update ddp call

Signed-off-by: Alexandros Koumparoulis <akoumparouli@nvidia.com>
NVIDIA · Jun 6, 2024 · 1a8b26f · 1a8b26f
1 parent bf3667d
commit 1a8b26f
Showing 1 changed file with 0 additions and 2 deletions.
diff --git a/nemo/collections/nlp/models/language_modeling/megatron_gpt_model.py b/nemo/collections/nlp/models/language_modeling/megatron_gpt_model.py
@@ -532,8 +532,6 @@ def setup_mcore_distributed_parallel(self):
                     config,
                     ddp_config,
                     model_chunk,
-                    data_parallel_group=parallel_state.get_data_parallel_group(with_context_parallel=True),
-                    expert_data_parallel_group=parallel_state.get_data_modulo_expert_parallel_group(),
                     # Turn off bucketing for model_chunk 2 onwards, since communication for these
                     # model chunks is overlapped with compute anyway.
                     disable_bucketing=(model_chunk_idx > 0),