mini-beatrix-3: the stage arms fitted on the finished core (weights, results, logs) (part 2)

mini-beatrix-3/arm_refit/solo_long/anchors/s3_rules_fresh_seed3_d20261005_step3200.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9ec19522dd9ad256c90c2537904da2694287c6efe37f82aee87f85695a4af4a5
+size 54818592

mini-beatrix-3/arm_refit/solo_long/anchors/s4_arith_fresh_seed4_d20261005_step3200.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0c9637b3a5e27ae2a19f517870eaecd2506ac65f6e880e486e5b6cc87a32a718
+size 54818584

mini-beatrix-3/arm_refit/solo_long/results/result_s3_rules_fresh_seed3_d20261005.json

@@ -1 +1 @@
-{"tag": "s3_rules_fresh_seed3_d20261005", "arm": "s3_rules", "phase": "curriculum_s3", "own_source": "rulechain-minted", "own_rows_held_out": true, "init": "fresh", "seed": 3, "data_seed": 20261005, "trunk_step": 245674, "steps": 3200, "ctx": 4096, "mb": 2, "lr": 0.001, "lambda": 2.0, "quiet": true, "compiled_chain": true, "replication": false, "spec": {"n_slots": 16, "K": 16, "D": 8, "hidden": 256}, "reads": [{"step": 0, "live_minus_masked_own": {"mean": 0.0, "se": 0.0, "masked": 0.9296131361815035, "live": 0.9296131361815035, "per_batch": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "live_minus_masked_fineweb": {"mean": 0.0, "se": 0.0, "masked": 0.9505967609111062, "live": 0.9505967609111062, "per_batch": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "battery_in": {"mean": 0.0, "se": 0.0, "n": 200, "acc_masked": 0.155, "acc_live": 0.155}, "battery_transfer": {"mean": 0.0, "se": 0.0, "n": 100, "acc_masked": 0.07, "acc_live": 0.07}, "battery_held-out": {"mean": 0.0, "se": 0.0, "n": 120, "acc_masked": 0.05, "acc_live": 0.05}, "gate_open_mean": 0.04742587317756678, "gate_open_max": 0.04742587317756678}, {"step": 200, "live_minus_masked_own": {"mean": -0.19916838919023563, "se": 0.0029478463818419427, "masked": 0.9296131361815035, "live": 0.7304447469912679, "per_batch": [-0.18861311522726587, -0.1942985176994887, -0.2126512163296862, -0.1898513903134782, -0.20438280642415485, -0.19834526947429054, -0.20656810397734315, -0.19863669407617757]}, "live_minus_masked_fineweb": {"mean": 0.000795924959277397, "se": 0.0002359271233041411, "masked": 0.9505967609111062, "live": 0.9513926858703836, "per_batch": [0.0015161990500065503, -0.0002265011511863868, 0.0013087879730659946, 0.00020096172753325536, 0.0010948415553926338, 0.00133355347479025, 0.0010468583958018751, 9.269864881500389e-05]}, "battery_in": {"mean": 0.22961505502462387, "se": 0.03524137425517963, "n": 200, "acc_masked": 0.155, "acc_live": 0.185}, "battery_transfer": {"mean": 0.05596338391304016, "se": 0.05618378400233961, "n": 100, "acc_masked": 0.07, "acc_live": 0.09}, "battery_held-out": {"mean": 0.0641354074080785, "se": 0.03950927132175443, "n": 120, "acc_masked": 0.05, "acc_live": 0.058333333333333334}, "gate_open_mean": 0.06336235477370164, "gate_open_max": 0.06533588989691239}, {"step": 400, "live_minus_masked_own": {"mean": -0.3492465664661437, "se": 0.004393077931582688, "masked": 0.9296131361815035, "live": 0.5803665697153597, "per_batch": [-0.337419900007755, -0.3384932437317899, -0.36666486185637226, -0.3324088415095655, -0.3545348394996929, -0.348017814919223, -0.35332029801929976, -0.3631127321854517]}, "live_minus_masked_fineweb": {"mean": 0.0006639712704029355, "se": 0.00015820478632910145, "masked": 0.9505967609111062, "live": 0.9512607321815091, "per_batch": [0.0014807706239288176, 0.0005778617068989966, 0.0008724679878270303, 0.00017705613906326967, 0.0003320125074908198, 0.0010005950627198779, 0.0006789875056063543, 0.0001920186296883175]}, "battery_in": {"mean": 0.5217548792064189, "se": 0.06131239265702742, "n": 200, "acc_masked": 0.155, "acc_live": 0.295}, "battery_transfer": {"mean": 0.06662610709667206, "se": 0.09284188636615415, "n": 100, "acc_masked": 0.07, "acc_live": 0.06}, "battery_held-out": {"mean": 0.1739781285325686, "se": 0.08293881820513324, "n": 120, "acc_masked": 0.05, "acc_live": 0.09166666666666666}, "gate_open_mean": 0.08292569250755885, "gate_open_max": 0.09417776196343719}], "inert_at_birth": true, "seconds": 293.88527607917786, "log": ["[    0.2s] THE JOB: s3_rules_fresh_seed3_d20261005: 3200 steps of 2 x 4096 bytes, reads at [0, 200, 400, 800, 1200, 1600, 2000, 2400, 2800, 3200]; quiet term on (lambda 2.0); adapter chain compiled; trunk step 245,674; card 31.4 GB capped at 0.92; host RSS 0.6 GB", "[   19.2s] mini-beatrix-3 @ 245,674: trunk loaded from /root/v3/runs/mission/mini-beatrix-3/checkpoints/step_00245674.safetensors and frozen (376.0M)", "[   31.5s] own rows (rulechain-minted; the hub tensors): 'If someone is hoream, then they are hortil. If someone is hortil, then they are babig. If someone is babig, then they are banng. Vin is hoream. So Vin is hortil. So Vin i'", "[   66.2s] THE BARE TRUNK (the arm masked): own rows 0.9296 bpb, web holdout 0.9506 bpb; battery acc in 0.155 (n=200), transfer 0.070 (n=100), held-out 0.050 (n=120); probes mean 0.411", "[   75.1s] at birth: own rows live-minus-masked +0.00000 +- 0.00000 (0.9296 -> 0.9296); web holdout +0.00000 +- 0.00000 (the quiet bar +.012); battery in +0.000 (se 0.000; acc 0.155 -> 0.155); battery transfer +0.000 (se 0.000; acc 0.070 -> 0.070); battery held-out +0.000 (se 0.000; acc 0.050 -> 0.050); gates open mean 0.047 / widest 0.047; host RSS 2.3 GB", "[   91.0s] step 1/3200: train bpb 0.9197; quiet term 0.00000; 15.85 s/step; 0.3 min spent, about 845.3 min of steps left; card peak 13.6 GB", "[  114.4s] step 50/3200: train bpb 0.9050; quiet term 0.00047; 0.79 s/step; 0.7 min spent, about 41.3 min of steps left; card peak 13.7 GB", "[  138.0s] step 100/3200: train bpb 0.8557; quiet term 0.00084; 0.63 s/step; 1.0 min spent, about 32.5 min of steps left; card peak 13.7 GB", "[  161.4s] step 150/3200: train bpb 0.8053; quiet term 0.00084; 0.58 s/step; 1.4 min spent, about 29.2 min of steps left; card peak 13.7 GB", "[  184.9s] step 200/3200: train bpb 0.7506; quiet term 0.00109; 0.55 s/step; 1.8 min spent, about 27.4 min of steps left; card