| { |
| "run_id": "4ccb379a86f2", |
| "mode": "train_from_scratch", |
| "config_path": "/root/MDNAC/config/train.48gb.yaml", |
| "training_config_snapshot_path": "/root/MDNAC/data/checkpoints/protein_from_scratch/training_config.snapshot.json", |
| "checkpoint_path": "/root/MDNAC/data/checkpoints/protein_from_scratch/checkpoint_best.pt", |
| "best_checkpoint_path": "/root/MDNAC/data/checkpoints/protein_from_scratch/checkpoint_best.pt", |
| "final_checkpoint_path": "/root/MDNAC/data/checkpoints/protein_from_scratch/checkpoint_final.pt", |
| "model": { |
| "progen_model_size": "0.8B", |
| "context_length": 512, |
| "stride": 256, |
| "progen_config_overrides": { |
| "emb_dim": 1024, |
| "n_heads": 16, |
| "n_layers": 16, |
| "hidden_dim": 4096, |
| "head_dim": 64, |
| "n_kv_groups": 4, |
| "linear_key_head_dim": 64, |
| "linear_value_head_dim": 64, |
| "linear_num_key_heads": 16, |
| "linear_num_value_heads": 16 |
| } |
| }, |
| "optimizer": { |
| "type": "adamw", |
| "learning_rate": 0.0002, |
| "weight_decay": 0.1 |
| }, |
| "runtime": { |
| "device": "cuda", |
| "distributed": false, |
| "data_parallel": false, |
| "rank": 0, |
| "world_size": 1, |
| "mixed_precision": "bf16" |
| }, |
| "progress": { |
| "epoch": 0, |
| "global_step": 0, |
| "tokens_seen": 0, |
| "current_part_index": 0, |
| "current_part_uri": null, |
| "current_batch_index": 0, |
| "completed_parts": [] |
| }, |
| "metrics": { |
| "latest_train_loss": null, |
| "latest_val_loss": null, |
| "best_loss": null, |
| "best_metric_name": "val_loss", |
| "train_losses": [], |
| "val_losses": [] |
| }, |
| "timestamps": { |
| "created_at": "2026-06-16T06:42:09.041642+00:00", |
| "updated_at": "2026-06-16T06:42:09.041642+00:00" |
| } |
| } |
|
|
