Upload checkpoints/status.json with huggingface_hub

checkpoints/status.json

@@ -1,14 +1,12 @@
 {
+  "last_check": "2026-03-11T01:55:00Z",
   "step": 12373126,
-  "tokens_seen": 10831075761,
-  "tokens_target": 35000000000,
-  "progress_pct": 30.9,
-  "loss": null,
-  "gpu_util": 96,
-  "gpu_temp": 62,
-  "tok_per_sec": null,
-  "status": "training",
-  "box": "New RTX 4090 (Instance 32610565)",
-  "last_check_utc": "2026-03-11 01:05",
-  "note": "Resumed training after ~60h downtime. New instance. Step 12373126. Speed still ramping."
+  "tokens_processed": 10845087297,
+  "target_tokens": 35000000000,
+  "percent": 31.0,
+  "gpu": "RTX 4090",
+  "status": "RUNNING",
+  "instance_id": 32610565,
+  "latest_checkpoint": "pretrain_step12373126.pt",
+  "balance_usd": 76.03
 }