GEMMA DEVELOPER AGENTSAVED EVIDENCE

← Weight change

Trained / exported adapter / adapter-e4b_pure_opd_v6b

e4b_pure_opd_v6b

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?

Scale prediction distillation to a larger repair corpus.

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?

505 trajectories (184 resolved, 321 unresolved), 4 epochs, 505 steps and 11.06 million active loss-token updates. A historical weight-norm cap was used; its former quantization explanation was later withdrawn. Exported low-rank adapter, rank 64, scaling alpha 32.

Did it improve?

Historical measurements · The family ledger reports several epoch/norm variants; it does not unambiguously assign one task score to this exact 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

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 They may contain original predictions and superseded counts; use the measured-outcome summary and corrections above for the current result.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
| 8 | **`e4b_pure_opd_v6b`** | Nov 2 | Scaled OPD + `||ΔW||_F<=0.28` | `E4B` `r=64, a=32` | `opd_e4b_pure_v6b` | `505` (`184` res + `321` unres, `89` repos) | `2.76M` | `4` (`505`) | **`11.06M`** (`ep2`: `5.53M`) | **`8.51h`** | `ep1: 6–8/45`, `ep2_n180: 9/45`, **`ep2_n200: 12/45`**, `ep3: 8/45`, `ep4: 7/45` | **`0.05`** (`57954119`) |
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_v6b",
  "wrx90_path": "/home/keith/gemma-v2/adapters/e4b_pure_opd_v6b",
  "files": {
    "adapter_model.safetensors": {
      "size_bytes": 279129832,
      "sha256": "118c2c9b9fa4c27aebe49b30d63a1224a2f0231ca392dc64d0ee24d993377b81"
    }
  },
  "sub_checkpoints": {
    "epoch_1": {
      "adapter_model.safetensors": {
        "size_bytes": 279129832
      }
    },
    "epoch_2": {
      "adapter_model.safetensors": {
        "size_bytes": 279129832
      }
    },
    "epoch_3": {
      "adapter_model.safetensors": {
        "size_bytes": 279129832
      }
    },
    "epoch_4": {
      "adapter_model.safetensors": {
        "size_bytes": 279129832
      }
    }
  }
}
Training metrics — original archive, not task improvement
{
  "dataset_size": 505,
  "elapsed_seconds": 30620.49,
  "epochs": 4,
  "export": {
    "adapter_dir": "/home/keith/gemma-v2/adapters/e4b_pure_opd_v6b",
    "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": 25694,
      "ce_loss": 0.2838,
      "cispo_loss": 0.0,
      "delta_w_frob": 0.00041,
      "delta_w_pre_proj": 0.00041,
      "delta_w_projected": false,
      "elapsed_sec": 285.4,
      "epoch": 1,
      "event": "opt_step",
      "grad_norm": 0.264,
      "loss": 0.6209,
      "lr_a": 8e-08,
      "opt_step": 1,
      "rkl_loss": 0.7761,
      "total_opt_steps": 508,
      "unique_active_tokens": 2764116
    },
    {
      "active_tokens_cum": 41196,
      "ce_loss": 0.2746,
      "cispo_loss": 0.0,
      "delta_w_frob": 0.00097,
      "delta_w_pre_proj": 0.00097,
      "delta_w_projected": false,
      "elapsed_sec": 344.9,
      "epoch": 1,
      "event": "opt_step",
      "grad_norm": 0.2512,
      "loss": 0.5527,
      "lr_a": 1.6e-07,
      "opt_step": 2,
      "rkl_loss": 0.6909,
      "total_opt_steps": 508,
      "unique_active_tokens": 2764116
    },
    {
      "active_tokens_cum": 64037,
      "ce_loss": 0.3454,
      "cispo_loss": 0.0,