peak 13.7 GB", "[  192.3s] after 200 steps: own rows live-minus-masked -0.19917 +- 0.00295 (0.9296 -> 0.7304); web holdout +0.00080 +- 0.00024 (the quiet bar +.012); battery in +0.230 (se 0.035; acc 0.155 -> 0.185); battery transfer +0.056 (se 0.056; acc 0.070 -> 0.090); battery held-out +0.064 (se 0.040; acc 0.050 -> 0.058); gates open mean 0.063 / widest 0.065; host RSS 4.9 GB", "[  215.8s] step 250/3200: train bpb 0.6934; quiet term 0.00151; 0.56 s/step; 2.3 min spent, about 27.7 min of steps left; card peak 13.7 GB", "[  239.6s] step 300/3200: train bpb 0.6492; quiet term 0.00137; 0.55 s/step; 2.7 min spent, about 26.5 min of steps left; card peak 13.7 GB", "[  263.1s] step 350/3200: train bpb 0.6121; quiet term 0.00128; 0.54 s/step; 3.1 min spent, about 25.5 min of steps left; card peak 13.7 GB", "[  286.3s] step 400/3200: train bpb 0.5784; quiet term 0.00130; 0.53 s/step; 3.5 min spent, about 24.6 min of steps left; card peak 13.7 GB", "[  293.7s] after 400 steps: own rows live-minus-masked -0.34925 +- 0.00439 (0.9296 -> 0.5804); web holdout +0.00066 +- 0.00016 (the quiet bar +.012); battery in +0.522 (se 0.061; acc 0.155 -> 0.295); battery transfer +0.067 (se 0.093; acc 0.070 -> 0.060); battery held-out +0.174 (se 0.083; acc 0.050 -> 0.092); gates open mean 0.083 / widest 0.094; host RSS 4.9 GB", "[  293.9s] anchor saved: /root/refit/long/anchors/s3_rules_fresh_seed3_d20261005_step400.safetensors"]}
+{"tag": "s3_rules_fresh_seed3_d20261005", "arm": "s3_rules", "phase": "curriculum_s3", "own_source": "rulechain-minted", "own_rows_held_out": true, "init": "fresh", "seed": 3, "data_seed": 20261005, "trunk_step": 245674, "steps": 3200, "ctx": 4096, "mb": 2, "lr": 0.001, "lambda": 2.0, "quiet": true, "compiled_chain": true, "replication": false, "spec": {"n_slots": 16, "K": 16, "D": 8, "hidden": 256}, "reads": [{"step": 0, "live_minus_masked_own": {"mean": 0.0, "se": 0.0, "masked": 0.9296131361815035, "live": 0.9296131361815035, "per_batch": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "live_minus_masked_fineweb": {"mean": 0.0, "se": 0.0, "masked": 0.9505967609111062, "live": 0.9505967609111062, "per_batch": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "battery_in": {"mean": 0.0, "se": 0.0, "n": 200, "acc_masked": 0.155, "acc_live": 0.155}, "battery_transfer": {"mean": 0.0, "se": 0.0, "n": 100, "acc_masked": 0.07, "acc_live": 0.07}, "battery_held-out": {"mean": 0.0, "se": 0.0, "n": 120, "acc_masked": 0.05, "acc_live": 0.05}, "gate_open_mean": 0.04742587317756678, "gate_open_max": 0.04742587317756678}, {"step": 200, "live_minus_masked_own": {"mean": -0.19916838919023563, "se": 0.0029478463818419427, "masked": 0.9296131361815035, "live": 0.7304447469912679, "per_batch": [-0.18861311522726587, -0.1942985176994887, -0.2126512163296862, -0.1898513903134782, -0.20438280642415485, -0.19834526947429054, -0.20656810397734315, -0.19863669407617757]}, "live_minus_masked_fineweb": {"mean": 0.000795924959277397, "se": 0.0002359271233041411, "masked": 0.9505967609111062, "live": 0.9513926858703836, "per_batch": [0.0015161990500065503, -0.0002265011511863868, 0.0013087879730659946, 0.00020096172753325536, 0.0010948415553926338, 0.00133355347479025, 0.0010468583958018751, 9.269864881500389e-05]}, "battery_in": {"mean": 0.22961505502462387, "se": 0.03524137425517963, "n": 200, "acc_masked": 0.155, "acc_live": 0.185}, "battery_transfer": {"mean": 0.05596338391304016, "se": 0.05618378400233961, "n": 100, "acc_masked": 0.07, "acc_live": 0.09}, "battery_held-out": {"mean": 0.0641354074080785, "se": 0.03950927132175443, "n": 120, "acc_masked": 0.05, "acc_live": 0.058333333333333334}, "gate_open_mean": 0.06336235477370164, "gate_open_max": 0.06533588989691239}, {"step": 400, "live_minus_masked_own": {"mean": -0.3492465664661437, "se": 0.004393077931582688, "masked": 0.9296131361815035, "live": 0.5803665697153597, "per_batch": [-0.337419900007755, -0.3384932437317899, -0.36666486185637226, -0.3324088415095655, -0.3545348394996929, -0.348017814919223, -0.35332029801929976, -0.3631127321854517]}, "live_minus_masked_fineweb": {"mean": 0.0006639712704029355, "se": 0.00015820478632910145, "masked": 0.9505967609111062, "live": 0.9512607321815091, "per_batch": [0.0014807706239288176, 0.0005778617068989966, 0.0008724679878270303, 0.00017705613906326967, 0.0003320125074908198, 0.0010005950627198779, 0.0006789875056063543, 0.0001920186296883175]}, "battery_in": {"mean": 0.5217548792064189, "se": 0.06131239265702742, "n": 200, "acc_masked": 0.155, "acc_live": 0.295}, "battery_transfer": {"mean": 0.06662610709667206, "se": 0.09284188636615415, "n": 100, "acc_masked": 0.07, "acc_live": 0.06}, "battery_held-out": {"mean": 0.1739781285325686, "se": 0.08293881820513324, "n": 120, "acc_masked": 0.05, "acc_live": 0.09166666666666666}, "gate_open_mean": 0.08292569250755885, "gate_open_max": 0.09417776196343719}, {"step": 800, "live_minus_masked_own": {"mean": -0.5572450662740864, "se": 0.005377580426106261, "masked": 0.9296131361815035, "live": 0.3723680699074171, "per_batch": [-0.5373290507220039, -0.5487909204800812, -0.5757709568081606, -0.5373354570757485, -0.5568974517154878, -0.5617716980149161, -0.5643730216005416, -0.5756919737757518]}, "live_minus_masked_fineweb": {"mean": 0.0009511822973438111, "se": 0.00011207823571925854, "masked": 0.9505967609111062, "live": 0.9515479432084499, "per_batch": [0.0008889783223098302, 0.0008287843945078421, 0.0006155259074379638, 0.0005122503255948763, 0.0013130015480122825, 0.0014436223713424345, 0.0010169334145517794, 0.00099036209499348]}, "battery_in": {"mean": 1.4775561818480492, "se": 0.08141909038782355, "n": 200, "acc_masked": 0.155, "acc_live": 0.6}, "battery_transfer": {"mean": 0.6305893403291702, "se": 0.1291595386339828, "n": 100, "acc_masked": 0.07, "acc_live": 0.12}, "battery_held-out": {"mean": 1.153956864774227, "se": 0.12904627168876776, "n": 120, "acc_masked": 0.05, "acc_live": 0.26666666666666666}, "gate_open_mean": 0.1275808543276537, "gate_open_max": 0.18120973163973936}, {"step": 