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 14, 05:30 UTC
OffloadingConnector: Fix GPU block tracking bug (#25856)
Signed-off-by: Or Ozeri <oro@il.ibm.com>
Or Ozeri committed
Sep 30, 2025 at 22:53 UTC
cfd302db9b8341e78c0b460d53c9c82f7df69d7d
1 file changed
+3
-2
vllm/distributed/kv_transfer/kv_connector/v1/offloading_connector.py
+3
-2
@@ -278,8 +278,9 @@ class OffloadingConnectorScheduler:
278
req, start_idx=start_block_idx, end_idx=num_blocks)
279
store_output = self.manager.prepare_store(new_block_hashes)
280
if store_output is None:
281
- logger.warning("Cannot store %s blocks", num_new_blocks)
282
- break
281
+ logger.warning("Request %s: cannot store %s blocks", req_id,
282
+ num_new_blocks)
283
+ continue
284
285
self._next_stored_block_idx[req_id] = num_blocks
286