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
[Misc] Fix a config typo in disable_hybrid_kv_cache_manager configuration (#19383)
Signed-off-by: Siyuan Liu <lsiyuan@google.com>
Siyuan Liu committed
Jun 9, 2025 at 16:41 UTC
3a7cd627a86c61a6dbc8d365c2e3c921b94e9971
1 file changed
+3
-3
vllm/config.py
+3
-3
@@ -4497,13 +4497,13 @@ class VllmConfig:
4497
# warning message here and will log it later.
4498
if not (current_platform.is_cuda() or current_platform.is_rocm()):
4499
# Hybrid KV cache manager is not supported on non-GPU platforms.
4500
- self.disable_hybrid_kv_cache_manager = True
4500
+ self.scheduler_config.disable_hybrid_kv_cache_manager = True
4501
if self.kv_transfer_config is not None:
4502
# Hybrid KV cache manager is not compatible with KV transfer.
4503
- self.disable_hybrid_kv_cache_manager = True
4503
+ self.scheduler_config.disable_hybrid_kv_cache_manager = True
4504
if self.kv_events_config is not None:
4505
# Hybrid KV cache manager is not compatible with KV events.
4506
- self.disable_hybrid_kv_cache_manager = True
4506
+ self.scheduler_config.disable_hybrid_kv_cache_manager = True
4507
4508
def update_sizes_for_sequence_parallelism(self,
4509
possible_sizes: list) -> list: