tokenfold-select-v2 / training-report.json
snchimata's picture
Release Tokenfold Select v2 ONNX and Safetensors preview
78736d9 verified
Raw History Blame Contribute Delete
21.8 kB
{
"status": "completed_development_only",
"quality": "unqualified",
"qualification": false,
"mode": "supervised-foundation-feature-distilled",
"recipe": {
"seed": 0,
"epochs": 3,
"batch": 1,
"accumulation": 8,
"optimizer": "AdamW",
"learning_rate": 1e-05,
"weight_decay": 0.01,
"betas": [
0.9,
0.999
],
"eps": 1e-08,
"schedule": "constant",
"precision": "fp32",
"backend": "sdpa",
"gradient_checkpointing": true,
"clip_grad_norm": 1.0,
"student_token_cap": 8192,
"scoring_window_profile": "source-tail-1536-capacity-v2",
"scoring_context_token_cap": 1536,
"scoring_window_code_sha256": "fa879fb896e305a60ff10f3f5a6a23887da11b9741a210e2dc058154e148d929",
"supervision": "row-global optional class-balanced BCE plus mean window-local hardest optional pair softplus",
"feature_weight": 0.1,
"distillation": "original foundation encoder pooled features; no teacher head/logits",
"feature_reduction": "row-mean cosine distance over all original groups",
"gradient_reduction": "row mean then accumulation-block mean; one live window graph",
"checkpoint_criterion": "first minimum DEV balanced BCE plus window-local hard-pair row mean; provisional only",
"early_stopping": "none",
"test_access": false,
"qualification": false,
"objective_profile": "optional-balanced-window-hardpair-headwarm-v4",
"progress_publication": "immutable ready-marked snapshots plus terminal-only report",
"hard_pair_weight": 1.0,
"hard_pair_margin": 1.0,
"hard_pair_loss": "softplus(margin + max known optional negative - min known optional positive)",
"hard_pair_reduction": "row mean over windows containing both known optional classes; zero if none; window-local, not cross-window",
"hard_pair_role_admission": "at least one pair-bearing window in each TRAIN/DEV role; per-family coverage separately reviewed",
"head_warmup": {
"source": "fresh frozen ORIGINAL initializer encoder; TRAIN only; no saved cache or trained parent",
"objective": "row-global optional balanced BCE plus L2/2 on weight and bias",
"l2": 0.0001,
"device": "cpu",
"precision": "float64",
"optimizer": "LBFGS",
"learning_rate": 1.0,
"max_iter": 200,
"max_eval": 500,
"history_size": 20,
"line_search_fn": "strong_wolfe",
"tolerance_grad": 1e-07,
"tolerance_change": 1e-09,
"closure_cap": 512,
"max_design_bytes": 134217728,
"installation": "head-only FP32 copy; encoder unchanged; before teacher and joint AdamW creation",
"selection": "warm-only state ineligible; first minimum among three full joint DEV epochs",
"convergence_certified": false
},
"joint_gradient_evidence": "per AdamW update head/encoder preclip and total pre/postclip L2 norms"
},
"approval_sha256": "6fc0e73fbbb0b4f6f892720ff396173e5bb890509c2125a51cb9fabd7af12875",
