Upload app.py with huggingface_hub

app.py

@@ -60,8 +60,8 @@ def upload_to_hub(local_path: str, repo_path: str) -> None:
         log(f"Upload failed for {repo_path}: {e}")
 
 
-def run_training() -> None:
-    """Run the full training pipeline."""
+def _run_training_inner() -> None:
+    """Run the full training pipeline (inner function)."""
     global training_status
     training_status = "LOADING DATA"
 
@@ -72,7 +72,9 @@ def run_training() -> None:
     log(f"Device: {device}")
     if device.type == "cuda":
         log(f"GPU: {torch.cuda.get_device_name(0)}")
-        log(f"VRAM: {torch.cuda.get_device_properties(0).total_mem / 1e9:.1f} GB")
+        props = torch.cuda.get_device_properties(0)
+        vram = getattr(props, "total_memory", getattr(props, "total_mem", 0))
+        log(f"VRAM: {vram / 1e9:.1f} GB")
 
     log("=" * 60)
     log("PHASE 1: Loading and preparing datasets")
@@ -185,6 +187,17 @@ def run_training() -> None:
     log("=" * 60)
 
 
+def run_training() -> None:
+    """Wrapper that catches all exceptions from the training pipeline."""
+    global training_status
+    try:
+        _run_training_inner()
+    except Exception as e:
+        training_status = f"CRASHED: {e}"
+        log(f"FATAL ERROR: {e}")
+        log(traceback.format_exc())
+
+
 def get_logs() -> str:
     """Return current training logs."""
     header = f"Status: {training_status}\n{'=' * 60}\n"