1200, "live_minus_masked_own": {"mean": -0.6726357912873537, "se": 0.0050909539875383285, "masked": 0.9296131361815035, "live": 0.25697734489414975, "per_batch": [-0.6547195926867329, -0.6732405333451369, -0.6888089403470223, -0.6536307920219518, -0.6627709174912125, -0.6895221738979818, -0.6747964388896615, -0.6835969416191308]}, "live_minus_masked_fineweb": {"mean": 0.0008754024418073658, "se": 0.00014486412513851473, "masked": 0.9505967609111062, "live": 0.9514721633529135, "per_batch": [0.0007708262411669775, 0.0011273462764056497, 0.0009668864631506846, 0.0004566139380407508, 0.0009745396911142734, 0.0016389086713971546, 0.0007476085833005852, 0.00032048966988285077]}, "battery_in": {"mean": 1.960613852776587, "se": 0.10256076521960343, "n": 200, "acc_masked": 0.155, "acc_live": 0.73}, "battery_transfer": {"mean": 0.8815224480628967, "se": 0.11607993547680281, "n": 100, "acc_masked": 0.07, "acc_live": 0.21}, "battery_held-out": {"mean": 1.6561269789934159, "se": 0.14207639071265543, "n": 120, "acc_masked": 0.05, "acc_live": 0.43333333333333335}, "gate_open_mean": 0.1713927274559158, "gate_open_max": 0.2897387929423398}, {"step": 1600, "live_minus_masked_own": {"mean": -0.7136420661870317, "se": 0.004987312607915046, "masked": 0.9296131361815035, "live": 0.21597106999447177, "per_batch": [-0.6972545568764362, -0.7140741376633486, -0.7295653029571653, -0.6953419593140224, -0.7017273956687508, -0.7302488695008509, -0.7169300815635764, -0.7239942259521035]}, "live_minus_masked_fineweb": {"mean": 0.0007527465649951603, "se": 0.00017171252942861024, "masked": 0.9505967609111062, "live": 0.9513495074761014, "per_batch": [0.0005285026861012732, 0.0006423552009725553, 0.0008802072071159017, 0.0009185593382583512, 0.0005966938071686467, 0.0017863837945120142, 0.0005610933984399047, 0.00010817708739263576]}, "battery_in": {"mean": 2.023766883965582, "se": 0.17521309630420598, "n": 200, "acc_masked": 0.155, "acc_live": 0.78}, "battery_transfer": {"mean": 0.7686387276649476, "se": 0.1907269318656029, "n": 100, "acc_masked": 0.07, "acc_live": 0.28}, "battery_held-out": {"mean": 1.7590389255434276, "se": 0.29957970144107743, "n": 120, "acc_masked": 0.05, "acc_live": 0.4666666666666667}, "gate_open_mean": 0.20135898469023938, "gate_open_max": 0.36319863706078725}, {"step": 2000, "live_minus_masked_own": {"mean": -0.7291049833304442, "se": 0.005555509858502334, "masked": 0.9296131361815035, "live": 0.20050815285105936, "per_batch": [-0.7101002635368499, -0.7319678360962246, -0.7431107640282769, -0.7088980833094876, -0.7161290078736888, -0.7512316128193678, -0.7335184746720664, -0.7378838243075916]}, "live_minus_masked_fineweb": {"mean": 0.0006626599026901131, "se": 0.00010374059122282537, "masked": 0.9505967609111062, "live": 0.9512594208137963, "per_batch": [0.0004846471101314087, 0.00026038173340625637, 0.0007823490787749465, 0.0008544098094864783, 0.001127260285080478, 0.0008767675540986009, 0.000536413888041154, 0.00037904976250158207]}, "battery_in": {"mean": 2.844541385974735, "se": 0.14491402705263298, "n": 200, "acc_masked": 0.155, "acc_live": 0.875}, "battery_transfer": {"mean": 0.9303885483741761, "se": 0.24688789354297472, "n": 100, "acc_masked": 0.07, "acc_live": 0.34}, "battery_held-out": {"mean": 1.8103366572099426, "se": 0.33700985372803915, "n": 120, "acc_masked": 0.05, "acc_live": 0.43333333333333335}, "gate_open_mean": 0.22207752851960566, "gate_open_max": 0.40633133998950055}, {"step": 2400, "live_minus_masked_own": {"mean": -0.7374527336698304, "se": 0.005535751370502486, "masked": 0.9296131361815035, "live": 0.19216040251167305, "per_batch": [-0.7189096233728387, -0.7405581545285083, -0.7521091972910581, -0.7148046124750571, -0.7277556370112926, -0.7587843169233335, -0.7405019162016762, -0.7461984115548795]}, "live_minus_masked_fineweb": {"mean": 0.00048349697615376885, "se": 0.00015045285185055513, "masked": 0.9505967609111062, "live": 0.9510802578872599, "per_batch": [0.0008041048841089804, -6.363358081906689e-06, 0.0007486404792058643, 0.0008152837564151527, 0.0009578573739805751, 0.00021394641767336076, 0.0005185276923513893, -0.00018402143642326507]}, "battery_in": {"mean": 2.9258429924584926, "se": 0.16647444837099906, "n": 200, "acc_masked": 0.155, "acc_live": 0.9}, "battery_transfer": {"mean": 0.6816713893413544, "se": 0.25107424713858995, "n": 100, "acc_masked": 0.07, "acc_live": 0.25}, "battery_held-out": {"mean": 1.169781177242597, "se": 0.3198557216637863, "n": 120, "acc_masked": 0.05, "acc_live": 0.39166666666666666}, "gate_open_mean": 0.2388701229043596, "gate_open_max": 0.4396489534105897}, {"step": 2800, "live_minus_masked_own": {"mean": -0.7429798274555486, "se": 0.005791561412590601, "masked": 0.9296131361815035, "live": 0.18663330872595507, "per_batch": [-0.7203817303707303, -0.7447609160654741, -0.7583708061371652, -0.7219063780684605, -0.7332839225830365, -0.7644031191140096, -0.747089582655988, -0.7536421646495236]}, "live_minus_masked_fineweb": {"mean": 0.0005125297974025306, "se": 0.00013308495414880595, "masked": 0.9505967609111062, "live": 0.9511092907085087, "per_batch": [-0.00016647920603518607, 0.0006501804115868204, 0.0006649709195610809, 0.0005655649473623736, 0.0010926057809315104, 0.00036460321982911825, 0.0007121801572228526, 0.00021661214876167456]}, "battery_in": {"mean": 3.248008658448234, "se": 0.17048545063490525, "n": 200, "acc_masked": 0.155, "acc_live": 0.885}, "battery_transfer": {"mean": 0.8529209530353546, "se": 0.16659463882093462, "n": 100, "acc_masked": 0.07, "acc_live": 0.31}, "battery_held-out": {"mean": 1.7855773378474018, "se": 0.2939575535449937, "n": 120, "acc_masked": 0.05, "acc_live": 0.44166666666666665}, "gate_open_mean": 0.2530877437444007, "gate_open_max": 0.4654702084544726}, {"step": 3200, "live_minus_masked_own": {"mean": -0.7453827824895483, "se": 0.0059446335687287025, "masked": 0.9296131361815035, "live": 0.1842303536919553, "per_batch": [-0.7227287131135609, -0.7488137732243816, -0.761573918515991, -0.722885926754281, -0.7349055469980219, -0.7668401562723983, -0.7504545737049383, -0.7548596513328127]}, "live_minus_masked_fineweb": {"mean": 