[Fix] Fix --disable-flashinfer (#1389)

This commit is contained in:
Lianmin Zheng
2024-09-11 04:36:21 -07:00
committed by GitHub
parent 224200e3c2
commit 15c75e4146

View File

@@ -140,11 +140,13 @@ class ServerArgs:
"The option '--disable-flashinfer' will be deprecated in the next release. "
"Please use '--attention-backend triton' instead."
)
self.attention_backend = "triton"
if self.disable_flashinfer_sampling:
logger.warning(
"The option '--disable-flashinfer-sampling' will be deprecated in the next release. "
"Please use '--sampling-backend pytorch' instead. "
)
self.sampling_backend = "pytorch"
# Model-specific patches
if "Alibaba-NLP/gte-Qwen2-1.5B-instruct" == self.model_path: