Upload checkpoints/status.json with huggingface_hub
Browse files- checkpoints/status.json +9 -16
checkpoints/status.json
CHANGED
|
@@ -1,19 +1,12 @@
|
|
| 1 |
{
|
| 2 |
-
"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
"training_active": true,
|
| 4 |
-
"
|
| 5 |
-
"
|
| 6 |
-
"latest_checkpoint": "pretrain_step12373125.pt",
|
| 7 |
-
"step": 12373125,
|
| 8 |
-
"tokens_seen": "10.86B",
|
| 9 |
-
"target_tokens": "35B",
|
| 10 |
-
"progress_pct": 31.0,
|
| 11 |
-
"speed_tok_s": 17400,
|
| 12 |
-
"loss_range": "4-11",
|
| 13 |
-
"next_checkpoint_eta": "2026-03-09T05:32:00Z",
|
| 14 |
-
"latest_inference": "inference/inference_results_step12373125_20260308_204821.json",
|
| 15 |
-
"gpu_util": "93%",
|
| 16 |
-
"gpu_temp": "72C",
|
| 17 |
-
"gpu_mem": "20578/24576 MiB",
|
| 18 |
-
"disk_usage": "84/100G"
|
| 19 |
}
|
|
|
|
| 1 |
{
|
| 2 |
+
"step": 12374115,
|
| 3 |
+
"seen_tok": 10832185419,
|
| 4 |
+
"loss": 10.558960914611816,
|
| 5 |
+
"tok_per_sec": 2639.961053823614,
|
| 6 |
+
"phase": "pretrain",
|
| 7 |
+
"updated": "2026-03-08T22:10:01.380762+00:00",
|
| 8 |
+
"target_tok": 35737600000,
|
| 9 |
"training_active": true,
|
| 10 |
+
"gpu": "RTX 3090",
|
| 11 |
+
"box": "Box 4 (CA)"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
}
|