feat: FSDP training failures now fallback to next strategy

train.py

@@ -176,6 +176,7 @@ def train(config_path):
     )
     
     # Manually wrap model with FSDP before passing to trainer
+    try:
         from torch.distributed.fsdp import FullyShardedDataParallel as FSDP
         from torch.distributed.fsdp.wrap import transformer_auto_wrap_policy
         from functools import partial
@@ -191,7 +192,6 @@ def train(config_path):
             return partial(
                 transformer_auto_wrap_policy,
                 transformer_layer_cls={Qwen3_5MoeDecoderLayer},
-            min_num_params=1,
             )
         
         # Wrap model with FSDP on CPU first
@@ -247,6 +247,20 @@ def train(config_path):
         
         print(f"Training complete! Model saved to {config['train_params']['output_dir']}")
         
+        # Cleanup distributed process group
+        if dist.is_initialized():
+            dist.destroy_process_group()
+        
+        # Return early - success
+        return
+    except Exception as e:
+        errors.append(("QLoRA 4-bit FSDP training", e))
+        print(f"✗ Failed during FSDP training: {e}")
+        # Cleanup distributed process group if initialized
+        if dist.is_initialized():
+            dist.destroy_process_group()
+        # Fall through to next strategy
+
 
 def main():
     parser = argparse.ArgumentParser(description="Train LoRA adapter")