A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
Handle triton kernel import exception (#25319)
Signed-off-by: Ming Yang <minos.future@gmail.com>
Ming Yang committed
Sep 23, 2025 at 00:56 UTC
ba8d2165b655793ff24cf074d041e0bddfc5a0cc
1 file changed
+8
-3
vllm/model_executor/layers/fused_moe/config.py
+8
-3
@@ -14,11 +14,16 @@ from vllm.model_executor.layers.quantization.utils.quant_utils import (
14
from vllm.utils import cdiv, has_triton_kernels
15
from vllm.utils.flashinfer import has_flashinfer_cutlass_fused_moe
16
17
-if has_triton_kernels():
18
- from triton_kernels.matmul_ogs import PrecisionConfig
19
-
17
logger = init_logger(__name__)
18
19
+if has_triton_kernels():
20
+ try:
21
+ from triton_kernels.matmul_ogs import PrecisionConfig
22
+ except ImportError:
23
+ logger.error(
24
+ "Failed to import Triton kernels. Please make sure your triton "
25
+ "version is compatible.")
26
+
27
28
def _get_config_dtype_str(
29
dtype: torch.dtype,