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 15, 06:00 UTC
[V1][NixlConnector] Drop `num_blocks` check (#19532)
Signed-off-by: NickLucche <nlucches@redhat.com>
Nicolò Lucchesi committed
Jun 12, 2025 at 14:36 UTC
1129e2b1abc50edd5038cafb185b7c2456e07e65
1 file changed
-2
vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py
-2
@@ -703,8 +703,6 @@ class NixlConnectorWorker:
703
assert self.block_size == remote_block_size, "Remote P worker with " \
704
"different block size is not supported"
705
706
- assert self.num_blocks >= nixl_agent_meta.num_blocks
707
-
706
# Create dst descs and xfer side handles. TP workers have same #blocks.
707
if engine_id in self.dst_num_blocks:
708
assert self.dst_num_blocks[engine_id] == nixl_agent_meta.num_blocks