A high-throughput and memory-efficient inference and serving engine for LLMs
[WIP][XPU] upgrade torch-xpu to 2.12 (#42262)
Signed-off-by: Kunshang Ji <kunshang.ji@intel.com> Signed-off-by: Kunshang Ji <jikunshang95@gmail.com>
Kunshang Ji committed
Jun 9, 2026 at 23:51 UTC
b4c6dc64548c46be62abcf5b0bafa7d6ec1a8ae6
5 files changed
+9
-9
docker/Dockerfile.xpu
+1
-1
@@ -203,7 +203,7 @@ RUN --mount=type=cache,target=/root/.cache/uv \
203
uv pip install -r /workspace/vllm/requirements/xpu.txt && \
204
uv pip install --no-build-isolation -r /workspace/vllm/requirements/test/xpu.txt && \
205
uv pip uninstall triton triton-xpu && \
206
- uv pip install triton-xpu==3.7.0 && \
206
+ uv pip install triton-xpu==3.7.1 && \
207
uv pip uninstall oneccl oneccl-devel && \
208
source /opt/intel/oneapi/setvars.sh --force && \
209
source /opt/intel/oneapi/ccl/2021.15/env/vars.sh --force && \
docs/getting_started/installation/gpu.xpu.inc.md
+2
-2
@@ -41,12 +41,12 @@ pip install -v -r requirements/xpu.txt
41
42
```bash
43
pip uninstall -y triton triton-xpu
44
- pip install triton-xpu==3.7.0 --extra-index-url https://download.pytorch.org/whl/xpu
44
+ pip install triton-xpu==3.7.1 --extra-index-url https://download.pytorch.org/whl/xpu
45
```
46
47
!!! note
48
- `triton` (without suffix) is for NVIDIA GPUs only. On XPU, using it instead of `triton-xpu` can cause correctness or runtime issues.
49
- - For torch 2.11 (the version used in `requirements/xpu.txt`), the matching package is `triton-xpu==3.7.0`. If you use a different version of torch, check the corresponding `triton-xpu` version in [docker/Dockerfile.xpu](https://github.com/vllm-project/vllm/blob/main/docker/Dockerfile.xpu).
49
+ - For torch 2.12 (the version used in `requirements/xpu.txt`), the matching package is `triton-xpu==3.7.1`. If you use a different version of torch, check the corresponding `triton-xpu` version in [docker/Dockerfile.xpu](https://github.com/vllm-project/vllm/blob/main/docker/Dockerfile.xpu).
50
51
- Finally, build and install vLLM XPU backend:
52
requirements/test/xpu.in
+1
-1
@@ -31,7 +31,7 @@ schemathesis
31
jiwer
32
bm25s
33
pystemmer
34
-mteb[bm25s]
34
+mteb[bm25s]>=2, <3 # required for mteb test
35
num2words
36
pqdm
37
requirements/test/xpu.txt
+3
-3
@@ -646,7 +646,7 @@ tokenizers==0.22.2
646
# via
647
# -c requirements/common.txt
648
# transformers
649
-torch==2.11.0+xpu
649
+torch==2.12.0+xpu
650
# via
651
# -c requirements/xpu.txt
652
# accelerate
@@ -654,7 +654,7 @@ torch==2.11.0+xpu
654
# sentence-transformers
655
# timm
656
# torchvision
657
-torchvision==0.26.0+xpu
657
+torchvision==0.27.0+xpu
658
# via timm
659
tqdm==4.67.3
660
# via
@@ -672,7 +672,7 @@ transformers==5.5.3
672
# via
673
# -c requirements/common.txt
674
# sentence-transformers
675
-triton-xpu==3.7.0
675
+triton-xpu==3.7.1
676
# via torch
677
typepy==1.3.4
678
# via
requirements/xpu.txt
+2
-2
@@ -12,9 +12,9 @@ jinja2>=3.1.6
12
datasets # for benchmark scripts
13
numba == 0.65.0 # Required for N-gram speculative decoding
14
--extra-index-url=https://download.pytorch.org/whl/xpu
15
-torch==2.11.0
15
+torch==2.12.0
16
torchaudio
17
torchvision
18
19
auto_round_lib>=0.13.0
20
-vllm_xpu_kernels @ https://github.com/vllm-project/vllm-xpu-kernels/releases/download/v0.1.9/vllm_xpu_kernels-0.1.9-cp38-abi3-manylinux_2_28_x86_64.whl
20
+vllm_xpu_kernels @ https://github.com/vllm-project/vllm-xpu-kernels/releases/download/v0.1.9.1/vllm_xpu_kernels-0.1.9.1-cp38-abi3-manylinux_2_28_x86_64.whl