Commit b10ca211 authored by mshoeybi's avatar mshoeybi
Browse files

fixed a bug

parent 8f7c0d11

megatron/model/transformer.py

+2 −1
Original line number Diff line number Diff line
@@ -214,8 +214,9 @@ class ParallelAttention(MegatronModule): 
        # is not provided, make sure there is no potential memory left

        # from previous inference.

        else:

            self.inference_key_memory = None

            self.inference_value_memory = None

            self.inference_current_sequence_len = None





        # =====================

        # Query, Key, and Value