[Distributed] Follow upstream TransformerArgs changes (#1161)

Author: kwen2501

dist_run.py

@@ -33,7 +33,7 @@
 )
 from torch.distributed.pipelining import PipelineStage, ScheduleGPipe
 from torchchat.cli.builder import _initialize_tokenizer, TokenizerArgs
-from torchchat.model import ModelArgs, Transformer
+from torchchat.model import ModelArgs, Transformer, TransformerArgs
 from torchchat.utils.build_utils import set_precision
 
 try:
@@ -239,8 +239,11 @@ def main(args):
     distribution, model_dtype = NAME_TO_DISTRIBUTION_AND_DTYPE[model_name]
     logger.info(f"Using HF model weights from {distribution} and dtype {model_dtype}")
 
-    config = ModelArgs.from_name(distribution).transformer_args["text"]
-    logger.info(f"Chat Model Config: {config}")
+    # Model-level config
+    model_config = ModelArgs.from_name(distribution)
+    # Transformer-level config
+    config = TransformerArgs.from_params(model_config.transformer_args["text"])
+    logger.info(f"Transformer Config: {config}")
 
     tokenizer = _build_chat_tokenizer(model_name)
 
@@ -282,6 +285,7 @@ def main(args):
     config.n_stages = pp_degree
 
     with device:
+        # TODO: we should create model instead of Transformer
         model = Transformer(config)
 
     # Distribute model on TP mesh