Trained / exported adapter / adapter-e4b_pure_opd_v7b
e4b_pure_opd_v7b
gemma-4-e4b-it-qat-w4a16-ctAuthor: Not recorded in adapter manifestState: Archived checkpointVerdict: Historical measurements
When
Recorded date: 2026-11-03
First archive commit: 2026-11-04T17:35:03-05:00 (offset shown).
Training run day: 2026-11-03 (America/New_York). For derived variants, this is the parent run day, not the variant’s creation day.
First Git archive commit time (UTC offset shown), not the time training changed weights. Per-update wall timestamps were not preserved.
Exact tensor-change timestamp: unavailable.
Why change it?
Test competence-gated training on the student’s own policy rather than adding teacher trajectories.
Which part changed?
Adapter matrices over the frozen base, targeting: ^model\.language_model\.layers\.\d+\.(self_attn\.(q_proj|k_proj|v_proj|o_proj)|mlp\.(gate_proj|up_proj|down_proj))$
The source describes adapter targets and mathematical changes. No before/after tensor comparison is available in this viewer.
How was it changed?
485 trajectories (212 resolved, 273 unresolved); 2 epochs, 243 optimizer steps. Exported low-rank adapter, rank 64, scaling alpha 32.
Did it improve?
Historical measurements · Family results: norm-0.20 variants at both epochs scored 8/45. These are not an exact score identity for this raw export.
An exported checkpoint is not proof of a promoted policy or a complete optimizer-step history.
Evaluation set: Historical sets vary; see source table · Batch: Not applicable. Local screening and historical task scores are not competition scores unless explicitly identified in the source. Different evaluation sets cannot be compared as a gain.
Original evidence
- v2/experiments/adapters/e4b_pure_opd_v7b/checkpoint_manifest.json
- v2/experiments/adapters/e4b_pure_opd_v7b/adapter_config.json
- docs/v2-post-training-and-lb-postmortem.md
- v2/experiments/adapters/e4b_pure_opd_v7b/train_metrics.json
Literal source excerpts follow. They may contain original predictions and superseded counts; use the measured-outcome summary and corrections above for the current result.
Historical training and evaluation ledger — original table rows
| 10 | `e4b_pure_opd_v7b` | Nov 3 | Pure On-Policy Competence-Gated | `E4B` `r=64, a=32` | `opd_e4b_pure_v7b` | `485` (`212` stu res + `273` unres, `101` repos) | `2.54M` | `2` (`243`) | `5.07M` | `2.99h` | `ep1_n200: 8/45`, `ep2_n200: 8/45` | — |
Adapter configuration — original metadata
{
"base_model_name_or_path": "gemma-4-e4b-it-qat-w4a16-ct",
"bias": "none",
"lora_alpha": 32,
"lora_dropout": 0.0,
"peft_type": "LORA",
"r": 64,
"target_modules": "^model\\.language_model\\.layers\\.\\d+\\.(self_attn\\.(q_proj|k_proj|v_proj|o_proj)|mlp\\.(gate_proj|up_proj|down_proj))$",
"task_type": "CAUSAL_LM",
"use_rslora": false
}Checkpoint identity — original manifest
{
"adapter_id": "e4b_pure_opd_v7b",
"wrx90_path": "/home/keith/gemma-v2/adapters/e4b_pure_opd_v7b",
"files": {
"adapter_model.safetensors": {
"size_bytes": 279129832,
"sha256": "61951fc74a1aabdcbc22e6a4a9cdb451bdded55c9dbfdb1ed0f8bc3b168848ca"
}
},
"sub_checkpoints": {
"epoch_1": {
"adapter_model.safetensors": {
"size_bytes": 279129832
}
},
"epoch_1_norm020": {
"adapter_model.safetensors": {
"size_bytes": 279129832
}
},
"epoch_2": {
"adapter_model.safetensors": {
"size_bytes": 279129832
}
},
"epoch_2_norm020": {
"adapter_model.safetensors": {
"size_bytes": 279129832
}
},
"final_norm020": {
"adapter_model.safetensors": {
"size_bytes": 279129832
}
}
}
}Training metrics — original archive, not task improvement
{
"dataset_size": 485,
"elapsed_seconds": 10765.13,
"epochs": 2,
"export": {
"adapter_dir": "/home/keith/gemma-v2/adapters/e4b_pure_opd_v7b",
"keqv_v_proj_cloned_layers": 0,
"safetensors_bytes": 279129832,
"tensor_count": 516
},
"final_delta_w_frobenius_norm": 0.28,
"final_rms_layer_frobenius_norm": 0.017432,
"history_head": [
{
"active_tokens_cum": 23157,
