Fix: num_workers <= num_shards, add empty_check=False for webdataset

scripts/training/train_flux_lora.py

@@ -53,14 +53,14 @@ def create_webdataset(data_dir, resolution=1024, batch_size=1):
     print(f"  Found {len(tar_files)} shards")
 
     dataset = (
-        wds.WebDataset([str(f) for f in tar_files], shardshuffle=True)
+        wds.WebDataset([str(f) for f in tar_files], shardshuffle=True, empty_check=False)
         .shuffle(1000)
         .decode("pil")
         .map(preprocess)
         .select(lambda x: x is not None)
         .batched(batch_size, collation_fn=collate_batch)
     )
-    return dataset
+    return dataset, len(tar_files)
 
 
 def pack_latents(latents):
@@ -209,9 +209,10 @@ def main():
 
     # Dataset
     print(f"  Loading dataset from {args.data_dir}")
-    train_dataset = create_webdataset(args.data_dir, args.resolution, args.batch_size)
+    train_dataset, num_shards = create_webdataset(args.data_dir, args.resolution, args.batch_size)
+    num_workers = min(2, num_shards)
     train_dataloader = torch.utils.data.DataLoader(
-        train_dataset, batch_size=None, num_workers=4, pin_memory=True
+        train_dataset, batch_size=None, num_workers=num_workers, pin_memory=True
     )
 
     # Training loop