0.0005136046889704371, "se": 0.00013688517362490973, "masked": 0.9505967609111062, "live": 0.9511103656000767, "per_batch": [0.0003635713239238392, 0.00013216866718801512, 0.000949258241437323, 0.0006590375181063646, 0.0007126101138501539, 0.00031662006023847056, 0.001031207974573456, -5.563638755412548e-05]}, "battery_in": {"mean": 3.6467077191849238, "se": 0.22014458800310382, "n": 200, "acc_masked": 0.155, "acc_live": 0.835}, "battery_transfer": {"mean": 0.8980604529380798, "se": 0.23222848177968905, "n": 100, "acc_masked": 0.07, "acc_live": 0.33}, "battery_held-out": {"mean": 1.9070637497973317, "se": 0.3806980165399829, "n": 120, "acc_masked": 0.05, "acc_live": 0.475}, "probes_masked": {"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_live": {"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": 0.5, "n": 30, "families": {"category": 0.8333333333333334, "antonym": 0.5, "odd_one_out": 0.5, "property": 0.5, "analogy": 0.16666666666666666}}, "p3_logic": {"acc": 0.2, "n": 30, "families": {"d1": 0.25, "d2": 0.25, "d3": 0.25, "d4": 0.25, "d5": 0.0, "d6": 0.0, "negation": 0.25, "contrapositive": 0.25}}, "p4_arithmetic": {"acc": 0.26666666666666666, "n": 30, "families": {"add2": 0.0, "add3": 0.0, "sub2": 0.5, "sub3": 0.4, "mul1": 0.25, "mul2": 0.3333333333333333, "word": 0.4}}, "p5_causal": {"acc": 0.6, "n": 30, "families": {"effect": 0.8571428571428571, "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": 0.4, "n": 30, "families": {"logic_arith": 0.25, "causal_category": 0.5, "belief_causal": 0.2857142857142857, "three_step": 0.5714285714285714}}, "p8_articulation": {"acc": 0.5333333333333333, "n": 30, "families": {"definition": 0.5, "register": 0.5714285714285714, "grammar": 0.875, "paraphrase": 0.14285714285714285}}}, "probes_mean_masked": 0.41111111111111115, "probes_mean_live": 0.4148148148148148, "gate_open_mean": 0.2655766731078919, "gate_open_max": 0.488023977678262}], "inert_at_birth": true, "seconds": 1715.4674272537231, "log": ["[    0.2s] THE JOB: s3_rules_fresh_seed3_d20261005: 3200 steps of 2 x 4096 bytes, reads at [0, 200, 400, 800, 1200, 1600, 2000, 2400, 2800, 3200]; quiet term on (lambda 2.0); adapter chain compiled; trunk step 245,674; card 31.4 GB capped at 0.92; host RSS 0.6 GB", "[   19.2s] mini-beatrix-3 @ 245,674: trunk loaded from /root/v3/runs/mission/mini-beatrix-3/checkpoints/step_00245674.safetensors and frozen (376.0M)", "[   31.5s] own rows (rulechain-minted; the hub tensors): 'If someone is hoream, then they are hortil. If someone is hortil, then they are babig. If someone is babig, then they are banng. Vin is hoream. So Vin is hortil. So Vin i'", "[   66.2s] THE BARE TRUNK (the arm masked): own rows 0.9296 bpb, web holdout 0.9506 bpb; battery acc in 0.155 (n=200), transfer 0.070 (n=100), held-out 0.050 (n=120); probes mean 0.411", "[   75.1s] at birth: own rows live-minus-masked +0.00000 +- 0.00000 (0.9296 -> 0.9296); web holdout +0.00000 +- 0.00000 (the quiet bar +.012); battery in +0.000 (se 0.000; acc 0.155 -> 0.155); battery transfer +0.000 (se 0.000; acc 0.070 -> 0.070); battery held-out +0.000 (se 0.000; acc 0.050 -> 0.050); gates open mean 0.047 / widest 0.047; host RSS 2.3 GB", "[   91.0s] step 1/3200: train bpb 0.9197; quiet term 0.00000; 15.85 s/step; 0.3 min spent, about 845.3 min of steps left; card peak 13.6 GB", "[  114.4s] step 50/3200: train bpb 0.9050; quiet term 0.00047; 0.79 s/step; 0.7 min spent, about 41.3 min of steps left; card peak 13.7 GB", "[  138.0s] step 100/3200: train bpb 0.8557; quiet term 0.00084; 0.63 s/step; 1.0 min spent, about 32.5 min of steps left; card peak 13.7 GB", "[  161.4s] step 150/3200: train bpb 0.8053; quiet term 0.00084; 0.58 s/step; 1.4 min spent, about 29.2 min of steps left; card peak 13.7 GB", "[  184.9s] step 200/3200: train bpb 0.7506; quiet term 0.00109; 0.55 s/step; 1.8 min spent, about 27.4 min of steps left; card peak 13.7 GB", "[  192.3s] after 200 steps: own rows live-minus-masked -0.19917 +- 0.00295 (0.9296 -> 0.7304); web holdout +0.00080 +- 0.00024 (the quiet bar +.012); battery in +0.230 (se 0.035; acc 0.155 -> 0.185); battery transfer +0.056 (se 0.056; acc 0.070 -> 0.090); battery held-out +0.064 (se 0.040; acc 0.050 -> 0.058); gates open mean 0.063 / widest 0.065; host RSS 4.9 GB", "[  215.8s] step 250/3200: train bpb 0.6934; quiet term 0.00151; 0.56 s/step; 2.3 min spent, about 27.7 min of steps left; card peak 13.7 GB", "[  239.6s] step 300/3200: train bpb 0.6492; quiet term 0.00137; 0.55 s/step; 2.7 min spent, about 26.5 min of steps left; card peak 13.7 GB", "[  263.1s] step 350/3200: train bpb 0.6121; quiet term 0.00128; 0.54 s/step; 3.1 min spent, about 25.5 min of steps left; card peak 13.7 GB", "[  286.3s] step 400/3200: train bpb 0.5784; quiet term 0.00130; 0.53 s/step; 3.5 min spent, about 24.6 min of steps left; card peak 13.7 GB", "[  293.7s] after 400 steps: own rows live-minus-masked -0.34925 +- 0.00439 (0.9296 -> 0.5804); web holdout +0.00066 +- 0.00016 (the quiet bar +.012); battery in +0.522 (se 0.061; acc 0.155 -> 0.295); battery transfer +0.067 (se 0.093; acc 0.070 -> 0.060); battery held-out +0.174 (se 0.083; acc 0.050 -> 0.092); gates open mean 0.083 / widest 0.094; host RSS 4.9 GB", "[  293.9s] anchor saved: /root/refit/long/anchors/s3_rules_fresh_seed3_d20261005_step400.safetensors", "[  317.3s] step 450/3200: train bpb 0.5499; quiet term 0.00135; 0.54 s/step; 4.0 min spent, about 24.7 min of steps left; card peak 13.7 GB", "[  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", "[ 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"], "train_bpb_first25": 0.9177436314245245, "train_bpb_last25": 0.1801375947844765, "descended": true, "quiet_first25": 0.0004210810794029385, "quiet_last25": 0.001174506489187479, "boundary_gauge": {"armed_fineweb_bpb": 0.9510053916895562, "armed_stage_bpb": 0.18425568216813823, "s3_rules": {"masked_fineweb_bpb": 0.9505967609111061, "masked_stage_bpb": 0.9296131361815035, "fineweb_delta": 0.0004086307784501031, "stage_delta": -0.7453574540133653, "write_own": 0.033664023014353006, "write_neutral": 0.006031400203482917, "selectivity": 5.581460668936086}, "bare_fineweb_bpb": 0.9505967609111061}}

