[Bugfix][Attention] Explicitly report support for kv_cache_dtype bfloat16 (#32795)

Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
This commit is contained in:
Matthew Bonanni
2026-01-22 19:05:18 +00:00
committed by GitHub
parent 744ef30484
commit 955b43a5a5
13 changed files with 31 additions and 11 deletions
+3 -2
View File
@@ -16,6 +16,7 @@ import torch
from vllm import envs
from vllm.logger import init_logger
from vllm.v1.attention.backend import is_quantized_kv_cache
from vllm.v1.attention.backends.registry import AttentionBackendEnum
from .interface import CpuArchEnum, Platform, PlatformEnum
@@ -198,13 +199,13 @@ class CpuPlatform(Platform):
if (
scheduler_config.enable_chunked_prefill
or cache_config.enable_prefix_caching
) and cache_config.cache_dtype != "auto":
) and is_quantized_kv_cache(cache_config.cache_dtype):
raise RuntimeError(
"Chunked-prefill and prefix-cache on the CPU "
"backend is not compatible with FP8 KV cache."
)
if cache_config.cache_dtype != "auto":
if cache_config.cache_dtype.startswith("fp8"):
logger.warning(
"CPU backend doesn't support KV cache quantization fallback to auto."
)