"input_preflight": {
"kind": "clean-source-context-packing-not-training",
"training_approved": false,
"qualification": false,
"redistribution_cleared": false,
"legacy_trainer_integrated": false,
"manifest_sha256": "ec3ec873a98846218b3e70a07fa37129bc756d4e199b72e6a2174638cb534b47",
"rows": {
"train": 143,
"validation": 87
},
"source_clusters": {
"train": 12,
"validation": 6
},
"original_source_admission_refusals": {
"train": 9,
"validation": 7
},
"original_masked_groups": {
"train": 1021,
"validation": 48
},
"admitted_counts": {
"train": 143,
"validation": 87
},
"packing_refusals": [],
"encoder_windows": {
"train": 377,
"validation": 87
},
"source_context_rows": {
"train": 54,
"validation": 0
},
"max_window_tokens": 8159,
"scoring_window_profile": "source-tail-1536-capacity-v2"
},
"approval_kind": "select-v2-clean-context-development-training-approval",
"G5_complete": false,
"development_only": true,
"release_approved": false,
"initialization_recipe": {
"kind": "untrained compact12 initialization recipe, NOT a training recipe or deployment approval",
"foundation_repo": "ibm-granite/granite-embedding-reranker-english-r2",
"foundation_revision": "d09d3d6971b689bf9c23839e45a470874d46e13a",
"foundation_weights_sha256": "e72c8f7087dbfbdbfb9d75077f7b7b05b01b459730f735b468cbde71d66090e6",
"seed": 0,
"device": "cpu",
"dtype": "torch.float32",
"attention": "sdpa",
"retained_original_encoder_layers": [
0,
1,
2,
3,
4,
5,
6,
7,
8,
9,
10,
11
],
"model_class": "JointGroupModel",
"head": "torch.nn.Linear(768,1), fresh seed0",
"optimizer_steps": 0,
"encoder_forwards": 0,
"teacher_parents": [],
"local_training_data_parents": [],
"source_sha256": {
"eval/research/joint_group_prototype.py": "a90cf1ed3274e4dcd9a968e2b51176abf8832bfa36076e07be1467d1c49a4d64",
"eval/learned/training/train_select_v2_joint.py": "78521741b6a5cc7d92278d32b4eff0408f2cdda32a1e38dc7e10b679d445b28f"
},
"canonical_state_tensor_sha256": "ee8d442944df727d0c81101f6ab4521f1493170645054747109b4a8294c726d2",
"initialization_weights_sha256": "4edc5fe495eb8e6b6492890735a4514bc23eb2f23609a7c1ae624f3935c8e0d2",
"tokenizer_sha256": {
"tokenizer.json": "39bc3f741c62ba10b6f412f2bfaf2ee0e3241bd3801c3a056bb4e0dbdabcf416",
"tokenizer_config.json": "fd514e3fd112962cfd80ee86828cc970dfdfa2b8176eb5a469d200b241c656b5",
"special_tokens_map.json": "ea97ecdbcc73713039d8d64dbb05e3689495c96657fbd9a18f5bed381be81049"
},
"environment": [
"torch==2.11.0+cu128",
"transformers==5.19.0",
"safetensors==0.8.0",
"tokenizers==0.23.2",
"huggingface_hub==1.33.0",
"numpy==2.5.3"
],
"verification_report_sha256": "76d21f95a511badd96541e0655d3aac32791b580b7e8944ed65e797d888f4e97",
"qualification": false,
"training_approved": false,
"redistribution_cleared": false,
"provider_license_evidence": "Pinned provider card identifies Apache-2.0; full upstream data lineage and future training clearance are not certified."
