| { | |
| "seed": 42, | |
| "tokenizer_name": "gpt2", | |
| "train_batch_size": 16, | |
| "gradient_acc": 64, | |
| "eval_batch_size": 128, | |
| "max_lr": 0.0001, | |
| "min_lr": 0.00001, | |
| "weight_decay": 0.01, | |
| "out_dir": "out/gsm_sft_2026-08-03_16-35-18", | |
| "save_every": 500, | |
| "lr_warmup_ratio": 0.03, | |
| "num_ce_chunks": 8, | |
| "max_checkpoints": 10, | |
| "resume_from_checkpoint": "out/gsm_sft_2026-08-03_16-35-18/checkpoint_00736000.pt", | |
| "pretrained_model_path": null, | |
| "model_cfg": { | |
| "num_blocks": 12, | |
| "model_dim": 768, | |
| "num_query_heads": 12, | |
| "num_kv_heads": 4, | |
| "attention_dim": 64, | |
| "max_len": 1024, | |
| "rope_theta": 10000.0, | |
| "hidden_dim": 2048 | |
| } | |
| } |