diff --git a/python/sglang/srt/lora/backend/flashinfer_backend.py b/python/sglang/srt/lora/backend/flashinfer_backend.py index 5374a3e0a..91c15be3c 100644 --- a/python/sglang/srt/lora/backend/flashinfer_backend.py +++ b/python/sglang/srt/lora/backend/flashinfer_backend.py @@ -1,10 +1,13 @@ from typing import Tuple import torch -from flashinfer import SegmentGEMMWrapper from sglang.srt.lora.backend import BaseLoraBackend from sglang.srt.lora.lora import LoraBatchInfo +from sglang.srt.utils import is_flashinfer_available + +if is_flashinfer_available(): + from flashinfer import SegmentGEMMWrapper class FlashInferLoraBackend(BaseLoraBackend):