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 3, 13:00 UTC
Support bf16 for mamba ssm cache (#41680)
Signed-off-by: Qi Zhou <qizzzh@google.com>
Qi Zhou committed
May 16, 2026 at 17:54 UTC
504a26ce2be2415118b73966480b4fc04d9b7bf8
1 file changed
+1
-1
vllm/config/cache.py
+1
-1
@@ -32,7 +32,7 @@ CacheDType = Literal[
32
"fp8_per_token_head",
33
"nvfp4",
34
]
35
-MambaDType = Literal["auto", "float32", "float16"]
35
+MambaDType = Literal["auto", "float32", "float16", "bfloat16"]
36
MambaCacheMode = Literal["all", "align", "none"]
37
PrefixCachingHashAlgo = Literal["sha256", "sha256_cbor", "xxhash", "xxhash_cbor"]
38
KVOffloadingBackend = Literal["native", "lmcache"]