      "delta_w_frob": 0.00175,
      "delta_w_pre_proj": 0.00175,
      "delta_w_projected": false,
      "elapsed_sec": 488.6,
      "epoch": 1,
      "event": "opt_step",
      "grad_norm": 0.2354,
      "loss": 0.5874,
      "lr_a": 2.4e-07,
      "opt_step": 3,
      "rkl_loss": 0.7342,
      "total_opt_steps": 508,
      "unique_active_tokens": 2764116
    }
  ],
  "history_tail": [
    {
      "active_tokens_cum": 10962106,
      "ce_loss": 0.3549,
      "cispo_loss": 0.0,
      "delta_w_frob": 0.28,
      "delta_w_pre_proj": 0.28081,
      "delta_w_projected": true,
      "elapsed_sec": 30429.9,
      "epoch": 4,
      "event": "opt_step",
      "grad_norm": 0.4008,
      "loss": 0.3079,
      "lr_a": 4e-07,
      "opt_step": 501,
      "rkl_loss": 0.3391,
      "total_opt_steps": 508,
      "unique_active_tokens": 2764116
    },
    {
      "active_tokens_cum": 10981323,
      "ce_loss": 0.25,
      "cispo_loss": 0.0,
      "delta_w_frob": 0.28,
      "delta_w_pre_proj": 0.28083,
      "delta_w_projected": true,
      "elapsed_sec": 30466.5,
      "epoch": 4,
      "event": "opt_step",
      "grad_norm": 0.4386,
      "loss": 0.2424,
      "lr_a": 4e-07,
      "opt_step": 502,
      "rkl_loss": 0.2699,
      "total_opt_steps": 508,
      "unique_active_tokens": 2764116
    },
    {
      "active_tokens_cum": 11007916,
      "ce_loss": 0.4995,
      "cispo_loss": 0.0,
      "delta_w_frob": 0.28,
      "delta_w_pre_proj": 0.28085,
      "delta_w_projected": true,
      "elapsed_sec": 30520.6,
      "epoch": 4,
      "event": "opt_step",
      "grad_norm": 0.6963,
      "loss": 0.4054,
      "lr_a": 4e-07,
      "opt_step": 503,
      "rkl_loss": 0.4561,
      "total_opt_steps": 508,
      "unique_active_tokens": 2764116
    },
    {
      "active_tokens_cum": 11028115,
      "ce_loss": 0.4048,
      "cispo_loss": 0.0,
      "delta_w_frob": 0.28,
      "delta_w_pre_proj": 0.28084,
      "delta_w_projected": true,
      "elapsed_sec": 30569.8,
      "epoch": 4,
      "event": "opt_step",
      "grad_norm": 0.4246,
      "loss": 0.2858,
      "lr_a": 4e-07,
      "opt_step": 504,
      "rkl_loss": 0.3146,
      "total_opt_steps": 508,
      "unique_active_tokens": 2764116
    },
    {
      "active_tokens_cum": 11056464,
      "ce_loss": 0.4044,
      "cispo_loss": 0.0,
      "delta_w_frob": 0.28,
      "delta_w_pre_proj": 0.28083,
      "delta_w_projected": true,
      "elapsed_sec": 30620.2,
      "epoch": 4,
      "event": "opt_step",
      "grad_norm": 0.3473,
      "loss": 0.3741,
      "lr_a": 4e-07,
      "opt_step": 505,
      "rkl_loss": 0.4209,
      "total_opt_steps": 508,
      "unique_active_tokens": 2764116
    }
  ],
  "injected_module_count": 258,
  "max_qat_frobenius_norm": 0.28,
  "mode": "dual_gpu_opd",
  "opt_steps": 505,
  "preflight_gate": {
    "dataset_size": 505,
    "eligible_unique_active_tokens": 2764116,
    "epochs": 4,
    "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_resolved_rows": 184,
    "n_unresolved_opd_rows": 321,
    "opd_weight": 1.0,
    "passed": true,
    "planned_active_token_updates": 11056464,
    "resolved_instances": 50,
    "unique_instances": 125,
    "violations": []
  },
  "student_alias": "gemma-4-e4b-it-qat-w4a16-ct",
  "total_active_token_updates": 11056464,
  "trust_region_projections": 455,
  "unique_active_tokens": 2764116
}