100/4000: train bpb s8 0.0227; quiet KL s8 0.00035; 0.78 s/step; the job: 192.1 min spent, about 10.7 min of steps left; card peak 14.5 GB +[11562.6s] stage 8/8 s8_register step 3150/4000: train bpb s8 0.0221; quiet KL s8 0.00031; 0.78 s/step; the job: 192.7 min spent, about 10.1 min of steps left; card peak 14.5 GB +[11601.0s] stage 8/8 s8_register step 3200/4000: train bpb s8 0.0231; quiet KL s8 0.00035; 0.78 s/step; the job: 193.3 min spent, about 9.5 min of steps left; card peak 14.5 GB +[11639.7s] stage 8/8 s8_register step 3250/4000: train bpb s8 0.0227; quiet KL s8 0.00033; 0.78 s/step; the job: 194.0 min spent, about 8.9 min of steps left; card peak 14.5 GB +[11678.2s] stage 8/8 s8_register step 3300/4000: train bpb s8 0.0226; quiet KL s8 0.00032; 0.78 s/step; the job: 194.6 min spent, about 8.3 min of steps left; card peak 14.5 GB +[11716.9s] stage 8/8 s8_register step 3350/4000: train bpb s8 0.0228; quiet KL s8 0.00034; 0.78 s/step; the job: 195.3 min spent, about 7.7 min of steps left; card peak 14.5 GB +[11755.3s] stage 8/8 s8_register step 3400/4000: train bpb s8 0.0221; quiet KL s8 0.00032; 0.78 s/step; the job: 195.9 min spent, about 7.1 min of steps left; card peak 14.5 GB +[11794.0s] stage 8/8 s8_register step 3450/4000: train bpb s8 0.0227; quiet KL s8 0.00034; 0.78 s/step; the job: 196.6 min spent, about 6.5 min of steps left; card peak 14.5 GB +[11832.4s] stage 8/8 s8_register step 3500/4000: train bpb s8 0.0224; quiet KL s8 0.00034; 0.77 s/step; the job: 197.2 min spent, about 5.9 min of steps left; card peak 14.5 GB +[11871.2s] stage 8/8 s8_register step 3550/4000: train bpb s8 0.0224; quiet KL s8 0.00035; 0.77 s/step; the job: 197.9 min spent, about 5.3 min of steps left; card peak 14.5 GB +[11909.7s] stage 8/8 s8_register step 3600/4000: train bpb s8 0.0227; quiet KL s8 0.00035; 0.77 s/step; the job: 198.5 min spent, about 4.7 min of steps left; card peak 14.5 GB +[11948.4s] stage 8/8 s8_register step 3650/4000: train bpb s8 0.0227; quiet KL s8 0.00030; 0.77 s/step; the job: 199.1 min spent, about 4.2 min of steps left; card peak 14.5 GB +[11986.9s] stage 8/8 s8_register step 3700/4000: train bpb s8 0.0222; quiet KL s8 0.00033; 0.77 s/step; the job: 199.8 min spent, about 3.6 min of steps left; card peak 14.5 GB +[12025.7s] stage 8/8 s8_register step 3750/4000: train bpb s8 0.0221; quiet KL s8 0.00035; 0.77 s/step; the job: 200.4 min spent, about 3.0 min of steps left; card peak 14.5 GB +[12064.1s] stage 8/8 s8_register step 3800/4000: train bpb s8 0.0224; quiet KL s8 0.00034; 0.77 s/step; the job: 201.1 min spent, about 2.4 min of steps left; card peak 14.5 GB +[12102.9s] stage 8/8 s8_register step 3850/4000: train bpb s8 0.0230; quiet KL s8 0.00042; 0.77 s/step; the job: 201.7 min spent, about 1.8 min of steps left; card peak 14.5 GB +[12141.5s] stage 8/8 s8_register step 3900/4000: train bpb s8 0.0221; quiet KL s8 0.00035; 0.77 s/step; the job: 202.4 min spent, about 1.2 min of steps left; card peak 14.5 GB +[12180.3s] stage 8/8 s8_register step 3950/4000: train bpb s8 0.0224; quiet KL s8 0.00033; 0.77 s/step; the job: 203.0 min spent, about 0.6 min of steps left; card peak 14.5 GB +[12218.9s] stage 8/8 s8_register step 4000/4000: train bpb s8 0.0225; quiet KL s8 0.00034; 0.77 s/step; the job: 203.6 min spent, about 0.0 min of steps left; card peak 14.5 GB +[12220.1s] anchors saved: /root/refit/group/anchors/gXB_close8_.safetensors for ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail', 's7_mixed', 's8_register'] +[12467.9s] CLOSE 8/8: the group ['s1_perspective', 's2_concept', 's3_rules', 's4_arith', 's5_causal', 's6_tryfail', 's7_mixed', 's8_register'] all on, against arms off: +[12467.9s] s1_perspective rows: off 0.5538 -> all on 0.0484 (-0.5055 +- 0.0048); with s1_perspective masked 0.4889 (its share -0.4405); s1_perspective alone 0.0522 (-0.5017); gate mean 0.212 / widest 0.555 +[12467.9s] s2_concept rows: off 0.7212 -> all on 0.1212 (-0.6000 +- 0.0038); with s2_concept masked 0.2402 (its share -0.1190); s2_concept alone 0.3603 (-0.3608); gate mean 0.273 / widest 0.487 +[12467.9s] s3_rules rows: off 0.9296 -> all on 0.2222 (-0.7074 +- 0.0055); with s3_rules masked 0.3137 (its share -0.0915); s3_rules alone 0.8702 (-0.0594); gate mean 0.124 / widest 0.175 +[12467.9s] s4_arith rows: off 1.2339 -> all on 0.3266 (-0.9073 +- 0.0062); with s4_arith masked 0.4909 (its share -0.1643); s4_arith alone 1.1550 (-0.0789); gate mean 0.108 / widest 0.167 +[12467.9s] s5_causal rows: off 0.6604 -> all on 0.0428 (-0.6176 +- 0.0069); with s5_causal masked 0.4673 (its share -0.4245); s5_causal alone 0.1418 (-0.5187); gate mean 0.186 / widest 0.360 +[12467.9s] s6_tryfail rows: off 0.6728 -> all on 0.0543 (-0.6185 +- 0.0036); with s6_tryfail masked 0.4663 (its share -0.4120); s6_tryfail alone 0.1437 (-0.5291); gate mean 0.229 / widest 0.581 +[12467.9s] s7_mixed rows: off 0.7830 -> all on 0.7198 (-0.0632 +- 0.0390); with s7_mixed masked 0.7767 (its share -0.0569); s7_mixed alone 0.7254 (-0.0576); gate mean 0.253 / widest 0.467 +[12467.9s] s8_register rows: off 0.4907 -> all on 0.0208 (-0.4699 +- 0.0086); with s8_register masked 0.3781 (its share -0.3573); s8_register alone 0.0516 (-0.4390); gate mean 0.205 / widest 0.470 +[12467.9s] web holdout: off 0.9506 -> all on 0.9516 (+0.00103 +- 0.00020; the quiet bar +.012); each member's share: s1_perspective +0.00043, s2_concept -0.00073, s3_rules +0.00059, s4_arith +0.00017, s5_causal +0.00029, s6_tryfail +0.00025, s7_mixed -0.00070, s8_register +0.00046; alone: s1_perspective +0.00134, s2_concept +0.00086, s3_rules +0.00034, s4_arith +0.00006, s5_causal +0.00000, s6_tryfail +0.00011, s7_mixed -0.00032, s8_register +0.00005 +[12467.9s] a member's share on a PARTNER's rows (all on minus that member masked; near 0 = it stays out): s1_perspective on s2_concept -0.2289; s1_perspective on s3_rules -0.1943; s1_perspective on s4_arith -0.1300; s1_perspective on s5_causal -0.0282; s1_perspective on s6_tryfail -0.0206; s1_perspective on s7_mixed -0.0043; s1_perspective on s8_register -0.0214; s2_concept on s1_perspective -0.0042; s2_concept on s3_rules -0.1006; s2_concept on s4_arith -0.0957; s2_concept on s5_causal -0.0009; s2_concept on s6_tryfail -0.0023; s2_concept on s7_mixed -0.0004; s2_concept on s8_register -0.0005; s3_rules on s1_perspective +0.0000; s3_rules on s2_concept -0.0005; s3_rules on s4_arith -0.0004; s3_rules on s5_causal -0.0001; s3_rules on s6_tryfail -0.0003; s3_rules on s7_mixed -0.0001; s3_rules on s8_register +0.0000; s4_arith on s1_perspective +0.0000; s4_arith on s2_concept -0.0002; s4_arith on s3_rules +0.0001; s4_arith on s5_causal -0.0002; s4_arith on s6_tryfail -0.0039; s4_arith on s7_mixed -0.0000; s4_arith on s8_register -0.0001; s5_causal on s1_perspective +0.0001; s5_causal on s2_concept +0.0002; s5_causal on s3_rules +0.0004; s5_causal on s4_arith +0.0001; s5_causal on s6_tryfail -0.0003; s5_causal on s7_mixed +0.0001; s5_causal on s8_register -0.0001; s6_tryfail on s1_perspective +0.0000; s6_tryfail on s2_concept +0.0003; s6_tryfail on s3_rules +0.0000; s6_tryfail on s4_arith -0.0016; s6_tryfail on s5_causal +0.0000; s6_tryfail on s7_mixed -0.0001; s6_tryfail on s8_register -0.0005; s7_mixed on s1_perspective +0.0001; s7_mixed on s2_concept -0.0003; s7_mixed on s3_rules -0.0007; s7_mixed on s4_arith -0.0099; s7_mixed on s5_causal +0.0000; s7_mixed on s6_tryfail -0.0003; s7_mixed on s8_register -0.0002; s8_register on s1_perspective +0.0001; s8_register on s2_concept +0.0001; s8_register on s3_rules +0.0005; s8_register on s4_arith +0.0001; s8_register on s5_causal +0.0002; s8_register on s6_tryfail +0.0001; s8_register on s7_mixed +0.0002 +[12467.9s] a member ALONE on a partner's rows (alone minus off): s1_perspective on s2_concept -0.4772; s1_perspective on s3_rules -0.4262; s1_perspective on s4_arith -0.4807; s1_perspective on s5_causal -0.1919; s1_perspective on s6_tryfail -0.1690; s1_perspective on s7_mixed -0.0088; s1_perspective on s8_register -0.1105; s2_concept on s1_perspective -0.0300; s2_concept on s3_rules -0.3138; s2_concept on s4_arith -0.3577; s2_concept on s5_causal -0.0127; s2_concept on s6_tryfail -0.0062; s2_concept on s7_mixed +0.0002; s2_concept on s8_register -0.0193; s3_rules on s1_perspective -0.0003; s3_rules on s2_concept +0.0016; s3_rules on s4_arith -0.0047; s3_rules on s5_causal +0.0060; s3_rules on s6_tryfail -0.0011; s3_rules on s7_mixed -0.0007; s3_rules on s8_register -0.0044; s4_arith on s1_perspective -0.0013; s4_arith on s2_concept -0.0015; s4_arith on s3_rules +0.0005; s4_arith on s5_causal -0.0011; s4_arith on s6_tryfail -0.0134; s4_arith on s7_mixed -0.0003; s4_arith on s8_register -0.0014; s5_causal on s1_perspective -0.0022; s5_causal on s2_concept -0.0010; s5_causal on s3_rules -0.0018; s5_causal on s4_arith -0.0022; s5_causal on s6_tryfail -0.0023; s5_causal on s7_mixed -0.0001; s5_causal on s8_register -0.0009; s6_tryfail on s1_perspective +0.0029; s6_tryfail on s2_concept +0.0013; s6_tryfail on s3_rules -0.0003; s6_tryfail on s4_arith -0.0010; s6_tryfail on s5_causal -0.0026; s6_tryfail on s7_mixed +0.0001; s6_tryfail on s8_register -0.0001; s7_mixed on s1_perspective -0.0237; s7_mixed on s2_concept -0.0022; s7_mixed on s3_rules -0.0529; s7_mixed on s4_arith -0.0414; s7_mixed on s5_causal -0.0976; s7_mixed on s6_tryfail -0.0774; s7_mixed on s8_register -0.0061; s8_register on s1_perspective -0.0045; s8_register on s2_concept -0.0050; s8_register on s3_rules -0.0063; s8_register on s4_arith -0.0064; s8_register on s5_causal -0.0139; s8_register on s6_tryfail -0.0187; s8_register on s7_mixed +0.0001 +[12467.9s] s1_perspective battery accuracy, off -> all on: in 0.550 -> 1.000, transfer 0.463 -> 0.425; alone: in 0.925, transfer 0.463 +[12467.9s] s2_concept battery accuracy, off -> all on: in 0.298 -> 0.952, transfer 0.678 -> 0.720; alone: in 0.905, transfer 0.703 +[12467.9s] s3_rules battery accuracy, off -> all on: in 0.155 -> 0.920, transfer 0.070 -> 0.370, held-out 0.050 -> 0.458; alone: in 0.335, transfer 0.100, held-out 0.142 +[12467.9s] s4_arith battery accuracy, off -> all on: in 0.344 -> 0.656, transfer 0.242 -> 0.271, held-out 0.350 -> 0.325; alone: in 0.412, transfer 0.237, held-out 0.400 +[12467.9s] probes mean: off 0.411 -> all on 0.430 +[12467.9s] compiled chain: 28 cache entries (limit 256), 0 limit warnings; card peak 14.5 GB; host RSS 11.9 GB +[arms] adapter chain compile OFF on 256 wrapped blocks (8 arms) +[12471.8s] the write of s1_perspective (mean |delta| / |x|, all members live): own rows 1.64e-02; web 6.14e-03 (own over web 2.67x); partners' rows s2_concept 1.31e-02, s3_rules 1.50e-02, s4_arith 1.57e-02, s5_causal 1.42e-02, s6_tryfail 1.41e-02, s7_mixed 6.09e-03, s8_register 1.26e-02 +[12471.8s] the write of s2_concept (mean |delta| / |x|, all members live): own rows 1.12e-02; web 4.23e-03 (own over web 2.65x); partners' rows s1_perspective 8.16e-03, s3_rules 1.11e-02, s4_arith 1.14e-02, s5_causal 8.44e-03, s6_tryfail 8.74e-03, s7_mixed 4.29e-03, s8_register 8.23e-03 +[12471.8s] the write of s3_rules (mean |delta| / |x|, all members live): own rows 8.97e-03; web 1.50e-03 (own over web 5.97x); partners' rows s1_perspective 1.89e-03, s2_concept 1.94e-03, s4_arith 1.52e-03, s5_causal 2.20e-03, s6_tryfail 1.99e-03, s7_mixed 1.47e-03, s8_register 1.81e-03 +[12471.8s] the write of s4_arith (mean |delta| / |x|, all members live): own rows 8.04e-03; web 1.23e-03 (own over web 6.55x); partners' rows s1_perspective 1.64e-03, s2_concept 1.54e-03, s3_rules 1.63e-03, s5_causal 1.99e-03, s6_tryfail 3.98e-03, s7_mixed 1.24e-03, s8_register 1.79e-03 +[12471.8s] the write of s5_causal (mean |delta| / |x|, all members live): own rows 1.49e-02; web 3.35e-03 (own over web 4.43x); partners' rows s1_perspective 4.64e-03, s2_concept 3.67e-03, s3_rules 4.01e-03, s4_arith 3.97e-03, s6_tryfail 4.55e-03, s7_mixed 3.37e-03, s8_register 4.76e-03 +[12471.8s] the write of s6_tryfail (mean |delta| / |x|, all members live): own rows 2.06e-02; web 4.95e-03 (own over web 4.16x); partners' rows s1_perspective 6.90e-03, s2_concept 5.40e-03, s3_rules 5.94e-03, s4_arith 6.91e-03, s5_causal 7.94e-03, s7_mixed 4.94e-03, s8_register 7.25e-03 +[12471.8s] the write of s7_mixed (mean |delta| / |x|, all members live): own rows 3.75e-03; web 3.84e-03 (own over web 0.98x); partners' rows s1_perspective 6.99e-03, s2_concept 5.11e-03, s3_rules 7.15e-03, s4_arith 6.53e-03, s5_causal 6.55e-03, s6_tryfail 6.59e-03, s8_register 5.28e-03 +[12471.8s] the write of s8_register (mean |delta| / |x|, all members live): own rows 1.61e-02; web 3.78e-03 (own over web 4.26x); partners' rows s1_perspective 5.60e-03, s2_concept 4.38e-03, s3_rules 4.42e-03, s4_arith 4.65e-03, s5_causal 6.37e-03, s6_tryfail 5.93e-03, s7_mixed 3.74e-03 +[12471.8s] DONE gXB: 5 closes; 12472 s; card peak 14.5 GB; host RSS 11.9 GB; compiled chain: 28 cache entries (limit 256), 0 limit warnings diff --git a/mini-beatrix-3/arm_refit/logs/long_card0.log b/mini-beatrix-3/arm_refit/logs/long_card0.log index 