mini-beatrix-3/arm_refit/solo_long/results/result_s4_arith_fresh_seed4_d20261005.json

@@ -1 +1 @@
-{"tag": "s4_arith_fresh_seed4_d20261005", "arm": "s4_arith", "phase": "curriculum_s4", "own_source": "arith-synth", "own_rows_held_out": true, "init": "fresh", "seed": 4, "data_seed": 20261005, "trunk_step": 245674, "steps": 3200, "ctx": 4096, "mb": 2, "lr": 0.001, "lambda": 2.0, "quiet": true, "compiled_chain": true, "replication": false, "spec": {"n_slots": 16, "K": 16, "D": 8, "hidden": 256}, "reads": [{"step": 0, "live_minus_masked_own": {"mean": 0.0, "se": 0.0, "masked": 1.2338748670942346, "live": 1.2338748670942346, "per_batch": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "live_minus_masked_fineweb": {"mean": 0.0, "se": 0.0, "masked": 0.9505967609111062, "live": 0.9505967609111062, "per_batch": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "battery_in": {"mean": 0.0, "se": 0.0, "n": 160, "acc_masked": 0.34375, "acc_live": 0.34375}, "battery_transfer": {"mean": 0.0, "se": 0.0, "n": 240, "acc_masked": 0.24166666666666667, "acc_live": 0.24166666666666667}, "battery_held-out": {"mean": 0.0, "se": 0.0, "n": 80, "acc_masked": 0.35, "acc_live": 0.35}, "gate_open_mean": 0.04742587317756678, "gate_open_max": 0.04742587317756678}, {"step": 200, "live_minus_masked_own": {"mean": -0.4017842566083417, "se": 0.005061161440211174, "masked": 1.2338748670942346, "live": 0.8320906104858928, "per_batch": [-0.4037906707068013, -0.3960227303152808, -0.4177188576695766, -0.3889391088915436, -0.3791569076931177, -0.41942853720180395, -0.39627021334987234, -0.4129470270387373]}, "live_minus_masked_fineweb": {"mean": 0.0004098884015844573, "se": 0.00016837925112639325, "masked": 0.9505967609111062, "live": 0.9510066493126906, "per_batch": [-0.00011196070571184435, 0.0005283307034504858, 0.000547936725648901, 4.488747187525455e-05, 0.0011943335189168458, -0.00023080071745784636, 0.0006040890611556105, 0.0007022911547982513]}, "battery_in": {"mean": 0.024818579852581023, "se": 0.01973090979475853, "n": 160, "acc_masked": 0.34375, "acc_live": 0.39375}, "battery_transfer": {"mean": -0.019725271066029868, "se": 0.013360559288625333, "n": 240, "acc_masked": 0.24166666666666667, "acc_live": 0.225}, "battery_held-out": {"mean": 0.011111414432525635, "se": 0.021338248873477846, "n": 80, "acc_masked": 0.35, "acc_live": 0.3375}, "gate_open_mean": 0.06263439556862972, "gate_open_max": 0.06417519751234678}, {"step": 400, "live_minus_masked_own": {"mean": -0.6205602050918797, "se": 0.004969367370604802, "masked": 1.2338748670942346, "live": 0.6133146620023548, "per_batch": [-0.630155611633637, -0.6133550134318136, -0.6431927564822921, -0.5975174320834966, -0.6102137073181416, -0.6235301949874609, -0.6175454137157487, -0.6289715110824466]}, "live_minus_masked_fineweb": {"mean": 0.0007009797870854567, "se": 0.00020139235408143472, "masked": 0.9505967609111062, "live": 0.9512977406981916, "per_batch": [-2.416356244627771e-05, 9.364455339477828e-05, 0.001284882384596009, 0.000389024756251688, 0.0005107024817372352, 0.0006697004424598418, 0.001541136534381593, 0.0011429107063087862]}, "battery_in": {"mean": 0.12010455802083016, "se": 0.060308358618697024, "n": 160, "acc_masked": 0.34375, "acc_live": 0.425}, "battery_transfer": {"mean": -0.05276497701803843, "se": 0.024308311765648672, "n": 240, "acc_masked": 0.24166666666666667, "acc_live": 0.22916666666666666}, "battery_held-out": {"mean": -0.01434510350227356, "se": 0.03557401434071663, "n": 80, "acc_masked": 0.35, "acc_live": 0.325}, "gate_open_mean": 0.07970948941070174, "gate_open_max": 0.08995287884018897}], "inert_at_birth": true, "seconds": 267.8654670715332, "log": ["[    0.2s] THE JOB: s4_arith_fresh_seed4_d20261005: 3200 steps of 2 x 4096 bytes, reads at [0, 200, 400, 800, 1200, 1600, 2000, 2400, 2800, 3200]; quiet term on (lambda 2.0); adapter chain compiled; trunk step 245,674; card 31.4 GB capped at 0.92; host RSS 0.6 GB", "[   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)", "[   26.5s] own rows (arith-synth; the hub tensors): 'Count by 3s: 1 3s make 3; 2 3s make 6; 3 3s make 9; 4 3s make 12; 5 3s make 15. So 3 times 5 is 15. | \ufffdStart at 84 and take away 33. 84 - 33 = 51. | \ufffdCount by 4s: 1 4s ma'", "[   57.5s] THE BARE TRUNK (the arm masked): own rows 1.2339 bpb, web holdout 0.9506 bpb; battery acc in 0.344 (n=160), transfer 0.242 (n=240), held-out 0.350 (n=80); probes mean 0.411", "[   62.6s] at birth: own rows live-minus-masked +0.00000 +- 0.00000 (1.2339 -> 1.2339); web holdout +0.00000 +- 0.00000 (the quiet bar +.012); battery in +0.000 (se 0.000; acc 0.344 -> 0.344); battery transfer +0.000 (se 0.000; acc 0.242 -> 0.242); battery held-out +0.000 (se 0.000; acc 0.350 -> 0.350); gates open mean 0.047 / widest 0.047; host RSS 2.2 GB", "[   69.2s] step 1/3200: train bpb 1.2516; quiet term 0.00000; 6.60 s/step; 0.1 min spent, about 352.1 min of steps left; card peak 13.6 GB", "[   93.0s] step 50/3200: train bpb 1.2052; quiet term 0.00046; 0.61 s/step; 0.5 min spent, about 31.9 min of steps left; card peak 13.7 GB", "[  116.8s] step 100/3200: train bpb 1.1276; quiet term 0.00076; 0.54 s/step; 0.9 min spent, about 28.0 min of steps left; card peak 13.7 GB", "[  140.3s] step 150/3200: train bpb 0.9943; quiet term 0.00108; 0.52 s/step; 1.3 min spent, about 26.3 min of steps left; card peak 13.7 GB", "[  164.0s] step 200/3200: train bpb 0.8830; quiet term 0.00093; 0.51 s/step; 1.7 min spent, about 25.4 min of steps left; card peak 13.7 GB", "[  167.6s] after 200 steps: own rows live-minus-masked -0.40178 +- 0.00506 (1.2339 -> 0.8321); web holdout +0.00041 +- 0.00017 (the quiet bar +.012); battery in +0.025 (se 0.020; acc 0.344 -> 0.394); battery transfer -0.020 (se 0.013; acc 0.242 -> 0.225); battery held-out +0.011 (se 0.021; acc 0.350 -> 0.338); gates open mean 0.063 / widest 0.064; host RSS 4.9 GB", "[  191.5s] step 250/3200: train bpb 0.8105; quiet term 0.00099; 0.52 s/step; 2.1 min spent, about 25.4 min of steps left; card peak 13.7 GB", "[  215.8s] step 300/3200: train bpb 0.7561; quiet term 0.00112; 0.51 s/step; 2.6 min spent, about 24.7 min of steps left; card peak 13.7 GB", "[  239.8s] step 350/3200: train bpb 0.7015; quiet term 0.00130; 0.51 s/step; 3.0 min spent, about 24.1 min of steps left; card peak 13.7 GB", "[  264.1s] step 400/3200: train bpb 0.6432; quiet term 0.00136; 0.50 s/step; 3.4 min spent, about 23.5 min of steps left; card peak 13.7 GB", "[  267.7s] after 400 steps: own rows live-minus-masked -0.62056 +- 0.00497 (1.2339 -> 0.6133); web holdout +0.00070 +- 0.00020 (the quiet bar +.012); battery in +0.120 (se 0.060; acc 0.344 -> 0.425); battery transfer -0.053 (se 0.024; acc 0.242 -> 0.229); battery held-out -0.014 (se 0.036; acc 0.350 -> 0.325); gates open mean 0.080 / widest 0.090; host RSS 4.9 GB", "[  267.9s] anchor saved: /root/refit/long/anchors/s4_arith_fresh_seed4_d20261005_step400.safetensors"]}
+{"tag": "s4_arith_fresh_seed4_d20261005", "arm": "s4_arith", "phase": "curriculum_s4", "own_source": "arith-synth", "own_rows_held_out": true, "init": "fresh", "seed": 4, "data_seed": 20261005, "trunk_step": 245674, "steps": 3200, "ctx": 4096, "mb": 2, "lr": 0.001, "lambda": 2.0, "quiet": true, "compiled_chain": true, "replication": false, "spec": {"n_slots": 16, "K": 16, "D": 8, "hidden": 256}, "reads": [{"step": 0, "live_minus_masked_own": {"mean": 0.0, "se": 0.0, "masked": 1.2338748670942346, "live": 1.2338748670942346, "per_batch": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "live_minus_masked_fineweb": {"mean": 0.0, "se": 0.0, "masked": 0.9505967609111062, "live": 0.9505967609111062, "per_batch": [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}, "battery_in": {"mean": 0.0, "se": 0.0, "n": 160, "acc_masked": 0.34375, "acc_live": 0.34375}, "battery_transfer": {"mean": 0.0, "se": 0.0, "n": 240, "acc_masked": 0.24166666666666667, "acc_live": 0.24166666666666667}, "battery_held-out": {"mean": 0.0, "se": 0.0, "n": 80, "acc_masked": 0.35, "acc_live": 0.35}, "gate_open_mean": 0.04742587317756678, "gate_open_max": 0.04742587317756678}, {"step": 200, "live_minus_masked_own": {"mean": -0.4017842566083417, "se": 0.005061161440211174, "masked": 1.2338748670942346, "live": 0.8320906104858928, "per_batch": [-0.4037906707068013, -0.3960227303152808, -0.4177188576695766, -0.3889391088915436, -0.3791569076931177, -0.41942853720180395, -0.39627021334987234, -0.4129470270387373]}, "live_minus_masked_fineweb": {"mean": 0.0004098884015844573, "se": 0.00016837925112639325, "masked": 0.9505967609111062, "live": 0.9510066493126906, "per_batch": [-0.00011196070571184435, 0.0005283307034504858, 0.000547936725648901, 4.488747187525455e-05, 0.0011943335189168458, -0.00023080071745784636, 0.0006040890611556105, 0.0007022911547982513]}, "battery_in": {"mean": 0.024818579852581023, "se": 0.01973090979475853, "n": 160, "acc_masked": 0.34375, "acc_live": 0.39375}, "battery_transfer": {"mean": -0.019725271066029868, "se": 0.013360559288625333, "n": 240, "acc_masked": 0.24166666666666667, "acc_live": 0.225}, "battery_held-out": {"mean": 0.011111414432525635, "se": 0.021338248873477846, "n": 80, "acc_masked": 0.35, "acc_live": 0.3375}, "gate_open_mean": 0.06263439556862972, "gate_open_max": 0.06417519751234678}, {"step": 400, "live_minus_masked_own": {"mean": -0.6205602050918797, "se": 0.004969367370604802, "masked": 1.2338748670942346, "live": 0.6133146620023548, "per_batch": [-0.630155611633637, -0.6133550134318136, -0.6431927564822921, -0.5975174320834966, -0.6102137073181416, -0.6235301949874609, -0.6175454137157487, -0.6289715110824466]}, "live_minus_masked_fineweb": {"mean": 0.0007009797870854567, "se": 0.00020139235408143472, "masked": 0.9505967609111062, "live": 0.9512977406981916, "per_batch": [-2.416356244627771e-05, 9.364455339477828e-05, 0.001284882384596009, 0.000389024756251688, 0.0005107024817372352, 0.0006697004424598418, 0.001541136534381593, 0.0011429107063087862]}, "battery_in": {"mean": 0.12010455802083016, "se": 0.060308358618697024, "n": 160, "acc_masked": 0.34375, "acc_live": 0.425}, "battery_transfer": {"mean": -0.05276497701803843, "se": 0.024308311765648672, "n": 240, "acc_masked": 0.24166666666666667, "acc_live": 0.22916666666666666}, "battery_held-out": {"mean": -0.01434510350227356, "se": 0.03557401434071663, "n": 80, "acc_masked": 0.35, "acc_live": 0.325}, "gate_open_mean": 0.07970948941070174, "gate_open_max": 0.08995287884018897}, {"step": 800, "live_minus_masked_own": {"mean": -0.8449529484060907, "se": 0.004752980677640924, "masked": 1.2338748670942346, "live": 0.38892191868814396, "per_batch": [-0.8541921654417665, -0.8481287451029472, -0.8613239420077453, -0.8197335495322433, -0.8323397703513302, -0.8543917513080928, -0.8418115213671172, -0.8477021421374822]}, "live_minus_masked_fineweb": {"mean": 0.0006933480569534384, "se": 0.0002323573181582207, "masked": 0.9505967609111062, "live": 0.9512901089680597, "per_batch": [0.0005236011805518359, 0.00012399949127206433, 0.0018096014523785175, 0.0007504462970399084, -0.0001502268455284561, 0.0009236328264586868, 0.00021695611406347126, 0.0013487739393914788]}, "battery_in": {"mean": 0.8293458690633997, "se": 0.15933903625568097, "n": 160, "acc_masked": 0.34375, "acc_live": 0.4875}, "battery_transfer": {"mean": -0.04161893973747889, "se": 0.03481104143827223, "n": 240, "acc_masked": 0.24166666666666667, "acc_live": 0.24166666666666667}, "battery_held-out": {"mean": -0.20069611668586732, "se": 0.09017014823882243, "n": 80, "acc_masked": 0.35, "acc_live": 0.3125}, "gate_open_mean": 0.11542622571075042, "gate_open_max": 0.15665446631154095}, {"step": 1200, "live_minus_masked_own": {"mean": -0.8912366694940002, "se": 0.0054814006427228645, "masked": 1.2338748670942346, "live": 0.34263819760023445, "per_batch": [-0.9078616734092926, -0.8859598548045646, -0.9059407991818058, -0.867343463774932, -0.8718482053489813, -0.9057616792509322, -0.8931739680646442, -0.8920037121168483]}, "live_minus_masked_fineweb": {"mean": 