From e2b31243c092e9f4ade5ffe4bf9a5d5ddae06ca7 Mon Sep 17 00:00:00 2001 From: Seiji Eicher <58963096+eicherseiji@users.noreply.github.com> Date: Wed, 4 Mar 2026 22:24:08 -0800 Subject: [PATCH] [Docs] Update `CacheConfig` block_size docstring to remove inaccurate limit when using CUDA (#35632) Signed-off-by: Seiji Eicher --- vllm/config/cache.py | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/vllm/config/cache.py b/vllm/config/cache.py index d3ce9c067..8a94141c9 100644 --- a/vllm/config/cache.py +++ b/vllm/config/cache.py @@ -40,8 +40,7 @@ class CacheConfig: """Configuration for the KV cache.""" block_size: SkipValidation[BlockSize] = None # type: ignore[assignment] - """Size of a contiguous cache block in number of tokens. On CUDA devices, - only block sizes up to 32 are supported. + """Size of a contiguous cache block in number of tokens. This config has no static default. If left unspecified by the user, it will be set in `Platform.check_and_update_config()` based on the current