update from asus 106

backend/training.py

@@ -25,7 +25,7 @@ def models_dir(project_slug: str) -> str:
     return os.path.join(config.project_dir(project_slug), "models")
 
 
-def start(project_id: int, epochs: int = 50, overrides: Optional[dict] = None, batch_ids: Optional[list] = None) -> dict:
+def start(project_id: int, epochs: int = 50, overrides: Optional[dict] = None, batch_ids: Optional[list] = None, class_ids: Optional[list] = None) -> dict:
     project = projects.get(project_id)
     if project is None:
         raise TrainingError("No such project")
@@ -38,7 +38,7 @@ def start(project_id: int, epochs: int = 50, overrides: Optional[dict] = None, b
     settings = hardware.resolve(overrides, epochs)
     job = jobs.create(
         "train",
-        params={"project_id": project_id, "settings": settings, "batch_ids": batch_ids},
+        params={"project_id": project_id, "settings": settings, "batch_ids": batch_ids, "class_ids": class_ids},
         project_id=project_id,
         message=f"{counts['train']} train / {counts['val']} val",
     )
@@ -102,7 +102,8 @@ def _run_train(job) -> None:
     project = projects.get(job.params["project_id"])
     settings = job.params["settings"]
     batch_ids = job.params.get("batch_ids")
-    data_yaml = dataset.write_data_yaml(project, batch_ids=batch_ids)
+    class_ids = job.params.get("class_ids")
+    data_yaml = dataset.write_data_yaml(project, batch_ids=batch_ids, selected_class_ids=class_ids)
 
     # SAM3 and a training run must not hold VRAM at the same time (REQ-065).
     from backend.sam3_engine import release_engine
@@ -133,6 +134,10 @@ def _run_train(job) -> None:
     model.add_callback("on_fit_epoch_end", on_epoch)
     job.progress(0, settings["epochs"])
 
+    import torch
+    if torch.cuda.is_available():
+        torch.backends.cudnn.benchmark = True
+
     keep_run_dir = False
     try:
         model.train(
@@ -142,7 +147,7 @@ def _run_train(job) -> None:
             batch=settings["batch"],
             device=settings["device"],
             workers=settings.get("workers", 8),
-            cache=False,
+            cache="ram",
             project=os.path.join(out_dir, "runs"),
             name="train",
             exist_ok=True,