052526940f4bc6f2f2a41c65c401d0afea7e335a..9fff5e6da1bf2dfdd7812aad6295b7ac2c2f43f7 100644 --- a/mini-beatrix-3/arm_refit/logs/long_card0.log +++ b/mini-beatrix-3/arm_refit/logs/long_card0.log @@ -22,3 +22,66 @@ Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF [ 340.5s] step 500/3200: train bpb 0.5214; quiet term 0.00139; 0.53 s/step; 4.4 min spent, about 23.9 min of steps left; card peak 13.7 GB [ 364.0s] step 550/3200: train bpb 0.4924; quiet term 0.00145; 0.53 s/step; 4.8 min spent, about 23.2 min of steps left; card peak 13.7 GB [ 387.2s] step 600/3200: train bpb 0.4669; quiet term 0.00150; 0.52 s/step; 5.2 min spent, about 22.5 min of steps left; card peak 13.7 GB +[ 410.8s] step 650/3200: train bpb 0.4415; quiet term 0.00155; 0.52 s/step; 5.6 min spent, about 21.9 min of steps left; card peak 13.7 GB +[ 434.2s] step 700/3200: train bpb 0.4144; quiet term 0.00157; 0.51 s/step; 6.0 min spent, about 21.4 min of steps left; card peak 13.7 GB +[ 457.7s] step 750/3200: train bpb 0.3935; quiet term 0.00156; 0.51 s/step; 6.4 min spent, about 20.8 min of steps left; card peak 13.7 GB +[ 481.3s] step 800/3200: train bpb 0.3748; quiet term 0.00167; 0.51 s/step; 6.8 min spent, about 20.3 min of steps left; card peak 13.7 GB +[ 488.7s] after 800 steps: own rows live-minus-masked -0.55725 +- 0.00538 (0.9296 -> 0.3724); web holdout +0.00095 +- 0.00011 (the quiet bar +.012); battery in +1.478 (se 0.081; acc 0.155 -> 0.600); battery transfer +0.631 (se 0.129; acc 0.070 -> 0.120); battery held-out +1.154 (se 0.129; acc 0.050 -> 0.267); gates open mean 0.128 / widest 0.181; host RSS 4.9 GB +[ 512.3s] step 850/3200: train bpb 0.3561; quiet term 0.00169; 0.51 s/step; 7.3 min spent, about 20.1 min of steps left; card peak 13.7 GB +[ 536.3s] step 900/3200: train bpb 0.3332; quiet term 0.00166; 0.51 s/step; 7.7 min spent, about 19.6 min of steps left; card peak 13.7 GB +[ 560.0s] step 950/3200: train bpb 0.3168; quiet term 0.00175; 0.51 s/step; 8.1 min spent, about 19.1 min of steps left; card peak 13.7 GB +[ 584.1s] step 1000/3200: train bpb 0.3012; quiet term 0.00167; 0.51 s/step; 8.5 min spent, about 18.7 min of steps left; card peak 13.7 GB +[ 607.9s] step 1050/3200: train bpb 0.2874; quiet term 0.00172; 0.51 s/step; 8.9 min spent, about 18.2 min of steps left; card peak 13.7 GB +[ 631.9s] step 1100/3200: train bpb 0.2771; quiet term 0.00175; 0.51 s/step; 9.3 min spent, about 17.7 min of steps left; card peak 13.7 GB +[ 655.9s] step 1150/3200: train bpb 0.2633; quiet term 0.00179; 0.51 s/step; 9.7 min spent, about 17.3 min of steps left; card peak 13.7 GB +[ 679.7s] step 1200/3200: train bpb 0.2523; quiet term 0.00169; 0.50 s/step; 10.1 min spent, about 16.8 min of steps left; card peak 13.7 GB +[ 687.2s] after 1200 steps: own rows live-minus-masked -0.67264 +- 0.00509 (0.9296 -> 0.2570); web holdout +0.00088 +- 0.00014 (the quiet bar +.012); battery in +1.961 (se 0.103; acc 0.155 -> 0.730); battery transfer +0.882 (se 0.116; acc 0.070 -> 0.210); battery held-out +1.656 (se 0.142; acc 0.050 -> 0.433); gates open mean 0.171 / widest 0.290; host RSS 4.9 GB +[ 711.0s] step 1250/3200: train bpb 0.2449; quiet term 0.00174; 0.51 s/step; 10.6 min spent, about 16.5 min of steps left; card peak 13.7 GB +[ 734.5s] step 1300/3200: train bpb 0.2413; quiet term 0.00162; 0.51 s/step; 11.0 min spent, about 16.1 min of steps left; card peak 13.7 GB +[ 758.2s] step 1350/3200: train bpb 0.2333; quiet term 0.00170; 0.51 s/step; 11.4 min spent, about 15.6 min of steps left; card peak 13.7 GB +[ 781.7s] step 1400/3200: train bpb 0.2290; quiet term 0.00181; 0.50 s/step; 11.8 min spent, about 15.1 min of steps left; card peak 13.7 GB +[ 805.4s] step 1450/3200: train bpb 0.2219; quiet term 0.00198; 0.50 s/step; 12.2 min spent, about 14.7 min of steps left; card peak 13.7 GB +[ 829.0s] step 1500/3200: train bpb 0.2214; quiet term 0.00156; 0.50 s/step; 12.6 min spent, about 14.2 min of steps left; card peak 13.7 GB +[ 852.5s] step 1550/3200: train bpb 0.2167; quiet term 0.00152; 0.50 s/step; 13.0 min spent, about 13.8 min of steps left; card peak 13.7 GB +[ 876.1s] step 1600/3200: train bpb 0.2119; quiet term 0.00150; 0.50 s/step; 13.3 min spent, about 13.3 min of steps left; card peak 13.7 GB +[ 883.5s] after 1600 steps: own rows live-minus-masked -0.71364 +- 0.00499 (0.9296 -> 0.2160); web holdout +0.00075 +- 0.00017 (the quiet bar +.012); battery in +2.024 (se 0.175; acc 0.155 -> 0.780); battery transfer +0.769 (se 0.191; acc 0.070 -> 0.280); battery held-out +1.759 (se 0.300; acc 0.050 -> 0.467); gates open mean 0.201 / widest 0.363; host RSS 4.9 GB +[ 907.0s] step 1650/3200: train bpb 0.2084; quiet term 0.00152; 0.50 s/step; 13.9 min spent, about 13.0 min of steps left; card peak 13.7 GB +[ 930.6s] step 1700/3200: train bpb 0.2078; quiet term 0.00158; 0.50 s/step; 14.3 min spent, about 12.6 min of steps left; card peak 13.7 GB +[ 954.1s] step 1750/3200: train bpb 0.2031; quiet term 0.00149; 0.50 s/step; 14.7 min spent, about 12.1 min of steps left; card peak 13.7 GB +[ 977.7s] step 1800/3200: train bpb 0.2023; quiet term 0.00138; 0.50 s/step; 15.0 min spent, about 11.7 min of steps left; card peak 13.7 GB +[ 1001.3s] step 1850/3200: train bpb 0.2002; quiet term 0.00132; 0.50 s/step; 15.4 min spent, about 11.3 min of steps left; card peak 13.7 GB +[ 1024.7s] step 1900/3200: train bpb 0.1993; quiet term 0.00132; 0.50 s/step; 15.8 min spent, about 10.8 min of steps left; card peak 13.7 GB +[ 1048.3s] step 1950/3200: train bpb 0.1972; quiet term 0.00143; 0.50 s/step; 16.2 min spent, about 10.4 min of steps left; card peak 13.7 GB +[ 1071.7s] step 2000/3200: train bpb 0.1944; quiet term 0.00127; 0.50 s/step; 16.6 min spent, about 10.0 min of steps left; card peak 13.7 GB +[ 1079.1s] after 2000 steps: own rows live-minus-masked -0.72910 +- 0.00556 (0.9296 -> 0.2005); web holdout +0.00066 +- 0.00010 (the quiet bar +.012); battery in +2.845 (se 0.145; acc 0.155 -> 0.875); battery transfer +0.930 (se 0.247; acc 0.070 -> 0.340); battery held-out +1.810 (se 0.337; acc 0.050 -> 0.433); gates open mean 0.222 / widest 0.406; host RSS 4.9 GB +[ 1102.6s] step 2050/3200: train bpb 0.1968; quiet term 0.00144; 0.50 s/step; 17.1 min spent, about 9.6 min of steps left; card peak 13.7 GB +[ 1126.1s] step 2100/3200: train bpb 0.1936; quiet term 0.00123; 0.50 s/step; 17.5 min spent, about 9.2 min of steps left; card peak 13.7 GB +[ 1149.6s] step 2150/3200: train bpb 0.1931; quiet term 0.00134; 0.50 s/step; 17.9 min spent, about 8.7 min of