bug fix on args.deepspeed_config_dict (#328)

microsoft · Jan 10, 2024 · a4f8079 · a4f8079
1 parent d65921c
commit a4f8079
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/pretrain_gpt.py b/pretrain_gpt.py
@@ -38,7 +38,7 @@ def model_provider(pre_process=True, post_process=True):
     config = core_transformer_config_from_args(args)
     with deepspeed.zero.Init(sequence_data_parallel_group=mpu.get_sequence_data_parallel_group(),
                              remote_device=None if args.remote_device == 'none' else args.remote_device,
-                             config_dict_or_path=args.deepspeed_config,
+                             config_dict_or_path=args.deepspeed_config_dict,
                              enabled=args.zero_stage == 3,
                              mpu=mpu):
         if args.deepspeed and not args.no_pipeline_parallel: