shutdown handoff: exp_t1_fusion_sft_unified_20260428_h100/manifest.json
Browse files
exp_t1_fusion_sft_unified_20260428_h100/manifest.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"metric_name": "eval_loss",
|
| 3 |
+
"mode": "min",
|
| 4 |
+
"records": [
|
| 5 |
+
{
|
| 6 |
+
"path": "/workspace/dnathinker/runs/exp_t1_fusion_sft_unified_20260428_h100/ckpt_step004375_eval_loss2.1677.pt",
|
| 7 |
+
"step": 4375,
|
| 8 |
+
"metric": 2.167741537094116,
|
| 9 |
+
"metric_name": "eval_loss",
|
| 10 |
+
"extra": {
|
| 11 |
+
"eval_loss": 2.167741537094116,
|
| 12 |
+
"eval_runtime": 827.9196,
|
| 13 |
+
"eval_samples_per_second": 8.455,
|
| 14 |
+
"eval_steps_per_second": 1.057,
|
| 15 |
+
"epoch": 1.0
|
| 16 |
+
}
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"path": "/workspace/dnathinker/runs/exp_t1_fusion_sft_unified_20260428_h100/ckpt_step004000_eval_loss2.1688.pt",
|
| 20 |
+
"step": 4000,
|
| 21 |
+
"metric": 2.1687817573547363,
|
| 22 |
+
"metric_name": "eval_loss",
|
| 23 |
+
"extra": {
|
| 24 |
+
"eval_loss": 2.1687817573547363,
|
| 25 |
+
"eval_runtime": 826.5775,
|
| 26 |
+
"eval_samples_per_second": 8.469,
|
| 27 |
+
"eval_steps_per_second": 1.059,
|
| 28 |
+
"epoch": 0.9142857142857143
|
| 29 |
+
}
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"path": "/workspace/dnathinker/runs/exp_t1_fusion_sft_unified_20260428_h100/ckpt_step003500_eval_loss2.1702.pt",
|
| 33 |
+
"step": 3500,
|
| 34 |
+
"metric": 2.1702442169189453,
|
| 35 |
+
"metric_name": "eval_loss",
|
| 36 |
+
"extra": {
|
| 37 |
+
"eval_loss": 2.1702442169189453,
|
| 38 |
+
"eval_runtime": 824.716,
|
| 39 |
+
"eval_samples_per_second": 8.488,
|
| 40 |
+
"eval_steps_per_second": 1.061,
|
| 41 |
+
"epoch": 0.8
|
| 42 |
+
}
|
| 43 |
+
}
|
| 44 |
+
]
|
| 45 |
+
}
|