},
"source_sha256": {
"eval/research/select_v2_clean_training.py": "40813f8fa917ca59f5a1eb9db74ec0f2ef4295e369a3f67ee5efd95c23e39a22",
"eval/learned/training/train_select_v2_joint.py": "78521741b6a5cc7d92278d32b4eff0408f2cdda32a1e38dc7e10b679d445b28f",
"eval/research/select_v2_supervised_corpus.py": "17e3510aeb2e30361a7ed2353054e502c9502bb0d7200e120e9f27f72cbe2683",
"eval/research/select_v2_context_windows.py": "fa879fb896e305a60ff10f3f5a6a23887da11b9741a210e2dc058154e148d929",
"eval/research/joint_group_prototype.py": "a90cf1ed3274e4dcd9a968e2b51176abf8832bfa36076e07be1467d1c49a4d64",
"eval/research/select_v2_campaign.py": "98eafa4c96444353db26b40e2d1353fe3cab57da117add8fe8b492ae093b0ad9"
},
"joint_model_code_sha256": "a90cf1ed3274e4dcd9a968e2b51176abf8832bfa36076e07be1467d1c49a4d64",
"epochs": [
{
"epoch": 0,
"train_objective": 1.493269572704477,
"validation_known_bce": 0.40362530379374134,
"validation_optional_balanced_bce": 0.405680693193585,
"validation_window_hardpair": 1.9300384357404552,
"validation_combined_objective": 2.3357191139982008
},
{
"epoch": 1,
"train_objective": 1.2188779098405893,
"validation_known_bce": 0.5550572217024606,
"validation_optional_balanced_bce": 0.4080781711843507,
"validation_window_hardpair": 1.7441693844465578,
"validation_combined_objective": 2.1522475449346947
},
{
"epoch": 2,
"train_objective": 0.9576328355990761,
"validation_known_bce": 0.4114132965165565,
"validation_optional_balanced_bce": 0.3547273740043928,
"validation_window_hardpair": 1.7159358712784916,
"validation_combined_objective": 2.0706632323616625
}
],
"optimizer_steps": 54,
"encoder_forwards": 1769,
"teacher_forwards": 1131,
"redistribution_cleared": false,
"torch_version": "2.11.0+cu128",
"cuda_version": "12.8",
"admitted_counts": {
"train": 143,
"validation": 87
},
"window_hardpair_coverage": {
"train": {
"rows": 143,
"windows": 377,
"rows_with_pairs": 139,
"pair_windows": 146
},
"validation": {
"rows": 87,
"windows": 87,
"rows_with_pairs": 83,
"pair_windows": 83
}
},
"head_warmup": {
"closure_calls": 231,
"iterations": 200,
"starting_regularized_objective": 0.7344952593492643,
"starting_balanced_bce": 0.7344779931390228,
"regularized_objective": 0.25508015771677495,
"balanced_bce": 0.20164778963500846,
"gradient_l2": 0.00725551751843674,
"gradient_linf": 0.004432530809740342,
"convergence_certified": false,
"status": "completed_TRAIN_head_warmup",
"encoder_forwards": 377,
"teacher_forwards": 0,
"DEV_encoder_forwards": 0,
"rows": 143,
"groups": 8831,
"fitted_parameters": 769,
"installed_precision": "float32",
"cache_sha256": "46b6873f5b22e07a85c16c034d2606ce2a2b294225f3a6ee505ce8e9e0ebddb9"
},
"joint_gradient_evidence": [
{
"step": 1,
"head_preclip_l2": 1.0165764093399048,
"encoder_preclip_l2": 34.34036636352539,
"total_preclip_l2": 34.355411529541016,
"total_postclip_l2": 1.0
},
{
"step": 2,
"head_preclip_l2": 4.268938064575195,
"encoder_preclip_l2": 62.07624816894531,
"total_preclip_l2": 62.222862243652344,
"total_postclip_l2": 1.0
},
{
"step": 3,
"head_preclip_l2": 2.611722469329834,
"encoder_preclip_l2": 51.621131896972656,
"total_preclip_l2": 51.68716049194336,
"total_postclip_l2": 1.0
},
{
"step": 4,
"head_preclip_l2": 2.866244316101074,
"encoder_preclip_l2": 46.00312423706055,
"total_preclip_l2": 46.09233093261719,
"total_postclip_l2": 1.0
},
{
"step": 5,
"head_preclip_l2": 1.4913560152053833,
"encoder_preclip_l2": 40.564170837402344,
"total_preclip_l2": 40.591575622558594,
"total_postclip_l2": 1.0
},
{
"step": 6,
"head_preclip_l2": 3.1226420402526855,
"encoder_preclip_l2": 64.21011352539062,
"total_preclip_l2": 64.28599548339844,
"total_postclip_l2": 1.0
},
{
"step": 7,
"head_preclip_l2": 2.2895054817199707,
"encoder_preclip_l2": 40.162174224853516,
"total_preclip_l2": 40.227378845214844,
"total_postclip_l2": 0.9999999403953552
},
{
"step": 8,
"head_preclip_l2": 2.5300004482269287,
"encoder_preclip_l2": 62.54695510864258,
"total_preclip_l2": 62.59810256958008,
"total_postclip_l2": 1.0
},
{
"step": 9,
"head_preclip_l2": 1.420209527015686,
"encoder_preclip_l2": 42.02363586425781,
"total_preclip_l2": 42.04762649536133,
"total_postclip_l2": 1.0
},
{
"step": 10,
"head_preclip_l2": 1.2796767950057983,
"encoder_preclip_l2": 38.142173767089844,
"total_preclip_l2": 38.163631439208984,
"total_postclip_l2": 1.0000001192092896
},
{
"step": 11,
"head_preclip_l2": 3.592926502227783,
"encoder_preclip_l2": 65.71934509277344,
"total_preclip_l2": 65.8174819946289,
"total_postclip_l2": 0.9999999403953552
},
{
"step": 12,
"head_preclip_l2": 1.7335712909698486,
"encoder_preclip_l2": 52.37236785888672,
"total_preclip_l2": 52.40105056762695,
"total_postclip_l2": 1.0
},
{
"step": 13,
"head_preclip_l2": 2.1708743572235107,
"encoder_preclip_l2": 51.65614318847656,
"total_preclip_l2": 51.70174026489258,
"total_postclip_l2": 1.0
},
{
"step": 14,
"head_preclip_l2": 1.3823407888412476,
"encoder_preclip_l2": 39.098941802978516,
"total_preclip_l2": 39.12337112426758,
"total_postclip_l2": 1.0
},
{
"step": 15,
"head_preclip_l2": 1.2050923109054565,
"encoder_preclip_l2": 39.03215026855469,
"total_preclip_l2": 39.050750732421875,
"total_postclip_l2": 0.9999999403953552
},
{
"step": 16,
"head_preclip_l2": 1.6002739667892456,
"encoder_preclip_l2": 40.18476486206055,
"total_preclip_l2": 40.216617584228516,
"total_postclip_l2": 1.0
},
{
"step": 17,
"head_preclip_l2": 1.3009775876998901,
"encoder_preclip_l2": 46.21324920654297,
"total_preclip_l2": 46.23155975341797,
"total_postclip_l2": 1.0
},
{
"step": 18,
"head_preclip_l2": 1.8785698413848877,
"encoder_preclip_l2": 64.73999786376953,
"total_preclip_l2": 64.76724243164062,
"total_postclip_l2": 1.0
},
{
"step": 19,
"head_preclip_l2": 1.306772232055664,
"encoder_preclip_l2": 38.571048736572266,
"total_preclip_l2": 38.593177795410156,
"total_postclip_l2": 1.0
},
{
"step": 20,
"head_preclip_l2": 2.69498872756958,
"encoder_preclip_l2": 65.64171600341797,
"total_preclip_l2": 65.69701385498047,
"total_postclip_l2": 0.9999999403953552
},
{
"step": 21,
"head_preclip_l2": 2.575758695602417,
"encoder_preclip_l2": 51.75717544555664,
"total_preclip_l2": 51.821231842041016,
"total_postclip_l2": 1.0
},
{
"step": 22,
"head_preclip_l2": 1.6727721691131592,
"encoder_preclip_l2": 39.11467742919922,