steps left; card peak 13.7 GB +[ 1173.2s] step 2200/3200: train bpb 0.1911; quiet term 0.00123; 0.50 s/step; 18.3 min spent, about 8.3 min of steps left; card peak 13.7 GB +[ 1196.6s] step 2250/3200: train bpb 0.1908; quiet term 0.00116; 0.50 s/step; 18.7 min spent, about 7.9 min of steps left; card peak 13.7 GB +[ 1220.1s] step 2300/3200: train bpb 0.1888; quiet term 0.00126; 0.50 s/step; 19.1 min spent, about 7.5 min of steps left; card peak 13.7 GB +[ 1243.6s] step 2350/3200: train bpb 0.1892; quiet term 0.00125; 0.50 s/step; 19.5 min spent, about 7.0 min of steps left; card peak 13.7 GB +[ 1267.1s] step 2400/3200: train bpb 0.1865; quiet term 0.00137; 0.50 s/step; 19.9 min spent, about 6.6 min of steps left; card peak 13.7 GB +[ 1274.5s] after 2400 steps: own rows live-minus-masked -0.73745 +- 0.00554 (0.9296 -> 0.1922); web holdout +0.00048 +- 0.00015 (the quiet bar +.012); battery in +2.926 (se 0.166; acc 0.155 -> 0.900); battery transfer +0.682 (se 0.251; acc 0.070 -> 0.250); battery held-out +1.170 (se 0.320; acc 0.050 -> 0.392); gates open mean 0.239 / widest 0.440; host RSS 4.9 GB +[ 1298.1s] step 2450/3200: train bpb 0.1849; quiet term 0.00114; 0.50 s/step; 20.4 min spent, about 6.2 min of steps left; card peak 13.7 GB +[ 1322.4s] step 2500/3200: train bpb 0.1837; quiet term 0.00113; 0.50 s/step; 20.8 min spent, about 5.8 min of steps left; card peak 13.7 GB +[ 1346.8s] step 2550/3200: train bpb 0.1871; quiet term 0.00116; 0.50 s/step; 21.2 min spent, about 5.4 min of steps left; card peak 13.7 GB +[ 1370.8s] step 2600/3200: train bpb 0.1826; quiet term 0.00117; 0.50 s/step; 21.6 min spent, about 5.0 min of steps left; card peak 13.7 GB +[ 1395.0s] step 2650/3200: train bpb 0.1832; quiet term 0.00122; 0.50 s/step; 22.0 min spent, about 4.6 min of steps left; card peak 13.7 GB +[ 1419.0s] step 2700/3200: train bpb 0.1838; quiet term 0.00119; 0.50 s/step; 22.4 min spent, about 4.1 min of steps left; card peak 13.7 GB +[ 1443.2s] step 2750/3200: train bpb 0.1813; quiet term 0.00135; 0.50 s/step; 22.8 min spent, about 3.7 min of steps left; card peak 13.7 GB +[ 1467.1s] step 2800/3200: train bpb 0.1805; quiet term 0.00112; 0.50 s/step; 23.2 min spent, about 3.3 min of steps left; card peak 13.7 GB +[ 1474.6s] after 2800 steps: own rows live-minus-masked -0.74298 +- 0.00579 (0.9296 -> 0.1866); web holdout +0.00051 +- 0.00013 (the quiet bar +.012); battery in +3.248 (se 0.170; acc 0.155 -> 0.885); battery transfer +0.853 (se 0.167; acc 0.070 -> 0.310); battery held-out +1.786 (se 0.294; acc 0.050 -> 0.442); gates open mean 0.253 / widest 0.465; host RSS 4.9 GB +[ 1498.7s] step 2850/3200: train bpb 0.1820; quiet term 0.00103; 0.50 s/step; 23.7 min spent, about 2.9 min of steps left; card peak 13.7 GB +[ 1522.9s] step 2900/3200: train bpb 0.1797; quiet term 0.00111; 0.50 s/step; 24.1 min spent, about 2.5 min of steps left; card peak 13.7 GB +[ 1546.8s] step 2950/3200: train bpb 0.1803; quiet term 0.00119; 0.50 s/step; 24.5 min spent, about 2.1 min of steps left; card peak 13.7 GB +[ 1570.9s] step 3000/3200: train bpb 0.1805; quiet term 0.00107; 0.50 s/step; 24.9 min spent, about 1.7 min of steps left; card peak 13.7 GB +[ 1594.9s] step 3050/3200: train bpb 0.1816; quiet term 0.00106; 0.50 s/step; 25.3 min spent, about 1.2 min of steps left; card peak 13.7 GB +[ 1619.0s] step 3100/3200: train bpb 0.1785; quiet term 0.00107; 0.50 s/step; 25.7 min spent, about 0.8 min of steps left; card peak 13.7 GB +[ 1643.0s] step 3150/3200: train bpb 0.1790; quiet term 0.00110; 0.50 s/step; 26.1 min spent, about 0.4 min of steps left; card peak 13.7 GB +[ 1667.2s] step 3200/3200: train bpb 0.1788; quiet term 0.00109; 0.50 s/step; 26.5 min spent, about 0.0 min of steps left; card peak 13.7 GB +[ 1711.6s] after 3200 steps: own rows live-minus-masked -0.74538 +- 0.00594 (0.9296 -> 0.1842); web holdout +0.00051 +- 0.00014 (the quiet bar +.012); battery in +3.647 (se 0.220; acc 0.155 -> 0.835); battery transfer +0.898 (se 0.232; acc 0.070 -> 0.330); battery held-out +1.907 (se 0.381; acc 0.050 -> 0.475); probes mean 0.411 -> 0.415; gates open mean 0.266 / widest 0.488; host RSS 4.9 GB +[ 1711.7s] anchor saved: /root/refit/long/anchors/s3_rules_fresh_seed3_d20261005_step3200.safetensors +[arms] adapter chain compile OFF on 32 wrapped blocks (1 arms) +[ 1715.5s] the boundary gauge: web holdout delta +0.00041, own rows delta -0.74536, write own 3.37e-02 / neutral 6.03e-03 = selectivity 5.58x +[ 1715.5s] DONE s3_rules_fresh_seed3_d20261005: descended True (train bpb 0.9177 -> 0.1801); 1715 s; card peak 13.7 GB; host RSS 4.9 GB diff --git a/mini-beatrix-3/arm_refit/logs/long_card1.log b/mini-beatrix-3/arm_refit/logs/long_card1.log index 11bd7b4bc34da6bdf271ec40ebaa9a81b652fdb9..0b519ea7e1c5e99ec7c1fab7a9de4b3c5452c0e1 100644 --- a/mini-beatrix-3/arm_refit/logs/long_card1.log +++ b/mini-beatrix-3/arm_refit/logs/long_card1.log @@ -23,3 +23,65 @@ Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF [ 340.0s] step 550/3200: train bpb 0.5183; quiet term 0.00137; 0.50 s/step; 4.6 min spent, about 22.3 min of steps left; card peak 13.7 GB [ 363.9s] step 600/3200: train bpb 0.4861; quiet term 0.00132; 0.50 s/step; 5.0 min spent, about 21.8 min of steps left; card peak 13.7 GB [ 388.2s] step 650/3200: train bpb 0.4539; quiet term 0.00152; 0.50 s/step; 5.4 min spent, about 21.3 min of steps left; card peak 13.7 GB +[ 412.0s] step 700/3200: train bpb 0.4298; quiet term 0.00139; 0.50 s/step; 5.8 min spent, about 20.8 min of steps left; card peak 13.7 GB +[ 436.2s] step 750/3200: train bpb 0.4136; quiet term 0.00133; 0.50 s/step; 6.2 min spent, about 20.3 min of steps left; card peak 13.7 GB +[ 460.2s] step 800/3200: train bpb 0.4001; quiet term 0.00135; 0.50 s/step; 6.6 min spent, about 19.9 min of steps left; card peak 13.7 GB +[ 463.7s] after 800 steps: own rows live-minus-masked -0.84495 +- 0.00475 (1.2339 -> 0.3889); web holdout +0.00069 +- 0.00023 (the quiet bar +.012); battery in +0.829 (se 0.159; acc 0.344 -> 0.487); battery transfer -0.042 (se 0.035; acc 0.242 -> 0.242); battery held-out -0.201 (se 0.090; acc 0.350 -> 0.312); gates open mean 0.115 / widest 0.157; host RSS 5.0 GB +[ 488.0s] step 850/3200: train bpb 0.3906; quiet term 0.00138; 0.50 s/step; 7.1 min spent, about 19.6 min of steps left; card peak 13.7 GB +[ 512.2s] step 900/3200: train bpb 0.3794; quiet term 0.00128; 0.50 s/step; 7.5 min spent, about 19.1 min of steps left; card peak 13.7 GB +[ 535.9s] step 950/3200: train bpb 0.3858; quiet term 