diff --git a/vllm/distributed/kv_transfer/kv_connector/v1/mooncake/store/data.py b/vllm/distributed/kv_transfer/kv_connector/v1/mooncake/store/data.py index 8aa2625852f..2a625c06277 100644 --- a/vllm/distributed/kv_transfer/kv_connector/v1/mooncake/store/data.py +++ b/vllm/distributed/kv_transfer/kv_connector/v1/mooncake/store/data.py @@ -223,7 +223,6 @@ class ReqMeta: skip_save: bool | None = False, block_hashes: list[BlockHash] | None = None, is_last_chunk: bool | None = None, - discard_partial_chunks: bool = True, original_block_size: int | None = None, ) -> "ReqMeta | None": """Create ReqMeta from a RequestTracker.""" @@ -231,16 +230,8 @@ class ReqMeta: block_hashes = [] input_token_len = tracker.token_len - chunk_boundary = ( - cdiv(tracker.num_saved_tokens + 1, block_size) * block_size - if discard_partial_chunks - else 0 - ) - num_tokens_to_save = ( - (input_token_len // block_size * block_size) - if discard_partial_chunks - else input_token_len - ) + chunk_boundary = cdiv(tracker.num_saved_tokens + 1, block_size) * block_size + num_tokens_to_save = input_token_len // block_size * block_size skip_save = skip_save or num_tokens_to_save < chunk_boundary # A ReqMeta must never carry both a save AND a load.