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 12, 18:30 UTC
[CI] Bug: Fix triton import issue (#29202)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Wentao Ye committed
Nov 21, 2025 at 20:14 UTC
1d34eb11e057f6b42af36bdb13852d2701f04245
2 files changed
+2
-4
vllm/v1/worker/gpu/block_table.py
+1
-2
@@ -3,10 +3,9 @@
3
from collections.abc import Iterable
4
5
import torch
6
-import triton
7
-import triton.language as tl
6
7
from vllm.attention.backends.utils import PAD_SLOT_ID
8
+from vllm.triton_utils import tl, triton
9
from vllm.utils.math_utils import cdiv
10
from vllm.v1.utils import CpuGpuBuffer
11
vllm/v1/worker/gpu/input_batch.py
+1
-2
@@ -7,9 +7,8 @@ import numba
7
import numba.types as types
8
import numpy as np
9
import torch
10
-import triton
11
-import triton.language as tl
10
11
+from vllm.triton_utils import tl, triton
12
from vllm.utils import random_uuid
13
from vllm.utils.math_utils import cdiv
14
from vllm.v1.utils import CpuGpuBuffer