A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 2, 19:30 UTC
[KV Connector] MooncakeStore: drop dead discard_partial_chunks parameter (#43627)
Signed-off-by: Zhewen Li <zhewen@inferact.ai> Co-authored-by: Zhewen Li <zhewen@inferact.ai> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Zhewen Li committed
May 26, 2026 at 13:40 UTC
d98cbf472b65b124b140c4e582ea38416e5c5cae
1 file changed
+2
-11
vllm/distributed/kv_transfer/kv_connector/v1/mooncake/store/data.py
+2
-11
@@ -223,7 +223,6 @@ class ReqMeta:
223
skip_save: bool | None = False,
224
block_hashes: list[BlockHash] | None = None,
225
is_last_chunk: bool | None = None,
226
- discard_partial_chunks: bool = True,
226
original_block_size: int | None = None,
227
) -> "ReqMeta | None":
228
"""Create ReqMeta from a RequestTracker."""
@@ -231,16 +230,8 @@ class ReqMeta:
230
block_hashes = []
231
input_token_len = tracker.token_len
232
234
- chunk_boundary = (
235
- cdiv(tracker.num_saved_tokens + 1, block_size) * block_size
236
- if discard_partial_chunks
237
- else 0
238
- )
239
- num_tokens_to_save = (
240
- (input_token_len // block_size * block_size)
241
- if discard_partial_chunks
242
- else input_token_len
243
- )
233
+ chunk_boundary = cdiv(tracker.num_saved_tokens + 1, block_size) * block_size
234
+ num_tokens_to_save = input_token_len // block_size * block_size
235
236
skip_save = skip_save or num_tokens_to_save < chunk_boundary
237
# A ReqMeta must never carry both a save AND a load.