| { | |
| "base": "Qwen/Qwen3-8B-Base", | |
| "data": "data/v4", | |
| "out": "runs/assay-8b", | |
| "lora_r": 16, | |
| "lora_alpha": 32, | |
| "lr": 5e-05, | |
| "head_lr": 0.001, | |
| "epochs": 1.0, | |
| "batch_size": 4, | |
| "grad_accum": 2, | |
| "warmup": 0.03, | |
| "weight_decay": 0.0, | |
| "evidence_weight": 0.5, | |
| "score_sigma": 0.5, | |
| "hard_targets": false, | |
| "content_term": false, | |
| "max_state_tokens": 1024, | |
| "limit": null, | |
| "eval_every": 0, | |
| "log_every": 20, | |
| "seed": 0, | |
| "no_gradient_checkpointing": false, | |
| "checkpoint_every": 500, | |
| "resume": true, | |
| "quant": null, | |
| "eval_batch_size": 8 | |
| } |