Trained / exported adapter / adapter-e4b_true_opd_v4

e4b_true_opd_v4

gemma-4-e4b-it-qat-w4a16-ctAuthor: Not recorded in adapter manifestState: Archived checkpointVerdict: Historical measurements

When

Recorded date: 2026-11-01

First archive commit: 2026-11-04T17:35:03-05:00 (offset shown).

Training run day: 2026-11-01 (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 online prediction distillation with the self-audit instruction strategy.

Which part changed?

The source describes adapter targets and mathematical changes. No before/after tensor comparison is available in this viewer.

How was it changed?

Train with the self_audit prompt on 74 trajectories (30 resolved, 44 unresolved); 2 epochs, 37 optimizer steps. Exported low-rank adapter, rank 64, scaling alpha 32.

Did it improve?

Historical measurements · Historical local result: 6/45. No reliable improvement established.

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

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 They may contain original predictions and superseded counts; use the measured-outcome summary and corrections above for the current result.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

| 5 | `e4b_true_opd_v4` | Nov 1 | Online OPD (`self_audit` prompt) | `E4B` `r=64, a=32` | `opd_e4b_self_audit_v4` | `74` (`30` res + `44` unres, `149` tasks) | `0.45M` | `2` (`37`) | `0.91M` | `0.89h` | `6/45` (`19` errs, `26.8` tc) | — |

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_true_opd_v4",
  "wrx90_path": "/home/keith/gemma-v2/adapters/e4b_true_opd_v4",
  "files": {
    "adapter_model.safetensors": {
      "size_bytes": 279129832,
      "sha256": "b80f522074ef4f2be73e4c3a77a5d6b6c2ed11a69492e4f73f4369b8460d3b6b"
    }
  }
}

Training metrics — original archive, not task improvement

{
  "dataset_size": 74,
  "elapsed_seconds": 3207.1,
  "epochs": 2,
  "export": {
    "adapter_dir": "/home/keith/gemma-v2/adapters/e4b_true_opd_v4",
    "keqv_v_proj_cloned_layers": 0,
    "safetensors_bytes": 279129832,
    "tensor_count": 516
  },
  "history_head": [
    {
      "active_tokens_cum": 21918,
      "ce_loss": 0.2583,
      "elapsed_sec": 266.7,
      "epoch": 1,
      "event": "opt_step",
      "grad_norm": 0.2256,
      "loss": 0.4819,
      "lr_a": 1.67e-06,
      "opt_step": 1,
      "rkl_loss": 0.7171,
      "total_opt_steps": 38,
      "unique_active_tokens": 454211
    },
    {
      "active_tokens_cum": 51352,
      "ce_loss": 0.3877,
      "elapsed_sec": 383.2,
      "epoch": 1,
      "event": "opt_step",
      "grad_norm": 0.21,
      "loss": 0.6944,
      "lr_a": 3.33e-06,
      "opt_step": 2,
      "rkl_loss": 0.868,
      "total_opt_steps": 38,
      "unique_active_tokens": 454211
    },
    {
      "active_tokens_cum": 79436,
      "ce_loss": 0.4883,
      "elapsed_sec": 495.8,
      "epoch": 1,
      "event": "opt_step",
      "grad_norm": 0.1953,
      "loss": 0.5662,
      "lr_a": 5e-06,
      "opt_step": 3,
      "rkl_loss": 0.7077,
      "total_opt_steps": 38,
      "unique_active_tokens": 454211
    }
  ],
  "history_tail": [
    {
      "active_tokens_cum": 802461,
      "ce_loss": 0.3449,
      "elapsed_sec": 3002.5,
      "epoch": 2,
      "event": "opt_step",
      "grad_norm": 0.0996,
      "loss": 0.3681,
      "lr_a": 8.2e-07,
      "opt_step": 33,
      "rkl_loss": 0.4677,
      "total_opt_steps": 38,
      "unique_active_tokens": 454211
    },
    {
      "active_tokens_cum": 823280,
      "ce_loss": 0.3442,
      "elapsed_sec": 3049.3,
      "epoch": 2,
      "event": "opt_step",
      "grad_norm": 0.1201,
      "loss": 0.3718,
      "lr_a": 7.2e-07,
      "opt_step": 34,
      "rkl_loss": 0.4381,
      "total_opt_steps": 38,
      "unique_active_tokens": 454211
    },
    {
      "active_tokens_cum": 862093,
      "ce_loss": 0.3487,
      "elapsed_sec": 3117.0,
      "epoch": 2,
      "event": "opt_step",
      "grad_norm": 0.0889,
      "loss": 0.3299,
      "lr_a": 6.4e-07,
      "opt_step": 35,
      "rkl_loss": 0.3929,
      "total_opt_steps": 38,
      "unique_active_tokens": 454211
    },
    {
      "active_tokens_cum": 891082,
      "ce_loss": 0.3155,
      "elapsed_sec": 3161.2,
      "epoch": 2,
      "event": "opt_step",
      "grad_norm": 0.0947,
      "loss": 0.3311,
      "lr_a": 5.8e-07,
      "opt_step": 36,
      "rkl_loss": 0.3952,
      "total_opt_steps": 38,
      "unique_active_tokens": 454211
    },
    {
      "active_tokens_cum": 908422,
      "ce_loss": 0.2969,
      "elapsed_sec": 3207.0,
      "epoch": 2,
      "event": "opt_step",
      "grad_norm": 0.0967,
      "loss": 0.3243,
      "lr_a": 5.4e-07,
      "opt_step": 37,
      "rkl_loss": 0.4129,
      "total_opt_steps": 38,
      "unique_active_tokens": 454211
    }
  ],
  "injected_module_count": 258,
  "mode": "dual_gpu_opd",
  "opt_steps": 37,
  "student_alias": "gemma-4-e4b-it-qat-w4a16-ct",
  "total_active_token_updates": 908422,
  "unique_active_tokens": 454211
}