"total_preclip_l2": 39.150428771972656,
"total_postclip_l2": 1.0
},
{
"step": 23,
"head_preclip_l2": 0.969562828540802,
"encoder_preclip_l2": 30.527206420898438,
"total_preclip_l2": 30.542598724365234,
"total_postclip_l2": 0.9999999403953552
},
{
"step": 24,
"head_preclip_l2": 1.8666222095489502,
"encoder_preclip_l2": 69.53941345214844,
"total_preclip_l2": 69.56446838378906,
"total_postclip_l2": 0.9999998807907104
},
{
"step": 25,
"head_preclip_l2": 2.3292839527130127,
"encoder_preclip_l2": 67.52586364746094,
"total_preclip_l2": 67.56602478027344,
"total_postclip_l2": 1.0
},
{
"step": 26,
"head_preclip_l2": 3.2683966159820557,
"encoder_preclip_l2": 45.52153015136719,
"total_preclip_l2": 45.63871383666992,
"total_postclip_l2": 1.0
},
{
"step": 27,
"head_preclip_l2": 3.3452396392822266,
"encoder_preclip_l2": 46.847129821777344,
"total_preclip_l2": 46.96641540527344,
"total_postclip_l2": 1.0
},
{
"step": 28,
"head_preclip_l2": 2.055723190307617,
"encoder_preclip_l2": 46.38289260864258,
"total_preclip_l2": 46.42842483520508,
"total_postclip_l2": 1.0
},
{
"step": 29,
"head_preclip_l2": 1.6482597589492798,
"encoder_preclip_l2": 42.32853698730469,
"total_preclip_l2": 42.36061477661133,
"total_postclip_l2": 0.9999999403953552
},
{
"step": 30,
"head_preclip_l2": 1.2157902717590332,
"encoder_preclip_l2": 36.048770904541016,
"total_preclip_l2": 36.06926727294922,
"total_postclip_l2": 1.0
},
{
"step": 31,
"head_preclip_l2": 3.820483684539795,
"encoder_preclip_l2": 60.90419387817383,
"total_preclip_l2": 61.02390670776367,
"total_postclip_l2": 1.0
},
{
"step": 32,
"head_preclip_l2": 5.021407604217529,
"encoder_preclip_l2": 80.17427062988281,
"total_preclip_l2": 80.33136749267578,
"total_postclip_l2": 1.0
},
{
"step": 33,
"head_preclip_l2": 5.283500671386719,
"encoder_preclip_l2": 75.29106903076172,
"total_preclip_l2": 75.47622680664062,
"total_postclip_l2": 1.0
},
{
"step": 34,
"head_preclip_l2": 2.726933479309082,
"encoder_preclip_l2": 47.34463882446289,
"total_preclip_l2": 47.4231071472168,
"total_postclip_l2": 1.0
},
{
"step": 35,
"head_preclip_l2": 2.505049228668213,
"encoder_preclip_l2": 55.8592414855957,
"total_preclip_l2": 55.91538619995117,
"total_postclip_l2": 1.0
},
{
"step": 36,
"head_preclip_l2": 0.37975412607192993,
"encoder_preclip_l2": 9.456351280212402,
"total_preclip_l2": 9.463973999023438,
"total_postclip_l2": 0.9999998211860657
},
{
"step": 37,
"head_preclip_l2": 1.4222089052200317,
"encoder_preclip_l2": 25.609291076660156,
"total_preclip_l2": 25.648752212524414,
"total_postclip_l2": 0.9999998807907104
},
{
"step": 38,
"head_preclip_l2": 1.8505936861038208,
"encoder_preclip_l2": 25.482921600341797,
"total_preclip_l2": 25.550029754638672,
"total_postclip_l2": 0.9999999403953552
},
{
"step": 39,
"head_preclip_l2": 3.343179702758789,
"encoder_preclip_l2": 82.03128814697266,
"total_preclip_l2": 82.0993881225586,
"total_postclip_l2": 0.9999999403953552
},
{
"step": 40,
"head_preclip_l2": 2.5391385555267334,
"encoder_preclip_l2": 42.44383239746094,
"total_preclip_l2": 42.51971435546875,
"total_postclip_l2": 1.0
},
{
"step": 41,