"ce_loss": 0.307,
"cispo_loss": 0.0,
"delta_w_frob": 0.00084,
"delta_w_pre_proj": 0.00084,
"delta_w_projected": false,
"elapsed_sec": 213.5,
"epoch": 1,
"event": "opt_step",
"grad_norm": 0.2656,
"loss": 0.5288,
"lr_a": 1.7e-07,
"opt_step": 1,
"rkl_loss": 0.661,
"total_opt_steps": 244,
"unique_active_tokens": 2536401
},
{
"active_tokens_cum": 56206,
"ce_loss": 0.3545,
"cispo_loss": 0.0,
"delta_w_frob": 0.00207,
"delta_w_pre_proj": 0.00207,
"delta_w_projected": false,
"elapsed_sec": 269.2,
"epoch": 1,
"event": "opt_step",
"grad_norm": 0.2192,
"loss": 0.5233,
"lr_a": 3.3e-07,
"opt_step": 2,
"rkl_loss": 0.654,
"total_opt_steps": 244,
"unique_active_tokens": 2536401
},
{
"active_tokens_cum": 77170,
"ce_loss": 0.3247,
"cispo_loss": 0.0,
"delta_w_frob": 0.00371,
"delta_w_pre_proj": 0.00371,
"delta_w_projected": false,
"elapsed_sec": 317.1,
"epoch": 1,
"event": "opt_step",
"grad_norm": 0.2478,
"loss": 0.6098,
"lr_a": 5e-07,
"opt_step": 3,
"rkl_loss": 0.7621,
"total_opt_steps": 244,
"unique_active_tokens": 2536401
}
],
"history_tail": [
{
"active_tokens_cum": 4988710,
"ce_loss": 0.3496,
"cispo_loss": 0.0,
"delta_w_frob": 0.28,
"delta_w_pre_proj": 0.28101,
"delta_w_projected": true,
"elapsed_sec": 10602.9,
"epoch": 2,
"event": "opt_step",
"grad_norm": 0.3888,
"loss": 0.4454,
"lr_a": 4.1e-07,
"opt_step": 238,
"rkl_loss": 0.533,
"total_opt_steps": 244,
"unique_active_tokens": 2536401
},
{
"active_tokens_cum": 5010125,
"ce_loss": 0.3039,
"cispo_loss": 0.0,
"delta_w_frob": 0.28,
"delta_w_pre_proj": 0.281,
"delta_w_projected": true,
"elapsed_sec": 10653.6,
"epoch": 2,
"event": "opt_step",
"grad_norm": 0.3469,
"loss": 0.4622,
"lr_a": 4.1e-07,
"opt_step": 239,
"rkl_loss": 0.5505,
"total_opt_steps": 244,
"unique_active_tokens": 2536401
},
{
"active_tokens_cum": 5025815,
"ce_loss": 0.2638,
"cispo_loss": 0.0,
"delta_w_frob": 0.28,
"delta_w_pre_proj": 0.281,
"delta_w_projected": true,
"elapsed_sec": 10687.9,
"epoch": 2,
"event": "opt_step",
"grad_norm": 0.3438,
"loss": 0.3787,
"lr_a": 4e-07,
"opt_step": 240,
"rkl_loss": 0.4476,
"total_opt_steps": 244,
"unique_active_tokens": 2536401
},
{
"active_tokens_cum": 5047729,
"ce_loss": 0.4366,
"cispo_loss": 0.0,
"delta_w_frob": 0.28,
"delta_w_pre_proj": 0.28101,
"delta_w_projected": true,
"elapsed_sec": 10722.3,
"epoch": 2,
"event": "opt_step",
"grad_norm": 0.3927,
"loss": 0.4857,
"lr_a": 4e-07,
"opt_step": 241,
"rkl_loss": 0.5764,
"total_opt_steps": 244,
"unique_active_tokens": 2536401
},
{
"active_tokens_cum": 5063109,
"ce_loss": 0.2119,
"cispo_loss": 0.0,
"delta_w_frob": 0.28,
"delta_w_pre_proj": 0.281,
"delta_w_projected": true,
"elapsed_sec": 10745.8,
"epoch": 2,
"event": "opt_step",
"grad_norm": 0.3578,
"loss": 0.3433,
"lr_a": 4e-07,
"opt_step": 242,
"rkl_loss": 0.4121,
"total_opt_steps": 244,
"unique_active_tokens": 2536401
}
],
"injected_module_count": 258,
"max_qat_frobenius_norm": 0.28,
"mode": "dual_gpu_opd",
"opt_steps": 243,
"posthoc_norm020_export": {
"adapter_dir": "/home/keith/gemma-v2/adapters/e4b_pure_opd_v7b/final_norm020",
"keqv_v_proj_cloned_layers": 0,
"safetensors_bytes": 279129832,
"shrink_scale": 0.714286,
"target_frobenius_norm": 0.2,
"tensor_count": 516
},
"preflight_gate": {
"dataset_size": 485,
"eligible_unique_active_tokens": 2536401,
"epochs": 2,
"is_qat_model": true,
"kl_base_weight": 0.2,
"lora_scaling": 0.5,
"loraplus_ratio": 2.0,
"lr_a": 4e-06,
"lr_b": 8e-06,
"max_qat_frobenius_norm": 0.28,
"missing_superset_tasks": [],
"mode": "dual_gpu_opd",
"n_cispo_neg_rows": 0,
"n_offpolicy_teacher_rows": 0,
"n_resolved_rows": 212,
"n_unresolved_opd_rows": 273,
"opd_weight": 1.0,
"passed": true,
"planned_active_token_updates": 5072802,
"positive_mean_tool_calls": 7.93,
"resolved_instances": 78,
"unique_instances": 122,
"violations": []
},
"student_alias": "gemma-4-e4b-it-qat-w4a16-ct",
"total_active_token_updates": 5072802,
"trust_region_projections": 206,
"unique_active_tokens": 2536401
}