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 13, 13:15 UTC
[CI] Fix Plugin Tests Tests (#28413)
Signed-off-by: Robert Shaw <robertgshaw2@gmail.com>
Robert Shaw committed
Nov 10, 2025 at 17:36 UTC
30700b1cd7de51f191be718215a58f5a8ddcb8aa
1 file changed
+13
-11
vllm/config/vllm.py
+13
-11
@@ -608,17 +608,19 @@ class VllmConfig:
608
)
609
current_platform.check_and_update_config(self)
610
611
- assert (
612
- self.parallel_config.dcp_kv_cache_interleave_size
613
- <= self.cache_config.block_size
614
- and self.cache_config.block_size
615
- % self.parallel_config.dcp_kv_cache_interleave_size
616
- == 0
617
- ), (
618
- f"Block_size({self.cache_config.block_size}) should be "
619
- "greater than or equal to and divisible by dcp_kv_cache_interleave_size "
620
- f"({self.parallel_config.dcp_kv_cache_interleave_size})."
621
- )
611
+ # If DCP, ensure the block size is right.
612
+ if self.parallel_config.decode_context_parallel_size > 1:
613
+ assert (
614
+ self.parallel_config.dcp_kv_cache_interleave_size
615
+ <= self.cache_config.block_size
616
+ and self.cache_config.block_size
617
+ % self.parallel_config.dcp_kv_cache_interleave_size
618
+ == 0
619
+ ), (
620
+ f"Block_size({self.cache_config.block_size}) should be greater "
621
+ "than or equal to and divisible by dcp_kv_cache_interleave_size "
622
+ f"({self.parallel_config.dcp_kv_cache_interleave_size})."
623
+ )
624
625
assert (
626
self.parallel_config.dcp_kv_cache_interleave_size == 1