0.0005708426649606996, "se": 0.00027936669275304316, "masked": 0.9505967609111062, "live": 0.9511676035760668, "per_batch": [-0.00020465935452684825, -9.166675290994686e-05, 0.0017265338320117118, 0.0006922301697228628, 0.0005263529029657654, 0.0012913317340034203, -0.0005009854619633103, 0.0011276042503819417]}, "battery_in": {"mean": 1.5228036325836, "se": 0.2618638755535045, "n": 160, "acc_masked": 0.34375, "acc_live": 0.55625}, "battery_transfer": {"mean": -0.11427276382843653, "se": 0.04764889425041648, "n": 240, "acc_masked": 0.24166666666666667, "acc_live": 0.2375}, "battery_held-out": {"mean": -0.29411683678627015, "se": 0.10777543557118893, "n": 80, "acc_masked": 0.35, "acc_live": 0.2625}, "gate_open_mean": 0.13850363997821008, "gate_open_max": 0.20570247969270677}, {"step": 1600, "live_minus_masked_own": {"mean": -0.9134001220808109, "se": 0.006258105141294883, "masked": 1.2338748670942346, "live": 0.32047474501342377, "per_batch": [-0.9311297221924746, -0.9056571567948704, -0.9290980051465066, -0.8873727137792342, -0.889925774733461, -0.9313522247470283, -0.9178347968480073, -0.9148305824049046]}, "live_minus_masked_fineweb": {"mean": 0.00046869571926387577, "se": 0.00022047372831440914, "masked": 0.9505967609111062, "live": 0.95106545663037, "per_batch": [-1.4016586045384472e-05, 0.0003022595088914004, 0.0015311615406315982, 0.0005125082995711683, 0.00014222965226373674, 0.000690854308515898, -0.0004535182503252466, 0.0010380872806078356]}, "battery_in": {"mean": 2.7862645976621705, "se": 0.41182980019283233, "n": 160, "acc_masked": 0.34375, "acc_live": 0.6125}, "battery_transfer": {"mean": -0.11464335918426513, "se": 0.045989978660138486, "n": 240, "acc_masked": 0.24166666666666667, "acc_live": 0.22916666666666666}, "battery_held-out": {"mean": -0.35274759531021116, "se": 0.1293126799964322, "n": 80, "acc_masked": 0.35, "acc_live": 0.275}, "gate_open_mean": 0.157632666517308, "gate_open_max": 0.2632528794294279}, {"step": 2000, "live_minus_masked_own": {"mean": -0.93336637000309, "se": 0.006864575288618662, "masked": 1.2338748670942346, "live": 0.3005084970911446, "per_batch": [-0.9513535705662279, -0.923208889224337, -0.947181507932441, -0.9051474002177131, -0.9081884404459318, -0.956949413073014, -0.9383219721580631, -0.9365797664069916]}, "live_minus_masked_fineweb": {"mean": 0.00045089551489957413, "se": 0.0002575988043421469, "masked": 0.9505967609111062, "live": 0.9510476564260057, "per_batch": [-0.0006891344820072476, -0.00018737509811506126, 0.0011216708489272254, 0.0008389313709088464, 0.0007151898536130741, 0.0008878604350792685, -0.0002974439946672458, 0.0012174651854577334]}, "battery_in": {"mean": 3.8386816872630503, "se": 0.5610279441564884, "n": 160, "acc_masked": 0.34375, "acc_live": 0.65}, "battery_transfer": {"mean": -0.22394415785868962, "se": 0.05389359075962423, "n": 240, "acc_masked": 0.24166666666666667, "acc_live": 0.23333333333333334}, "battery_held-out": {"mean": -0.28236727714538573, "se": 0.11666123600511272, "n": 80, "acc_masked": 0.35, "acc_live": 0.275}, "gate_open_mean": 0.1758980117592324, "gate_open_max": 0.3260214406484525}, {"step": 2400, "live_minus_masked_own": {"mean": -0.9455977841941396, "se": 0.007041907781034426, "masked": 1.2338748670942346, "live": 0.2882770829000949, "per_batch": [-0.964693297391245, -0.9342975136343793, -0.9593786250207883, -0.9177437750300382, -0.9186718579075879, -0.9689457404164787, -0.950808579043475, -0.9502428851091247]}, "live_minus_masked_fineweb": {"mean": 0.00029085490935605585, "se": 0.0001828798861411164, "masked": 0.9505967609111062, "live": 0.9508876158204622, "per_batch": [-0.0001257193177808258, -0.0006100224626104378, 0.0005971237637957261, 0.0003044092920272412, 0.0008471005468249082, 0.0007297223876109316, -8.986093507579174e-05, 0.0006740860000566951]}, "battery_in": {"mean": 3.7111849593522033, "se": 0.45703217798019874, "n": 160, "acc_masked": 0.34375, "acc_live": 0.73125}, "battery_transfer": {"mean": -0.23503285199403762, "se": 0.05722902504886821, "n": 240, "acc_masked": 0.24166666666666667, "acc_live": 0.22083333333333333}, "battery_held-out": {"mean": -0.5187231808900833, "se": 0.18202673644317552, "n": 80, "acc_masked": 0.35, "acc_live": 0.325}, "gate_open_mean": 0.19363744784541762, "gate_open_max": 0.38186628601022293}, {"step": 2800, "live_minus_masked_own": {"mean": -0.9547530468239052, "se": 0.006912103068578779, "masked": 1.2338748670942346, "live": 0.2791218202703293, "per_batch": [-0.9716391822094725, -0.9405388285276847, -0.9703490113583624, -0.9318319488537787, -0.9255281827603787, -0.9768870823157316, -0.9624629403835305, -0.9587871981823026]}, "live_minus_masked_fineweb": {"mean": 0.0003210271156667971, "se": 0.00018642882363367963, "masked": 0.9505967609111062, "live": 0.950917788026773, "per_batch": [-0.0004483587707992953, -0.00016510334482833233, 0.0008613751068463626, -3.7234243911776765e-05, 0.0007031510680526321, 0.001062336834379618, 0.0003173079908418419, 0.0002747422847533265]}, "battery_in": {"mean": 4.219581255415757, "se": 0.5264318299026162, "n": 160, "acc_masked": 0.34375, "acc_live": 0.7625}, "battery_transfer": {"mean": -0.13889329135417938, "se": 0.048008332868833686, "n": 240, "acc_masked": 0.24166666666666667, "acc_live": 0.24166666666666667}, "battery_held-out": {"mean": -0.9279398292303085, "se": 0.2335540430100328, "n": 80, "acc_masked": 0.35, "acc_live": 0.3}, "gate_open_mean": 0.20990397784030917, "gate_open_max": 0.42474200735799067}, {"step": 3200, "live_minus_masked_own": {"mean": -0.9601798438210376, "se": 0.007344438968373747, "masked": 1.2338748670942346, "live": 0.27369502327319695, "per_batch": [-0.9793031375907151, -0.9457435179929146, -0.9756294591812973, -0.9339212585899543, -0.9304729849423234, -0.9836782257438454, -0.9683047180810502, -0.9643854484462007]}, "live_minus_masked_fineweb": {"mean": 0.00030336664720634343, "se": 0.00017525482570802188, "masked": 0.9505967609111062, "live": 