change debugging statement

llmfoundry/data/finetuning/dataloader.py

@@ -179,8 +179,6 @@ def build_finetuning_dataloader(
         padding/waste rates for different `cfg.dataset.packing_ratio` choices,
         given a starting workload YAML.
     """
-    print(f"🚨🚨🚨 build_finetuning_dataloader was called with the following arguments: tokenizer={tokenizer}, device_batch_size={device_batch_size}, dataset={dataset}, num_workers={num_workers}, drop_last={drop_last}, pin_memory={pin_memory}, prefetch_factor={prefetch_factor}, persistent_workers={persistent_workers}, timeout={timeout}")
-
     dataset_cfg = dataset
     is_streaming = (
         dataset_cfg.get('remote') is not None or

llmfoundry/utils/config_utils.py

@@ -429,6 +429,7 @@ def calculate_batch_size_info(
     int, Literal['auto']]]:
 
     world_size = dist.get_world_size()
+    print(f"🚨🚨🚨 world_size: {world_size}")
     if world_size % data_replication_degree != 0:
         raise ValueError(
             f'World size {world_size} is not divisible by data replication degree {data_replication_degree}.',
@@ -457,6 +458,10 @@ def calculate_batch_size_info(
         )
     else:
         raise ValueError(f'Not sure how to parse {device_microbatch_size=}')
+
+    print(f"🚨🚨🚨 device_batch_size: {device_batch_size}")
+    print(f"🚨🚨🚨 device_microbatch_size: {device_microbatch_size}")
+    print(f"🚨🚨🚨 device_grad_accum: {device_grad_accum}")
 
     return device_batch_size, device_microbatch_size, device_grad_accum