"head_preclip_l2": 1.0755114555358887,
"encoder_preclip_l2": 19.83341407775879,
"total_preclip_l2": 19.862552642822266,
"total_postclip_l2": 0.9999998807907104
},
{
"step": 42,
"head_preclip_l2": 0.8598279356956482,
"encoder_preclip_l2": 28.788406372070312,
"total_preclip_l2": 28.801244735717773,
"total_postclip_l2": 0.9999998807907104
},
{
"step": 43,
"head_preclip_l2": 0.40072038769721985,
"encoder_preclip_l2": 13.1334228515625,
"total_preclip_l2": 13.139534950256348,
"total_postclip_l2": 0.9999999403953552
},
{
"step": 44,
"head_preclip_l2": 3.324460506439209,
"encoder_preclip_l2": 57.445472717285156,
"total_preclip_l2": 57.541587829589844,
"total_postclip_l2": 0.9999999403953552
},
{
"step": 45,
"head_preclip_l2": 2.296888589859009,
"encoder_preclip_l2": 44.85797119140625,
"total_preclip_l2": 44.91674041748047,
"total_postclip_l2": 0.9999999403953552
},
{
"step": 46,
"head_preclip_l2": 0.8208965063095093,
"encoder_preclip_l2": 20.56821632385254,
"total_preclip_l2": 20.584590911865234,
"total_postclip_l2": 0.9999999403953552
},
{
"step": 47,
"head_preclip_l2": 2.849490165710449,
"encoder_preclip_l2": 56.22677993774414,
"total_preclip_l2": 56.2989387512207,
"total_postclip_l2": 0.9999999403953552
},
{
"step": 48,
"head_preclip_l2": 3.7718403339385986,
"encoder_preclip_l2": 77.8814926147461,
"total_preclip_l2": 77.9727783203125,
"total_postclip_l2": 0.9999999403953552
},
{
"step": 49,
"head_preclip_l2": 1.942216396331787,
"encoder_preclip_l2": 44.697425842285156,
"total_preclip_l2": 44.739601135253906,
"total_postclip_l2": 1.0
},
{
"step": 50,
"head_preclip_l2": 1.3803201913833618,
"encoder_preclip_l2": 30.323734283447266,
"total_preclip_l2": 30.355133056640625,
"total_postclip_l2": 0.9999999403953552
},
{
"step": 51,
"head_preclip_l2": 2.318917989730835,
"encoder_preclip_l2": 60.24837112426758,
"total_preclip_l2": 60.2929801940918,
"total_postclip_l2": 0.9999999403953552
},
{
"step": 52,
"head_preclip_l2": 0.3753167986869812,
"encoder_preclip_l2": 6.164238452911377,
"total_preclip_l2": 6.175653457641602,
"total_postclip_l2": 0.9999998807907104
},
{
"step": 53,
"head_preclip_l2": 1.1889710426330566,
"encoder_preclip_l2": 32.083858489990234,
"total_preclip_l2": 32.10588073730469,
"total_postclip_l2": 0.9999999403953552
},
{
"step": 54,
"head_preclip_l2": 2.0162439346313477,
"encoder_preclip_l2": 48.08409118652344,
"total_preclip_l2": 48.126346588134766,
"total_postclip_l2": 1.0
}
],
"tokenizer_sha256": {
"special_tokens_map.json": "ea97ecdbcc73713039d8d64dbb05e3689495c96657fbd9a18f5bed381be81049",
"tokenizer.json": "39bc3f741c62ba10b6f412f2bfaf2ee0e3241bd3801c3a056bb4e0dbdabcf416",
"tokenizer_config.json": "fd514e3fd112962cfd80ee86828cc970dfdfa2b8176eb5a469d200b241c656b5"
},
"config_sha256": "3f238bd83606d303cf8a71d4c82890f99c6539a0ca756a113be2e27168717ec0",
"parameter_count": 98864641,
"progress": {
"epoch": 2,
"rows_completed": 143
},
"selected_epoch": 2,
"checkpoint_sha256": "4dbdb5071e7d6316d8eb70dc161b1f2f77f555012f6616d7fa34311188e24cec",
"wall_seconds": 3548.015153999997,
"peak_allocated_bytes": 3791257088,
"peak_reserved_bytes": 4865392640
}