| { |
| "status": "complete", |
| "deployed_weight_revision": "37adef9b3917e4a3de0bb77f0db71089b7b82202", |
| "expanded_candidate_revision": "4304e2898cd3e864be4a0787ebd7f37a0189229f", |
| "candidate_branch": "v3-expanded-20261113", |
| "trainable_parameters": 83886080, |
| "optimizer_steps": 360, |
| "training_runtime_seconds": 822.6722, |
| "baseline_standard_acc": { |
| "arc_easy": 0.87, |
| "piqa": 0.79, |
| "sciq": 0.94, |
| "winogrande": 0.79 |
| }, |
| "expanded_standard_results": { |
| "arc_easy": { |
| "name": "arc_easy", |
| "alias": "arc_easy", |
| "sample_len": 100, |
| "acc,none": 0.86, |
| "acc_stderr,none": 0.03487350880197771, |
| "acc_norm,none": 0.89, |
| "acc_norm_stderr,none": 0.03144660377352203 |
| }, |
| "piqa": { |
| "name": "piqa", |
| "alias": "piqa", |
| "sample_len": 100, |
| "acc,none": 0.8, |
| "acc_stderr,none": 0.04020151261036845, |
| "acc_norm,none": 0.84, |
| "acc_norm_stderr,none": 0.03684529491774711 |
| }, |
| "sciq": { |
| "name": "sciq", |
| "alias": "sciq", |
| "sample_len": 100, |
| "acc,none": 0.94, |
| "acc_stderr,none": 0.02386832565759419, |
| "acc_norm,none": 0.96, |
| "acc_norm_stderr,none": 0.019694638556693237 |
| }, |
| "winogrande": { |
| "name": "winogrande", |
| "alias": "winogrande", |
| "sample_len": 100, |
| "acc,none": 0.79, |
| "acc_stderr,none": 0.04093601807403326 |
| } |
| }, |
| "standard_sample_count_per_task": 100, |
| "baseline_bf16_task_suite": 43, |
| "expanded_bf16_task_suite": 43, |
| "decision": "Retain existing V3 weights: no aggregate benchmark gain or BF16 task-suite gain. Expanded weights remain available on candidate branch.", |
| "precision_note": "The training script gate compares quantized baseline with BF16 candidate. Final release decision instead uses previously measured BF1643/60 and identical standard benchmark settings.", |
| "output_cap": 4096, |
| "default_output_tokens": 1024, |
| "generation_time_seconds": 110, |
| "zerogpu_allocation_seconds": 120, |
| "quota_reserve_seconds": 180 |
| } |
