fine-tune script used by the GPU training job

train_yolo.py

@@ -0,0 +1,96 @@
+"""Fine-tune YOLO11n card detector on the synthetic poker-table benchmark.
+
+Used for both the smoke (few epochs, fraction of train) and the real run.
+After a real run, uploads weights to the Hub model repo.
+"""
+import argparse
+import os
+
+parse = argparse.ArgumentParser()
+parse.add_argument("--data", default="/data/benchmark/data.yaml")
+parse.add_argument("--weights", default="stock-best.pt")
+parse.add_argument("--epochs", type=int, default=80)
+parse.add_argument("--patience", type=int, default=15)
+parse.add_argument("--batch", type=int, default=16)
+parse.add_argument("--imgsz", type=int, default=1280)
+parse.add_argument("--fraction", type=float, default=1.0, help="fraction of train set")
+parse.add_argument("--device", default=0, help="0 for GPU, cpu for CPU smoke")
+parse.add_argument("--project", default="/data/runs")
+parse.add_argument("--name", default="poker_cards_ft")
+parse.add_argument("--upload-model-repo", default=None,
+                   help="e.g. ThreeSixNine/yolo11n-poker-table-cards; skips upload if empty")
+parse.add_argument("--trackio-space-id", default=None,
+                   help="trackio space id; enables per-epoch metric logging")
+parse.add_argument("--trackio-project", default="yolo-poker-cards-ft")
+args = parse.parse_args()
+
+if args.trackio_space_id:
+    import trackio
+    trackio.init(project=args.trackio_project, space_id=args.trackio_space_id)
+
+DOWNLOAD = ("shrimantasatpati/yolov11_playing_cards_detection",
+            "detect/yolo11n_playing_cards/weights/best.pt")
+
+if not os.path.exists(args.weights):  # stock weights, idempotent download
+    from huggingface_hub import hf_hub_download
+    args.weights = hf_hub_download(*DOWNLOAD)
+print(f"weights: {args.weights}", flush=True)
+
+from ultralytics import YOLO
+model = YOLO(args.weights)
+
+if args.trackio_space_id:
+    import trackio
+
+    def _log_epoch(trainer):
+        m = getattr(trainer, "metrics", None) or {}
+        row = {"epoch": trainer.epoch,
+               "loss": float(trainer.loss) if trainer.loss is not None else None}
+        for k, v in (m.items() if isinstance(m, dict) else []):
+            try:
+                row[k] = round(float(v), 4)
+            except (TypeError, ValueError):
+                pass
+        trackio.log(row, step=trainer.epoch)
+
+    model.add_callback("on_fit_epoch_end", _log_epoch)
+
+model.train(data=args.data, epochs=args.epochs, patience=args.patience,
+            batch=args.batch, imgsz=args.imgsz, fraction=args.fraction,
+            device=args.device, project=args.project, name=args.name,
+            exist_ok=True)
+
+metrics = model.val(data=args.data, imgsz=args.imgsz, split="val")
+if args.trackio_space_id:
+    try:
+        d = metrics.results_dict
+        trackio.log({f"final/{k}": round(float(v), 4) for k, v in d.items()},
+                    step=args.epochs)
+        trackio.finish()
+    except Exception as e:
+        print("trackio final log failed:", e, flush=True)
+results_dir = os.path.join(args.project, args.name)
+print("== TRAIN DONE ==", flush=True)
+print("results_dir:", results_dir, flush=True)
+try:
+    d = metrics.results_dict
+    print("val metrics:", {k: round(float(v), 4) for k, v in d.items()}, flush=True)
+except Exception as e:
+    print("metrics read failed:", e, flush=True)
+
+if args.upload_model_repo:
+    from huggingface_hub import HfApi
+    api = HfApi()
+    api.upload_file(
+        path_or_fileobj=os.path.join(results_dir, "weights", "best.pt"),
+        path_in_repo="best.pt",
+        repo_id=args.upload_model_repo,
+        repo_type="model",
+    )
+    api.upload_file(
+        path_or_fileobj=os.path.join(results_dir, "weights", "last.pt"),
+        path_in_repo="last.pt",
+        repo_id=args.upload_model_repo,
+        repo_type="model",
+    )
+    print(f"uploaded weights to {args.upload_model_repo}", flush=True)