[CI/Build] Fix CPU platform pre-commit formatting (#52981)
Signed-off-by: mgoin <mgoin64@gmail.com> Co-authored-by: OpenAI Codex <codex@openai.com>
This commit is contained in:
co-authored by
OpenAI Codex
parent
9b5f3454f2
commit
f76d71d7ce
@@ -181,13 +181,16 @@ class CpuPlatform(Platform):
|
|||||||
"otherwise the performance is not optimized."
|
"otherwise the performance is not optimized."
|
||||||
)
|
)
|
||||||
|
|
||||||
# Accelerated GDN (AMX tiles or AVX-512BF16 VDPBF16PS) requires float32 SSM state.
|
# Accelerated GDN (AMX tiles or AVX-512BF16 VDPBF16PS) requires
|
||||||
|
# float32 SSM state.
|
||||||
if (
|
if (
|
||||||
torch.cpu._is_avx512_bf16_supported()
|
torch.cpu._is_avx512_bf16_supported()
|
||||||
and cache_config.mamba_ssm_cache_dtype != "float32"
|
and cache_config.mamba_ssm_cache_dtype != "float32"
|
||||||
):
|
):
|
||||||
cache_config.mamba_ssm_cache_dtype = "float32"
|
cache_config.mamba_ssm_cache_dtype = "float32"
|
||||||
logger.warning("Reset SSM cache type to float32 for accelerated GDN mamba attention.")
|
logger.warning(
|
||||||
|
"Reset SSM cache type to float32 for accelerated GDN mamba attention."
|
||||||
|
)
|
||||||
|
|
||||||
# Lagecy setting
|
# Lagecy setting
|
||||||
env_key = "VLLM_CPU_KVCACHE_SPACE"
|
env_key = "VLLM_CPU_KVCACHE_SPACE"
|
||||||
|
|||||||
Reference in New Issue
Block a user