0.00135; 0.50 s/step; 7.9 min spent, about 18.7 min of steps left; card peak 13.7 GB +[ 560.0s] step 1000/3200: train bpb 0.3768; quiet term 0.00123; 0.50 s/step; 8.3 min spent, about 18.2 min of steps left; card peak 13.7 GB +[ 584.0s] step 1050/3200: train bpb 0.3615; quiet term 0.00124; 0.50 s/step; 8.7 min spent, about 17.8 min of steps left; card peak 13.7 GB +[ 608.1s] step 1100/3200: train bpb 0.3665; quiet term 0.00126; 0.50 s/step; 9.1 min spent, about 17.4 min of steps left; card peak 13.7 GB +[ 632.0s] step 1150/3200: train bpb 0.3583; quiet term 0.00123; 0.50 s/step; 9.5 min spent, about 16.9 min of steps left; card peak 13.7 GB +[ 656.1s] step 1200/3200: train bpb 0.3504; quiet term 0.00121; 0.49 s/step; 9.9 min spent, about 16.5 min of steps left; card peak 13.7 GB +[ 659.6s] after 1200 steps: own rows live-minus-masked -0.89124 +- 0.00548 (1.2339 -> 0.3426); web holdout +0.00057 +- 0.00028 (the quiet bar +.012); battery in +1.523 (se 0.262; acc 0.344 -> 0.556); battery transfer -0.114 (se 0.048; acc 0.242 -> 0.237); battery held-out -0.294 (se 0.108; acc 0.350 -> 0.263); gates open mean 0.139 / widest 0.206; host RSS 5.0 GB +[ 683.5s] step 1250/3200: train bpb 0.3512; quiet term 0.00119; 0.50 s/step; 10.3 min spent, about 16.1 min of steps left; card peak 13.7 GB +[ 707.6s] step 1300/3200: train bpb 0.3427; quiet term 0.00124; 0.50 s/step; 10.8 min spent, about 15.7 min of steps left; card peak 13.7 GB +[ 731.7s] step 1350/3200: train bpb 0.3406; quiet term 0.00125; 0.50 s/step; 11.2 min spent, about 15.3 min of steps left; card peak 13.7 GB +[ 756.0s] step 1400/3200: train bpb 0.3449; quiet term 0.00137; 0.50 s/step; 11.6 min spent, about 14.9 min of steps left; card peak 13.7 GB +[ 780.3s] step 1450/3200: train bpb 0.3399; quiet term 0.00137; 0.50 s/step; 12.0 min spent, about 14.4 min of steps left; card peak 13.7 GB +[ 804.4s] step 1500/3200: train bpb 0.3316; quiet term 0.00121; 0.49 s/step; 12.4 min spent, about 14.0 min of steps left; card peak 13.7 GB +[ 828.8s] step 1550/3200: train bpb 0.3320; quiet term 0.00118; 0.49 s/step; 12.8 min spent, about 13.6 min of steps left; card peak 13.7 GB +[ 852.8s] step 1600/3200: train bpb 0.3266; quiet term 0.00113; 0.49 s/step; 13.2 min spent, about 13.2 min of steps left; card peak 13.7 GB +[ 856.4s] after 1600 steps: own rows live-minus-masked -0.91340 +- 0.00626 (1.2339 -> 0.3205); web holdout +0.00047 +- 0.00022 (the quiet bar +.012); battery in +2.786 (se 0.412; acc 0.344 -> 0.613); battery transfer -0.115 (se 0.046; acc 0.242 -> 0.229); battery held-out -0.353 (se 0.129; acc 0.350 -> 0.275); gates open mean 0.158 / widest 0.263; host RSS 5.0 GB +[ 880.5s] step 1650/3200: train bpb 0.3237; quiet term 0.00118; 0.50 s/step; 13.6 min spent, about 12.8 min of steps left; card peak 13.7 GB +[ 904.3s] step 1700/3200: train bpb 0.3263; quiet term 0.00125; 0.50 s/step; 14.0 min spent, about 12.4 min of steps left; card peak 13.7 GB +[ 928.3s] step 1750/3200: train bpb 0.3181; quiet term 0.00113; 0.49 s/step; 14.4 min spent, about 12.0 min of steps left; card peak 13.7 GB +[ 952.1s] step 1800/3200: train bpb 0.3144; quiet term 0.00110; 0.49 s/step; 14.8 min spent, about 11.5 min of steps left; card peak 13.7 GB +[ 976.1s] step 1850/3200: train bpb 0.3161; quiet term 0.00119; 0.49 s/step; 15.2 min spent, about 11.1 min of steps left; card peak 13.7 GB +[ 1000.0s] step 1900/3200: train bpb 0.3157; quiet term 0.00114; 0.49 s/step; 15.6 min spent, about 10.7 min of steps left; card peak 13.7 GB +[ 1023.7s] step 1950/3200: train bpb 0.3106; quiet term 0.00115; 0.49 s/step; 16.0 min spent, about 10.3 min of steps left; card peak 13.7 GB +[ 1047.6s] step 2000/3200: train bpb 0.3096; quiet term 0.00102; 0.49 s/step; 16.4 min spent, about 9.9 min of steps left; card peak 13.7 GB +[ 1051.1s] after 2000 steps: own rows live-minus-masked -0.93337 +- 0.00686 (1.2339 -> 0.3005); web holdout +0.00045 +- 0.00026 (the quiet bar +.012); battery in +3.839 (se 0.561; acc 0.344 -> 0.650); battery transfer -0.224 (se 0.054; acc 0.242 -> 0.233); battery held-out -0.282 (se 0.117; acc 0.350 -> 0.275); gates open mean 0.176 / widest 0.326; host RSS 5.0 GB +[ 1075.0s] step 2050/3200: train bpb 0.3119; quiet term 0.00108; 0.49 s/step; 16.9 min spent, about 9.5 min of steps left; card peak 13.7 GB +[ 1099.0s] step 2100/3200: train bpb 0.3094; quiet term 0.00104; 0.49 s/step; 17.3 min spent, about 9.0 min of steps left; card peak 13.7 GB +[ 1122.9s] step 2150/3200: train bpb 0.3081; quiet term 0.00110; 0.49 s/step; 17.7 min spent, about 8.6 min of steps left; card peak 13.7 GB +[ 1146.9s] step 2200/3200: train bpb 0.3028; quiet term 0.00103; 0.49 s/step; 18.1 min spent, about 8.2 min of steps left; card peak 13.7 GB +[ 1170.7s] step 2250/3200: train bpb 0.3037; quiet term 0.00114; 0.49 s/step; 18.5 min spent, about 7.8 min of steps left; card peak 13.7 GB +[ 1194.7s] step 2300/3200: train bpb 0.3023; quiet term 0.00115; 0.49 s/step; 18.9 min spent, about 7.4 min of steps left; card peak 13.7 GB +[ 1218.7s] step 2350/3200: train bpb 0.2990; quiet term 0.00107; 0.49 s/step; 19.3 min spent, about 7.0 min of steps left; card peak 13.7 GB +[ 1242.5s] step 2400/3200: train bpb 0.2946; quiet term 0.00111; 0.49 s/step; 19.7 min spent, about 6.6 min of steps left; card peak 13.7 GB +[ 1246.0s] after 2400 steps: own rows live-minus-masked -0.94560 +- 0.00704 (1.2339 -> 0.2883); web holdout +0.00029 +- 0.00018 (the quiet bar +.012); battery in +3.711 (se 0.457; acc 0.344 -> 0.731); battery transfer -0.235 (se 0.057; acc 0.242 -> 0.221); battery held-out -0.519 (se 0.182; acc 0.350 -> 0.325); gates open mean 0.194 / widest 0.382; host RSS 5.0 GB +[ 1270.0s] step 2450/3200: train bpb 0.2927; quiet term 0.00106; 0.49 s/step; 20.1 min spent, about 6.2 min of steps left; card peak 13.7 GB +[ 1293.7s] step 2500/3200: train bpb 0.2983; quiet term 0.00107; 0.49 s/step; 20.5 min spent, about 5.7 min of steps left; card peak 13.7 GB +[ 1317.6s] step 2550/3200: train bpb 0.2888; quiet term 0.00111; 0.49 s/step; 20.9 min spent, about 5.3 min of steps left; card peak 13.7 GB +[ 1341.5s] step 2600/3200: train bpb 0.2874; quiet term 0.00115; 0.49 s/step; 21.3 min spent, about 4.9 min of steps left; card peak 13.7 GB +[ 1365.5s] step 2650/3200: train bpb 0.2928; quiet term 0.00104; 0.49 s/step; 21.7 min spent, about 4.5 min of steps left; card peak 13.7 GB +[ 1389.4s] step 2700/3200: train bpb 0.2906; quiet term 0.00109; 0.49 s/step; 22.1 min spent, about 4.1 min of steps left; card