jc1122's picture
Circuit atlas networks (+ integer-matvec)
b9c5bae verified
Raw History Blame Contribute Delete
16 kB
{
"entry": "integer-matvec",
"hand_built": {
"hand_A": {
"file": "hand_A.safetensors",
"cfg": {
"vocab_size": 15,
"hidden_size": 128,
"num_hidden_layers": 5,
"num_attention_heads": 4,
"intermediate_size": 640,
"rotary_pct": 0.0,
"use_parallel_residual": false,
"max_position_embeddings": 512,
"bos_token_id": 0,
"eos_token_id": 0,
"tie_word_embeddings": false,
"hidden_act": "relu",
"no_layernorm": true,
"initializer_range": 0.02,
"attention_dropout": 0.0,
"hidden_dropout": 0.0
},
"roundtrip_ok": true,
"sha256": "a332abe6a7f42ea022325329419ff25b3754f0fc418f09820b353f5d365e5a96"
},
"hand_T": {
"file": "hand_T.safetensors",
"cfg": {
"vocab_size": 15,
"hidden_size": 128,
"num_hidden_layers": 5,
"num_attention_heads": 4,
"intermediate_size": 640,
"rotary_pct": 0.0,
"use_parallel_residual": false,
"max_position_embeddings": 512,
"bos_token_id": 0,
"eos_token_id": 0,
"tie_word_embeddings": false,
"hidden_act": "relu",
"no_layernorm": true,
"initializer_range": 0.02,
"attention_dropout": 0.0,
"hidden_dropout": 0.0
},
"roundtrip_ok": true,
"sha256": "57e79ca939ff7aa099c5a75042394c15ac614ee654159865ddec5e2ad0e2c001"
}
},
"config": "config.json",
"nets": [
{
"file": "train_T_r1_best_s0.safetensors",
"rung": "TRAIN",
"fmt": "T",
"retry": "r1",
"wd": 0.1,
"snapshot": "best",
"seed": 0,
"best_step": 12500,
"circuit_type": "T2 late-routing local adder (HAND-b variables)",
"acc_test_fp32": 1.0,
"recorded_test": 1.0,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "07d293e18e0e9e77bed474d334a8944af6fb7f3bc505bc69278c3350426f3b51"
},
{
"file": "train_T_r1_best_s1.safetensors",
"rung": "TRAIN",
"fmt": "T",
"retry": "r1",
"wd": 0.1,
"snapshot": "best",
"seed": 1,
"best_step": 30000,
"circuit_type": "unsolved (.9155)",
"acc_test_fp32": 0.9155,
"recorded_test": 0.9155,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "b5b2ced390bfe18821e45b4b5deb47056eccd730de0722685d6448482141989e"
},
{
"file": "train_T_r1_best_s2.safetensors",
"rung": "TRAIN",
"fmt": "T",
"retry": "r1",
"wd": 0.1,
"snapshot": "best",
"seed": 2,
"best_step": 21000,
"circuit_type": "T2 late-routing local adder (HAND-b variables)",
"acc_test_fp32": 1.0,
"recorded_test": 1.0,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "ee2e8a5579b24fda602d3da76fe12092992318861f5a432c62f87b9bcc0e18e4"
},
{
"file": "train_T_r1_final_s0.safetensors",
"rung": "TRAIN",
"fmt": "T",
"retry": "r1",
"wd": 0.1,
"snapshot": "final",
"seed": 0,
"best_step": 12500,
"circuit_type": "T2 late-routing local adder (HAND-b variables)",
"acc_test_fp32": 1.0,
"recorded_test": 1.0,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "7bd04dadf067b2aa334c755bc3e267ccfb7708e41c24d47e2f2c3828b1d4557b"
},
{
"file": "train_T_r1_final_s1.safetensors",
"rung": "TRAIN",
"fmt": "T",
"retry": "r1",
"wd": 0.1,
"snapshot": "final",
"seed": 1,
"best_step": 30000,
"circuit_type": "unsolved (.9155)",
"acc_test_fp32": 0.9155,
"recorded_test": 0.9155,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "b5b2ced390bfe18821e45b4b5deb47056eccd730de0722685d6448482141989e"
},
{
"file": "train_T_r1_final_s2.safetensors",
"rung": "TRAIN",
"fmt": "T",
"retry": "r1",
"wd": 0.1,
"snapshot": "final",
"seed": 2,
"best_step": 21000,
"circuit_type": "T2 late-routing local adder (HAND-b variables)",
"acc_test_fp32": 1.0,
