Trained / exported adapter / adapter-e4b_true_opd_v3

e4b_true_opd_v3

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?

Transfer the larger model’s predictions on student repair trajectories, including unresolved attempts.

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?

Hybrid online prediction distillation from the 31B teacher; 36 trajectories (12 resolved, 24 unresolved), 3 epochs, 27 optimizer steps. Exported low-rank adapter, rank 64, scaling alpha 32.

Did it improve?

Historical measurements · Historical local result: 11/45 and 2/8 cross-validation. Archived Kaggle score: 0.06. Not evidence of a reliable paired improvement.

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 They may contain original predictions and superseded counts; use the measured-outcome summary and corrections above for the current result.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

| 4 | **`e4b_true_opd_v3`** | Nov 1 | Hybrid Online `31B` OPD | `E4B` `r=64, a=32` | `opd_e4b_hybrid_v3` | `36` (`12` res + `24` unres, `28` tasks) | `0.23M` | `3` (`27`) | `0.70M` | `0.64h` | `11/45` (`5/15` + `6/30`), `2/8` CV | **`0.06`** (`57908452`) |

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_v3",
  "wrx90_path": "/home/keith/gemma-v2/adapters/e4b_true_opd_v3",
  "files": {
    "adapter_model.safetensors": {
      "size_bytes": 279129832,
      "sha256": "7b25d67290b0d147d2968d0d78d2244839b969e5640f6f45e43228337e786572"
    }
  }
}

Training metrics — original archive, not task improvement

{
  "dataset_size": 36,
  "elapsed_seconds": 2316.56,
  "epochs": 3,
  "export": {
    "adapter_dir": "/home/keith/gemma-v2/adapters/e4b_true_opd_v3",
    "keqv_v_proj_cloned_layers": 0,
    "safetensors_bytes": 279129832,
    "tensor_count": 516
  },
  "history_head": [
    {
      "ce_loss": 0.6015,
      "epoch": 1,
      "grad_norm": 0.2129,
      "loss": 0.7009,
      "lr_a": 2.5e-06,
      "opt_step": 1,
      "rkl_loss": 0.8761
    },
    {
      "ce_loss": 0.2632,
      "epoch": 1,
      "grad_norm": 0.209,
      "loss": 0.535,
      "lr_a": 5e-06,
      "opt_step": 2,
      "rkl_loss": 0.6687
    },
    {
      "ce_loss": 0.3731,
      "epoch": 1,
      "grad_norm": 0.2148,
      "loss": 0.3976,
      "lr_a": 5e-06,
      "opt_step": 3,
      "rkl_loss": 0.5383
    }
  ],
  "history_tail": [
    {
      "ce_loss": 0.2997,
      "epoch": 3,
      "grad_norm": 0.1973,
      "loss": 0.4484,
      "lr_a": 9.3e-07,
      "opt_step": 23,
      "rkl_loss": 0.5603
    },
    {
      "ce_loss": 0.352,
      "epoch": 3,
      "grad_norm": 0.165,
      "loss": 0.4988,
      "lr_a": 7.8e-07,
      "opt_step": 24,
      "rkl_loss": 0.6232
    },
    {
      "ce_loss": 0.2813,
      "epoch": 3,
      "grad_norm": 0.1885,
      "loss": 0.4798,
      "lr_a": 6.6e-07,
      "opt_step": 25,
      "rkl_loss": 0.5995
    },
    {
      "ce_loss": 0.3087,
      "epoch": 3,
      "grad_norm": 0.1826,
      "loss": 0.5359,
      "lr_a": 5.7e-07,
      "opt_step": 26,
      "rkl_loss": 0.6696
    },
    {
      "ce_loss": 0.5629,
      "epoch": 3,
      "grad_norm": 0.2422,
      "loss": 0.6895,
      "lr_a": 5.2e-07,
      "opt_step": 27,
      "rkl_loss": 0.8615
    }
  ],
  "injected_module_count": 258,
  "mode": "dual_gpu_opd",
  "opt_steps": 27,
  "student_alias": "gemma-4-e4b-it-qat-w4a16-ct"
}