peak 13.7 GB +[ 1413.5s] step 2750/3200: train bpb 0.2901; quiet term 0.00103; 0.49 s/step; 22.5 min spent, about 3.7 min of steps left; card peak 13.7 GB +[ 1437.5s] step 2800/3200: train bpb 0.2843; quiet term 0.00105; 0.49 s/step; 22.9 min spent, about 3.3 min of steps left; card peak 13.7 GB +[ 1441.1s] after 2800 steps: own rows live-minus-masked -0.95475 +- 0.00691 (1.2339 -> 0.2791); web holdout +0.00032 +- 0.00019 (the quiet bar +.012); battery in +4.220 (se 0.526; acc 0.344 -> 0.762); battery transfer -0.139 (se 0.048; acc 0.242 -> 0.242); battery held-out -0.928 (se 0.234; acc 0.350 -> 0.300); gates open mean 0.210 / widest 0.425; host RSS 5.0 GB +[ 1465.0s] step 2850/3200: train bpb 0.2926; quiet term 0.00099; 0.49 s/step; 23.4 min spent, about 2.9 min of steps left; card peak 13.7 GB +[ 1489.0s] step 2900/3200: train bpb 0.2884; quiet term 0.00098; 0.49 s/step; 23.8 min spent, about 2.5 min of steps left; card peak 13.7 GB +[ 1512.9s] step 2950/3200: train bpb 0.2829; quiet term 0.00105; 0.49 s/step; 24.2 min spent, about 2.0 min of steps left; card peak 13.7 GB +[ 1536.9s] step 3000/3200: train bpb 0.2811; quiet term 0.00102; 0.49 s/step; 24.6 min spent, about 1.6 min of steps left; card peak 13.7 GB +[ 1560.7s] step 3050/3200: train bpb 0.2839; quiet term 0.00092; 0.49 s/step; 25.0 min spent, about 1.2 min of steps left; card peak 13.7 GB +[ 1584.7s] step 3100/3200: train bpb 0.2847; quiet term 0.00104; 0.49 s/step; 25.4 min spent, about 0.8 min of steps left; card peak 13.7 GB +[ 1608.4s] step 3150/3200: train bpb 0.2835; quiet term 0.00107; 0.49 s/step; 25.8 min spent, about 0.4 min of steps left; card peak 13.7 GB +[ 1632.9s] step 3200/3200: train bpb 0.2845; quiet term 0.00097; 0.49 s/step; 26.2 min spent, about 0.0 min of steps left; card peak 13.7 GB +[ 1674.5s] after 3200 steps: own rows live-minus-masked -0.96018 +- 0.00734 (1.2339 -> 0.2737); web holdout +0.00030 +- 0.00018 (the quiet bar +.012); battery in +5.676 (se 0.623; acc 0.344 -> 0.806); battery transfer -0.166 (se 0.040; acc 0.242 -> 0.237); battery held-out -0.805 (se 0.234; acc 0.350 -> 0.325); probes mean 0.411 -> 0.422; gates open mean 0.224 / widest 0.459; host RSS 5.0 GB +[ 1674.7s] anchor saved: /root/refit/long/anchors/s4_arith_fresh_seed4_d20261005_step3200.safetensors +[arms] adapter chain compile OFF on 32 wrapped blocks (1 arms) +[ 1678.5s] the boundary gauge: web holdout delta +0.00020, own rows delta -0.96023, write own 2.68e-02 / neutral 4.82e-03 = selectivity 5.56x +[ 1678.5s] DONE s4_arith_fresh_seed4_d20261005: descended True (train bpb 1.2203 -> 0.2847); 1679 s; card peak 13.7 GB; host RSS 5.0 GB diff --git a/mini-beatrix-3/arm_refit/logs/ship_refit_0458.log b/mini-beatrix-3/arm_refit/logs/ship_refit_0458.log new file mode 100644 index 0000000000000000000000000000000000000000..31a25b979da5e05cc9bdb38613c64f89a91c33a7 --- /dev/null +++ b/mini-beatrix-3/arm_refit/logs/ship_refit_0458.log @@ -0,0 +1,265 @@ +[ 0.0s] 260 files, 8.36 GB, to AbstractPhil/alephllm-mini-beatrix-training/mini-beatrix-3/arm_refit; left out by name: 168 +[ 0.0s] 52.3 MB group/anchors/gA_close1_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gA_close2_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gA_close2_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gA_close3_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gA_close3_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gA_close3_s3_rules.safetensors +[ 0.0s] 52.3 MB group/anchors/gA_close4_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gA_close4_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gA_close4_s3_rules.safetensors +[ 0.0s] 52.3 MB group/anchors/gA_close4_s4_arith.safetensors +[ 0.0s] 52.3 MB group/anchors/gB_close1_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gB_close2_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gB_close2_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gB_close3_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gB_close3_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gB_close3_s3_rules.safetensors +[ 0.0s] 52.3 MB group/anchors/gB_close4_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gB_close4_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gB_close4_s3_rules.safetensors +[ 0.0s] 52.3 MB group/anchors/gB_close4_s4_arith.safetensors +[ 0.0s] 52.3 MB group/anchors/gFA_close1_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gFA_close2_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gFA_close2_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gFA_close3_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gFA_close3_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gFA_close3_s3_rules.safetensors +[ 0.0s] 52.3 MB group/anchors/gFA_close4_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gFA_close4_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gFA_close4_s3_rules.safetensors +[ 0.0s] 52.3 MB group/anchors/gFA_close4_s4_arith.safetensors +[ 0.0s] 52.3 MB group/anchors/gFB_close1_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gFB_close2_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gFB_close2_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gFB_close3_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gFB_close3_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gFB_close3_s3_rules.safetensors +[ 0.0s] 52.3 MB group/anchors/gFB_close4_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gFB_close4_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gFB_close4_s3_rules.safetensors +[ 0.0s] 52.3 MB group/anchors/gFB_close4_s4_arith.safetensors +[ 0.0s] 52.3 MB group/anchors/gRA_close3_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gRA_close3_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gRA_close3_s3_rules.safetensors +[ 0.0s] 52.3 MB group/anchors/gRA_close4_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gRA_close4_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gRA_close4_s3_rules.safetensors +[ 0.0s] 52.3 MB group/anchors/gRA_close4_s4_arith.safetensors +[ 0.0s] 52.3 MB group/anchors/gRB_close3_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gRB_close3_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gRB_close3_s3_rules.safetensors +[ 0.0s] 52.3 MB group/anchors/gRB_close4_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gRB_close4_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gRB_close4_s3_rules.safetensors +[ 0.0s] 52.3 MB group/anchors/gRB_close4_s4_arith.safetensors +[ 0.0s] 