"recorded_test": 1.0,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "d16660eeee0cf9c0e98fa811e914015f89037ec50dae69ac80ae9208dcf58f71"
},
{
"file": "train_A_r3_best_s0.safetensors",
"rung": "TRAIN",
"fmt": "A",
"retry": "r3",
"wd": 0.1,
"snapshot": "best",
"seed": 0,
"best_step": 90500,
"circuit_type": "A1 route, unsolved (.9555; bit-2 carry errors)",
"acc_test_fp32": 0.9585,
"recorded_test": 0.9585,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "140b26e0746367db69e63823de7e6d5fd2805d7d5f35ec9a6cb72f03c104527d"
},
{
"file": "train_A_r3_best_s1.safetensors",
"rung": "TRAIN",
"fmt": "A",
"retry": "r3",
"wd": 0.1,
"snapshot": "best",
"seed": 1,
"best_step": 78500,
"circuit_type": "A1 exact products + 2-d nonlinear row-sum code",
"acc_test_fp32": 1.0,
"recorded_test": 1.0,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "93b4b089a897e3a5abbfee0ec11ddfe439720c17c4dc3221fadfe5746bfd9200"
},
{
"file": "train_A_r3_best_s2.safetensors",
"rung": "TRAIN",
"fmt": "A",
"retry": "r3",
"wd": 0.1,
"snapshot": "best",
"seed": 2,
"best_step": 73000,
"circuit_type": "A1 (.999)",
"acc_test_fp32": 0.9995,
"recorded_test": 0.9995,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "e3686e9590126a826cc293af399b11d2c8f0f26303a7c59fcad604dec981ed3c"
},
{
"file": "train_A_r3_final_s0.safetensors",
"rung": "TRAIN",
"fmt": "A",
"retry": "r3",
"wd": 0.1,
"snapshot": "final",
"seed": 0,
"best_step": 90500,
"circuit_type": "A1 route, unsolved (.9555; bit-2 carry errors)",
"acc_test_fp32": 0.9555,
"recorded_test": 0.9555,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "048a8797bdc786c20ce0c9f4fa3ece3078a51c75ee30f2788d942f616501b66f"
},
{
"file": "train_A_r3_final_s1.safetensors",
"rung": "TRAIN",
"fmt": "A",
"retry": "r3",
"wd": 0.1,
"snapshot": "final",
"seed": 1,
"best_step": 78500,
"circuit_type": "A1 exact products + 2-d nonlinear row-sum code",
"acc_test_fp32": 1.0,
"recorded_test": 1.0,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "5998b9bdf5ad1e4a5a448c089c84374f14be5ddada5b5f59928553ddd8073cca"
},
{
"file": "train_A_r3_final_s2.safetensors",
"rung": "TRAIN",
"fmt": "A",
"retry": "r3",
"wd": 0.1,
"snapshot": "final",
"seed": 2,
"best_step": 73000,
"circuit_type": "A1 (.999)",
"acc_test_fp32": 0.999,
"recorded_test": 0.999,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "f0213703ef084dd0cbe1357985cb94c1054fb482331125bf8ee6623e4e1ad2d5"
},
{
"file": "train_T_final_best_s0.safetensors",
"rung": "TRAIN",
"fmt": "T",
"retry": "final",
"wd": 0.1,
"snapshot": "best",
"seed": 0,
"best_step": 12000,
"circuit_type": "T2 late-routing local adder (HAND-b variables)",
"acc_test_fp32": 1.0,
"recorded_test": 1.0,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "6e0bc7562d31543d464ef8d6e43beaa8a87d7dd21712ddd3efdefaac38ac8395"
},
{
"file": "train_T_final_best_s1.safetensors",
"rung": "TRAIN",
"fmt": "T",
"retry": "final",
"wd": 0.1,
"snapshot": "best",
"seed": 1,
"best_step": 28500,
"circuit_type": "T2 late-routing local adder (HAND-b variables)",
"acc_test_fp32": 1.0,
"recorded_test": 1.0,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "132b6cb3d0d49d6aa888bc3c2592f7814759e7e6258ff829cbf35d98a325ab84"
},
{
"file": "train_T_final_best_s2.safetensors",
"rung": "TRAIN",
"fmt": "T",
"retry": "final",
"wd": 0.1,
"snapshot": "best",
"seed": 2,
"best_step": 24000,
"circuit_type": "T2 late-routing local adder (HAND-b variables)",
"acc_test_fp32": 1.0,
