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 7, 05:15 UTC
[Docs] Update `CacheConfig` block_size docstring to remove inaccurate limit when using CUDA (#35632)
Signed-off-by: Seiji Eicher <seiji@anyscale.com>
Seiji Eicher committed
Mar 4, 2026 at 22:24 UTC
e2b31243c092e9f4ade5ffe4bf9a5d5ddae06ca7
1 file changed
+1
-2
vllm/config/cache.py
+1
-2
@@ -40,8 +40,7 @@ class CacheConfig:
40
"""Configuration for the KV cache."""
41
42
block_size: SkipValidation[BlockSize] = None # type: ignore[assignment]
43
- """Size of a contiguous cache block in number of tokens. On CUDA devices,
44
- only block sizes up to 32 are supported.
43
+ """Size of a contiguous cache block in number of tokens.
44
45
This config has no static default. If left unspecified by the user, it will
46
be set in `Platform.check_and_update_config()` based on the current