52.3 MB group/anchors/gSA_close1_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gSA_close2_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gSA_close2_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gSA_close3_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gSA_close3_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gSA_close3_s3_rules.safetensors +[ 0.0s] 52.3 MB group/anchors/gSA_close4_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gSA_close4_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gSA_close4_s3_rules.safetensors +[ 0.0s] 52.3 MB group/anchors/gSA_close4_s4_arith.safetensors +[ 0.0s] 52.3 MB group/anchors/gSB_close1_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gSB_close2_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gSB_close2_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gSB_close3_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gSB_close3_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gSB_close3_s3_rules.safetensors +[ 0.0s] 52.3 MB group/anchors/gSB_close4_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gSB_close4_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gSB_close4_s3_rules.safetensors +[ 0.0s] 52.3 MB group/anchors/gSB_close4_s4_arith.safetensors +[ 0.0s] 52.3 MB group/anchors/gXA_close5_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gXA_close5_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gXA_close5_s3_rules.safetensors +[ 0.0s] 52.3 MB group/anchors/gXA_close5_s4_arith.safetensors +[ 0.0s] 52.3 MB group/anchors/gXA_close5_s5_causal.safetensors +[ 0.0s] 52.3 MB group/anchors/gXA_close6_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gXA_close6_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gXA_close6_s3_rules.safetensors +[ 0.0s] 52.3 MB group/anchors/gXA_close6_s4_arith.safetensors +[ 0.0s] 52.3 MB group/anchors/gXA_close6_s5_causal.safetensors +[ 0.0s] 52.3 MB group/anchors/gXA_close6_s6_tryfail.safetensors +[ 0.0s] 52.3 MB group/anchors/gXA_close7_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gXA_close7_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gXA_close7_s3_rules.safetensors +[ 0.0s] 52.3 MB group/anchors/gXA_close7_s4_arith.safetensors +[ 0.0s] 52.3 MB group/anchors/gXA_close7_s5_causal.safetensors +[ 0.0s] 52.3 MB group/anchors/gXA_close7_s6_tryfail.safetensors +[ 0.0s] 52.3 MB group/anchors/gXA_close7_s7_mixed.safetensors +[ 0.0s] 52.3 MB group/anchors/gXA_close8_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gXA_close8_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gXA_close8_s3_rules.safetensors +[ 0.0s] 52.3 MB group/anchors/gXA_close8_s4_arith.safetensors +[ 0.0s] 52.3 MB group/anchors/gXA_close8_s5_causal.safetensors +[ 0.0s] 52.3 MB group/anchors/gXA_close8_s6_tryfail.safetensors +[ 0.0s] 52.3 MB group/anchors/gXA_close8_s7_mixed.safetensors +[ 0.0s] 52.3 MB group/anchors/gXA_close8_s8_register.safetensors +[ 0.0s] 52.3 MB group/anchors/gXB_close5_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gXB_close5_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gXB_close5_s3_rules.safetensors +[ 0.0s] 52.3 MB group/anchors/gXB_close5_s4_arith.safetensors +[ 0.0s] 52.3 MB group/anchors/gXB_close5_s5_causal.safetensors +[ 0.0s] 52.3 MB group/anchors/gXB_close6_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gXB_close6_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gXB_close6_s3_rules.safetensors +[ 0.0s] 52.3 MB group/anchors/gXB_close6_s4_arith.safetensors +[ 0.0s] 52.3 MB group/anchors/gXB_close6_s5_causal.safetensors +[ 0.0s] 52.3 MB group/anchors/gXB_close6_s6_tryfail.safetensors +[ 0.0s] 52.3 MB group/anchors/gXB_close7_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gXB_close7_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gXB_close7_s3_rules.safetensors +[ 0.0s] 52.3 MB group/anchors/gXB_close7_s4_arith.safetensors +[ 0.0s] 52.3 MB group/anchors/gXB_close7_s5_causal.safetensors +[ 0.0s] 52.3 MB group/anchors/gXB_close7_s6_tryfail.safetensors +[ 0.0s] 52.3 MB group/anchors/gXB_close7_s7_mixed.safetensors +[ 0.0s] 52.3 MB group/anchors/gXB_close8_s1_perspective.safetensors +[ 0.0s] 52.3 MB group/anchors/gXB_close8_s2_concept.safetensors +[ 0.0s] 52.3 MB group/anchors/gXB_close8_s3_rules.safetensors +[ 0.0s] 52.3 MB group/anchors/gXB_close8_s4_arith.safetensors +[ 0.0s] 52.3 MB group/anchors/gXB_close8_s5_causal.safetensors +[ 0.0s] 52.3 MB group/anchors/gXB_close8_s6_tryfail.safetensors +[ 0.0s] 52.3 MB group/anchors/gXB_close8_s7_mixed.safetensors +[ 0.0s] 52.3 MB group/anchors/gXB_close8_s8_register.safetensors +[ 0.0s] 8.1 MB group/anchors/routed_gRA.dispatch.pt +[ 0.0s] 8.1 MB group/anchors/routed_gRB.dispatch.pt +[ 0.0s] 8.1 MB group/anchors/routed_gSA.dispatch.pt +[ 0.0s] 8.1 MB group/anchors/routed_gSB.dispatch.pt +[ 0.0s] 0.1 MB group/results/group_gA.json +[ 0.0s] 0.1 MB group/results/group_gB.json +[ 0.0s] 0.1 MB group/results/group_gFA.json +[ 0.0s] 0.1 MB group/results/group_gFB.json +[ 0.0s] 0.1 MB group/results/group_gRA.json +[ 0.0s] 0.1 MB group/results/group_gRB.json +[ 0.0s] 0.1 MB group/results/group_gSA.json +[ 0.0s] 0.1 MB group/results/group_gSB.json +[ 0.0s] 0.3 MB group/results/group_gXA.json +[ 0.0s] 0.3 MB group/results/group_gXB.json +[ 0.0s] 0.0 MB group/results/lines_gX_close8.json +[ 0.0s] 0.0 MB group/results/routed_gRA.json +[ 0.0s] 0.0 MB group/results/routed_gRB.json +[ 0.0s] 0.0 MB group/results/routed_gSA.json +[ 0.0s] 0.0 MB group/results/routed_gSB.json +[ 0.0s] 52.3 MB solo/anchors/pc_s1_fresh_seed1_d20260930_step400.safetensors +[ 0.0s] 52.3 MB solo/anchors/s1_perspective_anchor_seed1_d20261005_step400.safetensors +[ 0.0s] 52.3 MB solo/anchors/s1_perspective_anchor_seed1_d20261005_step800.safetensors +[ 0.0s] 52.3 MB solo/anchors/s1_perspective_fresh_seed101_d20261006_step400.safetensors +[ 0.0s] 52.3 MB solo/anchors/s1_perspective_fresh_seed101_d20261006_step800.safetensors +[ 0.0s] 52.3 MB solo/anchors/s1_perspective_fresh_seed1_d20261005_step400.safetensors +[ 0.0s] 52.3 MB solo/anchors/s1_perspective_fresh_seed1_d20261005_step800.safetensors +[ 0.0s] 52.3 MB solo/anchors/s2_concept_anchor_seed2_d20261005_step400.safetensors +[ 0.0s] 52.3 MB solo/anchors/s2_concept_anchor_seed2_d20261005_step800.safetensors +[ 0.0s] 52.3 MB solo/anchors/s2_concept_fresh_seed102_d20261006_step400.safetensors +[ 0.0s] 52.3 MB solo/anchors/s2_concept_fresh_seed102_d20261006_step800.safetensors +[ 0.0s] 52.3 MB solo/anchors/s2_concept_fresh_seed2_d20261005_step400.safetensors +[ 0.0s] 52.3 MB solo/anchors/s2_concept_fresh_seed2_d20261005_step800.safetensors +[ 0.0s] 52.3 MB solo/anchors/s3_rules_anchor_seed3_d20261005_step400.safetensors +[ 0.0s] 52.3 MB solo/anchors/s3_rules_anchor_seed3_d20261005_step800.safetensors +[ 0.0s] 52.3 MB solo/anchors/s3_rules_fresh_seed103_d20261006_step400.safetensors +[ 0.0s] 52.3 MB solo/anchors/s3_rules_fresh_seed103_d20261006_step800.safetensors +[ 0.0s] 52.3 MB solo/anchors/s3_rules_fresh_seed3_d20261005_step400.safetensors +[ 0.0s] 52.3 MB solo/anchors/s3_rules_fresh_seed3_d20261005_step800.safetensors +[ 0.0s] 52.3 MB solo/anchors/s4_arith_anchor_seed4_d20261005_step400.safetensors +[ 0.0s] 52.3 MB solo/anchors/s4_arith_anchor_seed4_d20261005_step800.safetensors +[ 0.0s] 52.3 MB solo/anchors/s4_arith_fresh_seed104_d20261006_step400.safetensors +[ 0.0s] 52.3 MB solo/anchors/s4_arith_fresh_seed104_d20261006_step800.safetensors +[ 0.0s] 52.3 MB solo/anchors/s4_arith_fresh_seed4_d20261005_step400.safetensors +[ 0.0s] 52.3 MB solo/anchors/s4_arith_fresh_seed4_d20261005_step800.safetensors +[ 0.0s] 52.3 MB solo/anchors/s5_causal_fresh_seed5_d20261005_step400.safetensors +[ 0.0s] 52.3 MB solo/anchors/s5_causal_fresh_seed5_d20261005_step800.safetensors +[ 0.0s] 52.3 MB solo/anchors/s6_tryfail_fresh_seed6_d20261005_step400.safetensors +[ 0.0s] 52.3 MB solo/anchors/s6_tryfail_fresh_seed6_d20261005_step800.safetensors +[ 0.0s] 52.3 MB solo/anchors/s7_mixed_fresh_seed7_d20261005_step400.safetensors +[ 0.0s] 52.3 MB solo/anchors/s7_mixed_fresh_seed7_d20261005_step800.safetensors +[ 0.0s] 52.3 MB solo/anchors/s8_register_fresh_seed8_d20261005_step400.safetensors +[ 0.0s] 52.3 MB solo/anchors/s8_register_fresh_seed8_d20261005_step800.safetensors +[ 0.0s] 0.0 MB solo/results/result_pc_s1_fresh_seed1_d20260930.json +[ 0.0s] 0.0 MB solo/results/result_s1_perspective_anchor_seed1_d20261005.json +[ 0.0s] 0.0 MB solo/results/result_s1_perspective_fresh_seed101_d20261006.json +[ 0.0s] 0.0 MB solo/results/result_s1_perspective_fresh_seed1_d20261005.json +[ 0.0s] 0.0 MB solo/results/result_s2_concept_anchor_seed2_d20261005.json +[ 0.0s] 0.0 MB solo/results/result_s2_concept_fresh_seed102_d20261006.json +[ 0.0s] 0.0 MB solo/results/result_s2_concept_fresh_seed2_d20261005.json +[ 0.0s] 0.0 MB solo/results/result_s3_rules_anchor_seed3_d20261005.json +[ 0.0s] 0.0 MB solo/results/result_s3_rules_fresh_seed103_d20261006.json +[ 0.0s] 0.0 MB solo/results/result_s3_rules_fresh_seed3_d20261005.json +[ 0.0s] 0.0 MB solo/results/result_s4_arith_anchor_seed4_d20261005.json +[ 0.0s] 0.0 MB solo/results/result_s4_arith_fresh_seed104_d20261006.json +[ 0.0s] 0.0 MB solo/results/result_s4_arith_fresh_seed4_d20261005.json +[ 0.0s] 0.0 MB solo/results/result_s5_causal_fresh_seed5_d20261005.json +[ 0.0s] 0.0 MB solo/results/result_s6_tryfail_fresh_seed6_d20261005.json +[ 0.0s] 0.0 MB solo/results/result_s7_mixed_fresh_seed7_d20261005.json +[ 0.0s] 0.0 MB solo/results/result_s8_register_fresh_seed8_d20261005.json +[ 0.0s] 52.3 MB solo_long/anchors/s3_rules_fresh_seed3_d20261005_step3200.safetensors +[ 0.0s] 52.3 MB solo_long/anchors/s3_rules_fresh_seed3_d20261005_step400.safetensors +[ 0.0s] 52.3 MB solo_long/anchors/s4_arith_fresh_seed4_d20261005_step3200.safetensors +[ 0.0s] 52.3 MB solo_long/anchors/s4_arith_fresh_seed4_d20261005_step400.safetensors +[ 0.0s] 0.0 MB solo_long/results/result_s3_rules_fresh_seed3_d20261005.json +[ 0.0s] 0.0 MB solo_long/results/result_s4_arith_fresh_seed4_d20261005.json +[ 0.0s] 0.0 MB logs/chain10_card0.log +[ 0.0s] 0.0 MB logs/chain10_card1.log +[ 0.0s] 0.0 MB logs/chain11_card0.log +[ 0.0s] 0.0 MB logs/chain11_card1.log +[ 0.0s] 0.0 MB logs/chain3_card0.log +[ 0.0s] 0.0 MB logs/chain3_card1.log +[ 0.0s] 0.0 MB logs/chain4_card0.log +[ 0.0s] 0.0 MB logs/chain4_card1.log +[ 0.0s] 0.0 MB logs/chain5_card0.log +[ 0.0s] 0.0 MB logs/chain5_card1.log +[ 0.0s] 0.0 MB logs/chain6_card0.log +[ 0.0s] 0.0 MB logs/chain6_card1.log +[ 0.0s] 0.0 MB logs/chain7_card0.log +[ 0.0s] 0.0 MB logs/chain7_card1.log +[ 0.0s] 0.0 MB logs/chain8_card0.log +[ 0.0s] 0.0 MB logs/chain8_card1.log +[ 0.0s] 0.0 MB logs/chain9_card0.log +[ 0.0s] 0.0 MB logs/chain9_card1.log +[ 0.0s] 0.0 MB logs/group_A.log +[ 0.0s] 0.0 MB logs/group_B.log +[ 0.0s] 0.0 MB logs/group_EA.log +[ 0.0s] 0.0 MB logs/group_EB.log +[ 0.0s] 0.0 MB logs/group_FA.log +[ 0.0s] 0.0 MB logs/group_FB.log +[ 0.0s] 0.0 MB logs/group_RA.log +[ 0.0s] 0.0 MB logs/group_RB.log +[ 0.0s] 0.0 MB logs/group_SA.log +[ 0.0s] 0.0 MB logs/group_SB.log +[ 0.0s] 0.1 MB logs/group_XA.log +[ 0.0s] 0.1 MB logs/group_XB.log +[ 0.0s] 0.0 MB logs/group_runs.log +[ 0.0s] 0.0 MB logs/long_card0.log +[ 0.0s] 0.0 MB logs/long_card1.log +[ 0.0s] 0.0 MB logs/queueB_card0.log +[ 0.0s] 0.0 MB logs/queueB_card1.log +[ 0.0s] 0.0 MB logs/queue_card0.log +[ 0.0s] 0.0 MB logs/queue_card1.log +[ 0.0s] 0.0 MB logs/routed_gRA.log +[ 0.0s] 0.0 MB logs/routed_gRB.log +[ 0.0s] 0.0 MB logs/routed_gSA.log +[ 0.0s] 0.0 MB logs/routed_gSB.log +[ 0.0s] 0.0 MB logs/runB_card0_01_arms3_rules_seed3_dataseed20261005.log +[ 0.0s] 0.0 MB logs/runB_card0_02_arms1_perspective_initanchor_seed1_dataseed20261005.log +[ 0.0s] 0.0 MB logs/runB_card0_03_arms3_rules_initanchor_seed3_dataseed20261005.log +[ 0.0s] 0.0 MB logs/runB_card0_04_arms5_causal_seed5_dataseed20261005.log +[ 0.0s] 0.0 MB logs/runB_card0_05_arms7_mixed_seed7_dataseed20261005.log +[ 0.0s] 0.0 MB logs/runB_card1_01_arms4_arith_seed4_dataseed20261005.log +[ 0.0s] 0.0 MB logs/runB_card1_02_arms2_concept_initanchor_seed2_dataseed20261005.log +[ 0.0s] 0.0 MB logs/runB_card1_03_arms4_arith_initanchor_seed4_dataseed20261005.log +[ 0.0s] 0.0 MB logs/runB_card1_04_arms6_tryfail_seed6_dataseed20261005.log +[ 0.0s] 0.0 MB logs/runB_card1_05_arms8_register_seed8_dataseed20261005.log +[ 0.0s] 0.0 MB logs/runC_card0_01.log +[ 0.0s] 0.0 MB logs/runC_card0_02.log +[ 0.0s] 0.0 MB logs/runC_card1_01.log +[ 0.0s] 0.0 MB logs/runC_card1_02.log +[ 0.0s] 0.0 MB logs/run_card0_01_armpc_s1.log +[ 0.0s] 0.0 MB logs/run_card0_02_arms1_perspective_seed1_dataseed20261005.log +[ 0.0s] 0.0 MB logs/run_card1_01_arms2_concept_seed2_dataseed20261005.log +[ 0.0s] 0.0 MB logs/ship_refit_0458.log +[ 0.0s] the folder's README: /root/refit/README_arm_refit.md +/root/venv/lib/python3.12/site-packages/huggingface_hub/hf_api.py:11754: UserWarning: Warnings while validating metadata in README.md: +- empty or missing yaml metadata in repo card + warnings.warn(f"Warnings while validating metadata in README.md:\n{message}")