"recorded_test": 1.0,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "30b95959af030b81d1e2dc1916fe3fcede1f04ed2e3b0477e8346eacf5b9c59f"
},
{
"file": "train_T_final_best_s3.safetensors",
"rung": "TRAIN",
"fmt": "T",
"retry": "final",
"wd": 0.1,
"snapshot": "best",
"seed": 3,
"best_step": 14000,
"circuit_type": "T2 late-routing local adder (HAND-b variables)",
"acc_test_fp32": 1.0,
"recorded_test": 1.0,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "3d4a6855da4601605799463ce8bc8ed85e590bdb46708b6f01e291b136837ef1"
},
{
"file": "train_T_final_best_s4.safetensors",
"rung": "TRAIN",
"fmt": "T",
"retry": "final",
"wd": 0.1,
"snapshot": "best",
"seed": 4,
"best_step": 14000,
"circuit_type": "T2 late-routing local adder (HAND-b variables)",
"acc_test_fp32": 1.0,
"recorded_test": 1.0,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "8ce2ee80ff66022ae988d784a4ebe2aacbb8f501475490c02ba178b9d1438759"
},
{
"file": "train_T_final_best_s5.safetensors",
"rung": "TRAIN",
"fmt": "T",
"retry": "final",
"wd": 0.1,
"snapshot": "best",
"seed": 5,
"best_step": 21000,
"circuit_type": "T2 late-routing local adder (HAND-b variables)",
"acc_test_fp32": 1.0,
"recorded_test": 1.0,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "bbe55459fcce3db4f8289fab17511f2784f5f074a8c452b52bab71a5a3ae1ead"
},
{
"file": "train_T_final_best_s6.safetensors",
"rung": "TRAIN",
"fmt": "T",
"retry": "final",
"wd": 0.1,
"snapshot": "best",
"seed": 6,
"best_step": 25000,
"circuit_type": "T2 late-routing local adder (HAND-b variables)",
"acc_test_fp32": 1.0,
"recorded_test": 1.0,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "b6f67755430c3729e134721dc2fcdb0a2b8a5e0f0926f58cb4e58df2da6ea896"
},
{
"file": "train_T_final_best_s7.safetensors",
"rung": "TRAIN",
"fmt": "T",
"retry": "final",
"wd": 0.1,
"snapshot": "best",
"seed": 7,
"best_step": 29000,
"circuit_type": "T2 variables, subexact final (.9995; best 1.0)",
"acc_test_fp32": 1.0,
"recorded_test": 1.0,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "632124397e581600c2c31e5ccba2e5d422dbc8fbda27ec3a6836372aa277940e"
},
{
"file": "train_T_final_best_s8.safetensors",
"rung": "TRAIN",
"fmt": "T",
"retry": "final",
"wd": 0.1,
"snapshot": "best",
"seed": 8,
"best_step": 11500,
"circuit_type": "T2 late-routing local adder (HAND-b variables)",
"acc_test_fp32": 1.0,
"recorded_test": 1.0,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "32e28fec3a49336a0457b83c2f24be649df1bed1802d65c26557120f0479e5c3"
},
{
"file": "train_T_final_best_s9.safetensors",
"rung": "TRAIN",
"fmt": "T",
"retry": "final",
"wd": 0.1,
"snapshot": "best",
"seed": 9,
"best_step": 30000,
"circuit_type": "unsolved (.815; incomplete low-bit carry composition)",
"acc_test_fp32": 0.815,
"recorded_test": 0.815,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "082fcbfaffaac4ffc8c55d6e7bf7c081d2fdbb2a49996c3111303772b15e9676"
},
{
"file": "train_T_final_final_s0.safetensors",
"rung": "TRAIN",
"fmt": "T",
"retry": "final",
"wd": 0.1,
"snapshot": "final",
"seed": 0,
"best_step": 12000,
"circuit_type": "T2 late-routing local adder (HAND-b variables)",
"acc_test_fp32": 1.0,
"recorded_test": 1.0,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "01277e9c81fc386a2b93c265adf4b354e721cc9308d3c515dae3a3c18216bb64"
},
{
"file": "train_T_final_final_s1.safetensors",
"rung": "TRAIN",
"fmt": "T",
"retry": "final",
"wd": 0.1,
"snapshot": "final",
"seed": 1,
"best_step": 28500,
