Skip to content

Commit

Permalink
fix typo.
Browse files Browse the repository at this point in the history
  • Loading branch information
adoda committed Jan 30, 2025
1 parent e12d8fd commit bff3fdc
Showing 1 changed file with 1 addition and 1 deletion.
2 changes: 1 addition & 1 deletion chatlearn/synchronizer/megatron_vllm.py
Original file line number Diff line number Diff line change
Expand Up @@ -327,9 +327,9 @@ def regroup_qkv_tp_slices(self, name, param_data, tp_division):
param_data_shape = param_data.shape
# Regroup qkv tensors into different tp slices only for inference model which enables vLLM backend.
to_fix_qkv_ordering_dict = self.sync_map.to_fix_qkv_ordering_dict
# pylint: disable=too-many-nested-blocks
if "attention.query_key_value" in name or \
"self_attention.query_key_value" in name or \
"self_attention.linear_qkv" in name: # pylint: disable=too-many-nested-blocks
src_tp_size = self.src_module_args.args_dict["tensor_model_parallel_size"]
dst_tp_size = self.dst_module_args.args_dict["tensor_model_parallel_size"]
heads = self.src_module_args.args_dict["num_attention_heads"] // src_tp_size
Expand Down

0 comments on commit bff3fdc

Please sign in to comment.