Trained / exported adapter / adapter-e4b_pure_opd_v6a

e4b_pure_opd_v6a

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

When

Recorded date: 2026-11-02

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

Training run day: 2026-11-02 (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 pure online prediction distillation without the negative-example objective.

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?

119 trajectories (54 resolved, 65 unresolved); 2 epochs, 60 optimizer steps. Exported low-rank adapter, rank 64, scaling alpha 32.

Did it improve?

Historical measurements · Historical local result: 8/45 for the raw adapter.

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 They may contain original predictions and superseded counts; use the measured-outcome summary and corrections above for the current result.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

| 7 | `e4b_pure_opd_v6a` | Nov 2 | Pure Online OPD (no neg) | `E4B` `r=64, a=32` | `opd_e4b_pure_v6a` | `119` (`54` res + `65` unres, `168` tasks) | `0.61M` | `2` (`60`) | `1.23M` | `1.00h` | `8/45` raw (`norm025`: `11/45`) | — |

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_v6a",
  "wrx90_path": "/home/keith/gemma-v2/adapters/e4b_pure_opd_v6a",
  "files": {
    "adapter_model.safetensors": {
      "size_bytes": 279129832,
      "sha256": "f5b58835f3deaca422ed5619c26515190a7d9827834754b248ffe19cb67af427"
    }
  }
}

Training metrics — original archive, not task improvement

{
  "dataset_size": 119,
  "elapsed_seconds": 3607.09,
  "epochs": 2,
  "export": {
    "adapter_dir": "/home/keith/gemma-v2/adapters/e4b_pure_opd_v6a",
    "keqv_v_proj_cloned_layers": 0,
    "safetensors_bytes": 279129832,
    "tensor_count": 516
  },
  "history_head": [
    {
      "active_tokens_cum": 26164,
      "ce_loss": 0.3333,
      "cispo_loss": 0.0,
      "elapsed_sec": 253.2,
      "epoch": 1,
      "event": "opt_step",
      "grad_norm": 0.2432,
      "loss": 0.5125,
      "lr_a": 8.3e-07,
      "opt_step": 1,
      "rkl_loss": 0.6405,
      "total_opt_steps": 60,
      "unique_active_tokens": 614542
    },
    {
      "active_tokens_cum": 51612,
      "ce_loss": 0.4032,
      "cispo_loss": 0.0,
      "elapsed_sec": 377.5,
      "epoch": 1,
      "event": "opt_step",
      "grad_norm": 0.252,
      "loss": 0.5891,
      "lr_a": 1.67e-06,
      "opt_step": 2,
      "rkl_loss": 0.7364,
      "total_opt_steps": 60,
      "unique_active_tokens": 614542
    },
    {
      "active_tokens_cum": 74593,
      "ce_loss": 0.3468,
      "cispo_loss": 0.0,
      "elapsed_sec": 471.3,
      "epoch": 1,
      "event": "opt_step",
      "grad_norm": 0.25,
      "loss": 0.6088,
      "lr_a": 2.5e-06,
      "opt_step": 3,
      "rkl_loss": 0.7609,
      "total_opt_steps": 60,
      "unique_active_tokens": 614542
    }
  ],
  "history_tail": [
    {
      "active_tokens_cum": 1148683,
      "ce_loss": 0.3404,
      "cispo_loss": 0.0,
      "elapsed_sec": 3437.7,
      "epoch": 2,
      "event": "opt_step",
      "grad_norm": 0.0903,
      "loss": 0.3094,
      "lr_a": 6.4e-07,
      "opt_step": 55,
      "rkl_loss": 0.3571,
      "total_opt_steps": 60,
      "unique_active_tokens": 614542
    },
    {
      "active_tokens_cum": 1162490,
      "ce_loss": 0.5153,
      "cispo_loss": 0.0,
      "elapsed_sec": 3467.9,
      "epoch": 2,
      "event": "opt_step",
      "grad_norm": 0.1338,
      "loss": 0.3748,
      "lr_a": 5.9e-07,
      "opt_step": 56,
      "rkl_loss": 0.4296,
      "total_opt_steps": 60,
      "unique_active_tokens": 614542
    },
    {
      "active_tokens_cum": 1185737,
      "ce_loss": 0.3504,
      "cispo_loss": 0.0,
      "elapsed_sec": 3508.0,
      "epoch": 2,
      "event": "opt_step",
      "grad_norm": 0.1074,
      "loss": 0.342,
      "lr_a": 5.6e-07,
      "opt_step": 57,
      "rkl_loss": 0.3883,
      "total_opt_steps": 60,
      "unique_active_tokens": 614542
    },
    {
      "active_tokens_cum": 1199301,
      "ce_loss": 0.4556,
      "cispo_loss": 0.0,
      "elapsed_sec": 3548.1,
      "epoch": 2,
      "event": "opt_step",
      "grad_norm": 0.1177,
      "loss": 0.3687,
      "lr_a": 5.3e-07,
      "opt_step": 58,
      "rkl_loss": 0.4209,
      "total_opt_steps": 60,
      "unique_active_tokens": 614542
    },
    {
      "active_tokens_cum": 1221473,
      "ce_loss": 0.3703,
      "cispo_loss": 0.0,
      "elapsed_sec": 3590.6,
      "epoch": 2,
      "event": "opt_step",
      "grad_norm": 0.085,
      "loss": 0.3352,
      "lr_a": 5.2e-07,
      "opt_step": 59,
      "rkl_loss": 0.383,
      "total_opt_steps": 60,
      "unique_active_tokens": 614542
    }
  ],
  "injected_module_count": 258,
  "mode": "dual_gpu_opd",
  "opt_steps": 60,
  "student_alias": "gemma-4-e4b-it-qat-w4a16-ct",
  "total_active_token_updates": 1229084,
  "unique_active_tokens": 614542
}