"circuit_type": "T2 late-routing local adder (HAND-b variables)",
"acc_test_fp32": 1.0,
"recorded_test": 1.0,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "8e88eacddfa664403d1f9004e4edb043ad353ff8a3b3d53d9971c9dc085c4e85"
},
{
"file": "train_T_final_final_s2.safetensors",
"rung": "TRAIN",
"fmt": "T",
"retry": "final",
"wd": 0.1,
"snapshot": "final",
"seed": 2,
"best_step": 24000,
"circuit_type": "T2 late-routing local adder (HAND-b variables)",
"acc_test_fp32": 1.0,
"recorded_test": 1.0,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "fdb4ebe831f4623c4c2315685443252dc532254331e04162059a3f587c816ba0"
},
{
"file": "train_T_final_final_s3.safetensors",
"rung": "TRAIN",
"fmt": "T",
"retry": "final",
"wd": 0.1,
"snapshot": "final",
"seed": 3,
"best_step": 14000,
"circuit_type": "T2 late-routing local adder (HAND-b variables)",
"acc_test_fp32": 1.0,
"recorded_test": 1.0,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "27d9d71d743dc0b39b3f6ce5ce56035667d21b5a04a385d8270e46486983cd56"
},
{
"file": "train_T_final_final_s4.safetensors",
"rung": "TRAIN",
"fmt": "T",
"retry": "final",
"wd": 0.1,
"snapshot": "final",
"seed": 4,
"best_step": 14000,
"circuit_type": "T2 late-routing local adder (HAND-b variables)",
"acc_test_fp32": 1.0,
"recorded_test": 1.0,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "f45cd49ce761d66a424898d921d365a23df331f22c59dfbacd39a05e15c5d203"
},
{
"file": "train_T_final_final_s5.safetensors",
"rung": "TRAIN",
"fmt": "T",
"retry": "final",
"wd": 0.1,
"snapshot": "final",
"seed": 5,
"best_step": 21000,
"circuit_type": "T2 late-routing local adder (HAND-b variables)",
"acc_test_fp32": 1.0,
"recorded_test": 1.0,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "31a0c862b6c5ac7c77bbac664912a6fb8aa152b8e6334c167d0e2c5bb29e1d18"
},
{
"file": "train_T_final_final_s6.safetensors",
"rung": "TRAIN",
"fmt": "T",
"retry": "final",
"wd": 0.1,
"snapshot": "final",
"seed": 6,
"best_step": 25000,
"circuit_type": "T2 late-routing local adder (HAND-b variables)",
"acc_test_fp32": 1.0,
"recorded_test": 1.0,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "a3dce037b5bba9b9143022f203a6e6884c05581fb83d3a4eca84fcdc4c1f229e"
},
{
"file": "train_T_final_final_s7.safetensors",
"rung": "TRAIN",
"fmt": "T",
"retry": "final",
"wd": 0.1,
"snapshot": "final",
"seed": 7,
"best_step": 29000,
"circuit_type": "T2 variables, subexact final (.9995; best 1.0)",
"acc_test_fp32": 0.9995,
"recorded_test": 0.9995,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "82aca6d0fac97d90f2196de163688a0f1ba0b6b0036b3c5383a5287cea1b998b"
},
{
"file": "train_T_final_final_s8.safetensors",
"rung": "TRAIN",
"fmt": "T",
"retry": "final",
"wd": 0.1,
"snapshot": "final",
"seed": 8,
"best_step": 11500,
"circuit_type": "T2 late-routing local adder (HAND-b variables)",
"acc_test_fp32": 1.0,
"recorded_test": 1.0,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "31b680cad73194d2f591802f6cc29fbea4dc363fa5bb77de21db7cd17253e639"
},
{
"file": "train_T_final_final_s9.safetensors",
"rung": "TRAIN",
"fmt": "T",
"retry": "final",
"wd": 0.1,
"snapshot": "final",
"seed": 9,
"best_step": 30000,
"circuit_type": "unsolved (.815; incomplete low-bit carry composition)",
"acc_test_fp32": 0.815,
"recorded_test": 0.815,
"trained_tf32": true,
"bit_identical": true,
"roundtrip_ok": true,
"sha256": "082fcbfaffaac4ffc8c55d6e7bf7c081d2fdbb2a49996c3111303772b15e9676"
}
],
"roundtrip_all_ok": true
}