Undo parallel state changes

Signed-off-by: MaximumEntropy <sandeep.subramanian.1@umontreal.ca>

megatron/core/parallel_state.py

@@ -53,7 +53,6 @@ def initialize_model_parallel(
     pipeline_model_parallel_size: int = 1,
     virtual_pipeline_model_parallel_size: Optional[int] = None,
     pipeline_model_parallel_split_rank: Optional[int] = None,
-    untie_embeddings_and_output_weights: bool = False,
 ) -> None:
     """
     Initialize model data parallel groups.
@@ -94,9 +93,6 @@ def initialize_model_parallel(
             pipeline_model_parallel_split_rank is 3, then ranks 0-2
             will be the encoder and ranks 3-7 will be the decoder.
 
-        untie_embeddings_and_output_weights: whether to use separate embedding and output layer.
-                this affects the computation of embedding groups
-
     Let's say we have a total of 16 GPUs denoted by g0 ... g15 and we
     use 2 GPUs to parallelize the model tensor, and 4 GPUs to parallelize
     the model pipeline. The present function will
@@ -204,16 +200,10 @@ def initialize_model_parallel(
         # Setup embedding group (to exchange gradients between
         # first and last stages).
         if len(ranks) > 1:
-            if untie_embeddings_and_output_weights:
-                embedding_ranks = [ranks[0]]
-            else:
             embedding_ranks = [ranks[0], ranks[-1]]
             position_embedding_ranks = [ranks[0]]
             if pipeline_model_parallel_split_rank is not None:
                 if ranks[pipeline_model_parallel_split_rank] not in embedding_ranks:
-                    if untie_embeddings_and_output_weights:
-                        embedding_ranks = [ranks[0], ranks[pipeline_model_parallel_split_rank]]
-                    else:
                     embedding_ranks = [ranks[0],
                                     ranks[pipeline_model_parallel_split_rank],
                                     ranks[-1]]