0.9509001275583124, "per_batch": [0.0005004695140107263, -0.0005845690302827, 0.0005913623449917971, 9.269864881511491e-05, 0.000973077838581915, 0.0001434335308195811, -2.2873692564706616e-05, 0.0007333340232790198]}, "battery_in": {"mean": 5.676389947077123, "se": 0.6227717757446342, "n": 160, "acc_masked": 0.34375, "acc_live": 0.80625}, "battery_transfer": {"mean": -0.16589549779891968, "se": 0.03995100235216166, "n": 240, "acc_masked": 0.24166666666666667, "acc_live": 0.2375}, "battery_held-out": {"mean": -0.8051506698131561, "se": 0.23375744604561652, "n": 80, "acc_masked": 0.35, "acc_live": 0.325}, "probes_masked": {"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_live": {"p0_grounding": {"acc": 0.3333333333333333, "n": 30, "families": {"count": 0.0, "permanence": 0.7142857142857143, "sequence": 0.25, "attribute": 0.42857142857142855}}, "p1_self_other": {"acc": 0.5333333333333333, "n": 30, "families": {"belief": 0.375, "reality": 0.6666666666666666, "perspective": 0.8, "reference": 0.7142857142857143, "second_order": 0.0}}, "p2_difference": {"acc": 0.5333333333333333, "n": 30, "families": {"category": 0.8333333333333334, "antonym": 0.5, "odd_one_out": 0.6666666666666666, "property": 0.5, "analogy": 0.16666666666666666}}, "p3_logic": {"acc": 0.2, "n": 30, "families": {"d1": 0.5, "d2": 0.25, "d3": 0.5, "d4": 0.0, "d5": 0.0, "d6": 0.0, "negation": 0.0, "contrapositive": 0.25}}, "p4_arithmetic": {"acc": 0.4, "n": 30, "families": {"add2": 0.4, "add3": 0.25, "sub2": 1.0, "sub3": 0.2, "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_masked": 0.41111111111111115, "probes_mean_live": 0.4222222222222222, "gate_open_mean": 0.22428606614484503, "gate_open_max": 0.45898487523781806}], "inert_at_birth": true, "seconds": 1678.4997999668121, "log": ["[    0.2s] THE JOB: s4_arith_fresh_seed4_d20261005: 3200 steps of 2 x 4096 bytes, reads at [0, 200, 400, 800, 1200, 1600, 2000, 2400, 2800, 3200]; quiet term on (lambda 2.0); adapter chain compiled; trunk step 245,674; card 31.4 GB capped at 0.92; host RSS 0.6 GB", "[   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)", "[   26.5s] own rows (arith-synth; the hub tensors): 'Count by 3s: 1 3s make 3; 2 3s make 6; 3 3s make 9; 4 3s make 12; 5 3s make 15. So 3 times 5 is 15. | \ufffdStart at 84 and take away 33. 84 - 33 = 51. | \ufffdCount by 4s: 1 4s ma'", "[   57.5s] THE BARE TRUNK (the arm masked): own rows 1.2339 bpb, web holdout 0.9506 bpb; battery acc in 0.344 (n=160), transfer 0.242 (n=240), held-out 0.350 (n=80); probes mean 0.411", "[   62.6s] at birth: own rows live-minus-masked +0.00000 +- 0.00000 (1.2339 -> 1.2339); web holdout +0.00000 +- 0.00000 (the quiet bar +.012); battery in +0.000 (se 0.000; acc 0.344 -> 0.344); battery transfer +0.000 (se 0.000; acc 0.242 -> 0.242); battery held-out +0.000 (se 0.000; acc 0.350 -> 0.350); gates open mean 0.047 / widest 0.047; host RSS 2.2 GB", "[   69.2s] step 1/3200: train bpb 1.2516; quiet term 0.00000; 6.60 s/step; 0.1 min spent, about 352.1 min of steps left; card peak 13.6 GB", "[   93.0s] step 50/3200: train bpb 1.2052; quiet term 0.00046; 0.61 s/step; 0.5 min spent, about 31.9 min of steps left; card peak 13.7 GB", "[  116.8s] step 100/3200: train bpb 1.1276; quiet term 0.00076; 0.54 s/step; 0.9 min spent, about 28.0 min of steps left; card peak 13.7 GB", "[  140.3s] step 150/3200: train bpb 0.9943; quiet term 0.00108; 0.52 s/step; 1.3 min spent, about 26.3 min of steps left; card peak 13.7 GB", "[  164.0s] step 200/3200: train bpb 0.8830; quiet term 0.00093; 0.51 s/step; 1.7 min spent, about 25.4 min of steps left; card peak 13.7 GB", "[  167.6s] after 200 steps: own rows live-minus-masked -0.40178 +- 0.00506 (1.2339 -> 0.8321); web holdout +0.00041 +- 0.00017 (the quiet bar +.012); battery in +0.025 (se 0.020; acc 0.344 -> 0.394); battery transfer -0.020 (se 0.013; acc 0.242 -> 0.225); battery held-out +0.011 (se 0.021; acc 0.350 -> 0.338); gates open mean 0.063 / widest 0.064; host RSS 4.9 GB", "[  191.5s] step 250/3200: train bpb 0.8105; quiet term 0.00099; 0.52 s/step; 2.1 min spent, about 25.4 min of steps left; card peak 13.7 GB", "[  215.8s] step 300/3200: train bpb 0.7561; quiet term 0.00112; 0.51 s/step; 2.6 min spent, about 24.7 min of steps left; card peak 13.7 GB", "[  239.8s] step 350/3200: train bpb 0.7015; quiet term 0.00130; 0.51 s/step; 3.0 min spent, about 24.1 min of steps left; card peak 13.7 GB", "[  264.1s] step 400/3200: train bpb 0.6432; quiet term 0.00136; 0.50 s/step; 3.4 min spent, about 23.5 min of steps left; card peak 13.7 GB", "[  267.7s] after 400 steps: own rows live-minus-masked -0.62056 +- 0.00497 (1.2339 -> 0.6133); web holdout +0.00070 +- 0.00020 (the quiet bar +.012); battery in +0.120 (se 0.060; acc 0.344 -> 0.425); battery transfer -0.053 (se 0.024; acc 0.242 -> 0.229); battery held-out -0.014 (se 0.036; acc 0.350 -> 0.325); gates open mean 0.080 / widest 0.090; host RSS 4.9 GB", "[  267.9s] anchor saved: /root/refit/long/anchors/s4_arith_fresh_seed4_d20261005_step400.safetensors", "[  292.1s] step 450/3200: train bpb 0.6063; quiet term 0.00133; 0.51 s/step; 3.8 min spent, about 23.4 min of steps left; card peak 13.7 GB", "[  315.9s] step 500/3200: train bpb 0.5524; quiet term 0.00131; 0.51 s/step; 4.2 min spent, about 22.8 min of steps left; card peak 13.7 GB", "[  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", "[ 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"], "train_bpb_first25": 1.2203467582126817, "train_bpb_last25": 0.2847064977709104, "descended": true, "quiet_first25": 0.00042157103773206474, "quiet_last25": 0.0009663339448161423, "boundary_gauge": {"armed_fineweb_bpb": 0.9507929716179093, "armed_stage_bpb": 0.27364914152662145, "s4_arith": {"masked_fineweb_bpb": 0.9505967609111061, "masked_stage_bpb": 1.2338748670942346, "fineweb_delta": 0.00019621070680320152, "stage_delta": -0.9602257255676131, "write_own": 0.026783284702105448, "write_neutral": 0.004821105276732851, "selectivity": 5.555424153744231}, "bare_fineweb_bpb": 0.9505967609111061}}