Revert "[v1] Add fp32 support to v1 engine through flex attn" (#19404)

This commit is contained in:
Isotr0py
2025-06-10 01:30:20 -07:00
committed by GitHub
parent 9368cc90b2
commit 5f1ac1e1d1
4 changed files with 7 additions and 38 deletions
-4
View File
@@ -233,10 +233,6 @@ class CudaPlatformBase(Platform):
logger.info_once("Using Triton backend on V1 engine.")
return ("vllm.v1.attention.backends."
"triton_attn.TritonAttentionBackend")
if dtype not in (torch.float16, torch.bfloat16):
logger.info_once(
f"Using FlexAttenion backend for {dtype} on V1 engine.")
return "vllm.v1.attention.backends.flex_attention.FlexAttentionBackend" # noqa: E501
if cls.is_device_capability(100):
# Prefer FlashInfer for V1 on Blackwell GPUs if installed
try: