skip lmslim import when infer non quantized models

vllm/model_executor/layers/quantization/utils/w8a8_utils.py

@@ -13,13 +13,14 @@ from vllm.model_executor.layers.quantization.input_quant_fp8 import QuantFP8
 from vllm.model_executor.layers.quantization.utils.quant_utils import (
     GroupShape)
 from vllm.platforms import current_platform
-
-from vllm.utils import W8a8GetCacheJSON
-from lmslim.layers.gemm.int8_utils import per_token_quant_int8
-
-from vllm.utils import direct_register_custom_op
+from vllm.utils import direct_register_custom_op, W8a8GetCacheJSON
 from vllm.utils.flashinfer import flashinfer_scaled_fp8_mm, has_flashinfer
 
+try:
+    from lmslim.layers.gemm.int8_utils import per_token_quant_int8
+except Exception:
+    print("INFO: Please install lmslim if you want to infer the quantitative model of moe.\n") 
+
 # Input scaling factors are no longer optional in _scaled_mm starting
 # from pytorch 2.5. Allocating a dummy tensor to pass as input_scale
 TORCH_DEVICE_IDENTITY = None