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
Enable prefix caching with full cuda graphs (#19617)
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
Woosuk Kwon committed
Jun 15, 2025 at 01:05 UTC
055915e6ce0bdd3a9f7222cd23084197faaed408
1 file changed
-1
vllm/config.py
-1
@@ -4495,7 +4495,6 @@ class VllmConfig:
4495
"full_cuda_graph is not supported with "
4496
"cascade attention. Disabling cascade attention.")
4497
self.model_config.disable_cascade_attn = True
4498
- self.cache_config.enable_prefix_caching = False
4498
4499
if (self.kv_events_config is not None
4500
and self.kv_events_config.enable_kv_cache_events