Revert "[v1] Add fp32 support to v1 engine through flex attn" (#19404)
This commit is contained in:
@@ -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:
|
||||
|
||||
Reference in New Issue
Block a user