LM/vLLM: Difference between revisions

From Fundamental Ramen
< LM
Jump to navigation Jump to search
← Older edit
VisualWikitext
 
(83 intermediate revisions by the same user not shown)
Line 1: Line 1:
== Experiment ==
== compose ==


<syntaxhighlight lang="bash">
<syntaxhighlight lang="yaml">
uv venv --python 3.12 --seed --managed-python
services:
source .venv/bin/activate
  vllm-xpu:
    image: vllm/vllm-openai-xpu:v0.30.0
    container_name: vllm-xpu-qwen38-exp


git clone https://github.com/vllm-project/vllm.git
    devices:
cd vllm
      - /dev/dri:/dev/dri
pip install --upgrade pip
pip install -v -r requirements/xpu.txt


pip uninstall -y triton triton-xpu
    volumes:
pip install triton-xpu==3.7.2 --extra-index-url https://download.pytorch.org/whl/xpu
      - ${HOME}/.cache/huggingface:/root/.cache/huggingface


vllm serve unsloth/gemma-4-E4B-it-qat-GGUF --device xpu
    environment:
      - HF_HOME=/root/.cache/huggingface
      - HF_TOKEN=${HF_TOKEN:-}
      - no_proxy=localhost,127.0.0.1
      - ZE_FLAT_DEVICE_HIERARCHY=COMPOSITE
      - ZE_AFFINITY_MASK=0
      - VLLM_WORKER_MULTIPROC_METHOD=spawn
      - PYTORCH_ALLOC_CONF=expandable_segments:True
      - VLLM_XPU_ENABLE_XPU_GRAPH=1
 
    ports:
      - "9000:9000"
 
    entrypoint: ["vllm", "serve"]
    command:
      - "SergiioB/Qwen3.8-27B-GPTQ-Int4-sym-G128-MTP-BF16"
      - "--served-model-name=qwen38"
      - "--port=9000"
      - "--host=0.0.0.0"
      - "--quantization=gptq"
      - "--dtype=float16"
      - "--max-model-len=131072"
      - "--gpu-memory-utilization=0.88"
      - "--kv-cache-dtype=fp8"
      - "--max-num-seqs=1"
      - "--max-num-batched-tokens=8192"
      - "--trust-remote-code"
      - "--enable-auto-tool-choice"
      - "--tool-call-parser=qwen3_xml"
      - "--reasoning-parser=qwen3"
      - "--limit-mm-per-prompt={\"image\":10}"
      - "--speculative-config={\"method\":\"mtp\",\"num_speculative_tokens\":4}"
 
    restart: no
</syntaxhighlight>
 
== Verify environments in container ==
 
{| class="wikitable"
! Command || Results in vllm v0.30.0
|-
| valign="top" | <syntaxhighlight lang="bash">
docker exec vllm-xpu-qwen38-prod bash -c \
  "pip list | awk 'NR<3 || tolower(\$0) ~ /torch|intel|triton|vllm|oneccl|mkl|dpc|level/'"
</syntaxhighlight>
| <pre>
Package                                  Version        Editable project location
---------------------------------------- --------------- -------------------------------------
dpcpp-cpp-rt                            2026.0.0
intel-cmplr-lib-rt                      2026.0.0
intel-cmplr-lib-ur                      2026.0.0
intel-cmplr-lic-rt                      2026.0.0
intel-opencl-rt                          2026.0.0
intel-openmp                            2026.0.0
intel-pti                                0.17.0
intel-sycl-rt                            2026.0.0
mkl                                      2026.0.0
oneccl                                  2022.0.0
oneccl-devel                            2022.0.0
onemkl-license                          2026.0.0
onemkl-sycl-blas                        2026.0.0
onemkl-sycl-dft                          2026.0.0
onemkl-sycl-lapack                      2026.0.0
onemkl-sycl-rng                          2026.0.0
onemkl-sycl-sparse                      2026.0.0
open_clip_torch                          2.32.0
torch                                    2.13.0+xpu
torchaudio                              2.11.0+xpu
torchcodec                              0.16.0+cpu
torchvision                              0.28.0+xpu
triton                                  3.7.2+xpu
triton-xpu                              3.7.2
vllm                                    0.30.0+xpu
vllm_test_utils                          0.1            /workspace/vllm/tests/vllm_test_utils
vllm-xpu-kernels                        0.1.14.1
</pre>
|-
| valign="top" | <syntaxhighlight lang="bash">
docker exec vllm-xpu-qwen38-prod bash -c 'pip show vllm-xpu-kernels'
</syntaxhighlight>
| <pre>
Name: vllm-xpu-kernels
Version: 0.1.14.1
Summary: A high-throughput and memory-efficient inference and serving engine for LLMs
Home-page: https://github.com/vllm-project/vllm-xpu-kernels
Author: vLLM Team
Author-email:
License-Expression: Apache-2.0
Location: /opt/venv/lib/python3.12/site-packages
Requires:
Required-by: vllm
</pre>
|-
| valign="top" | <syntaxhighlight lang="bash">
docker exec vllm-xpu-qwen38-prod bash -c 'lsb_release -a'
</syntaxhighlight>
| <pre>
Distributor ID: Ubuntu
Description:    Ubuntu 24.04.5 LTS
Release:        24.04
Codename:      noble
</pre>
|-
| valign="top" | <syntaxhighlight lang="bash">
docker exec -it vllm-xpu-qwen38-prod bash
</syntaxhighlight>
</syntaxhighlight>
|
|}
== About CCL_SYCL_* variables ==


== Path ==
<pre>
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:40:  118 |CCL_WARN| value of CCL_ATL_TRANSPORT changed to be ofi (default:mpi)
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:40:  118 |CCL_WARN| value of CCL_SYCL_ALLREDUCE_SIMPLE_THRESHOLD changed to be 4294967296 (default:4194304)
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:40:  118 |CCL_WARN| value of CCL_SYCL_REDUCE_SCATTER_SIMPLE_THRESHOLD changed to be 4294967296 (default:8388608)
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:40:  118 |CCL_WARN| value of CCL_SYCL_ALLGATHERV_SIMPLE_THRESHOLD changed to be 4294967296 (default:1048576)
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:40:  118 |CCL_WARN| value of CCL_SYCL_ALLTOALL_TMP_BUF changed to be 1 (default:0)
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:40:  117 |CCL_WARN| value of CCL_ATL_TRANSPORT changed to be ofi (default:mpi)
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:40:  117 |CCL_WARN| value of CCL_SYCL_ALLREDUCE_SIMPLE_THRESHOLD changed to be 4294967296 (default:4194304)
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:40:  117 |CCL_WARN| value of CCL_SYCL_REDUCE_SCATTER_SIMPLE_THRESHOLD changed to be 4294967296 (default:8388608)
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:40:  117 |CCL_WARN| value of CCL_SYCL_ALLGATHERV_SIMPLE_THRESHOLD changed to be 4294967296 (default:1048576)
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:40:  117 |CCL_WARN| value of CCL_SYCL_ALLTOALL_TMP_BUF changed to be 1 (default:0)
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:40:  118 |CCL_WARN| could not get local_idx/count from environment variables, trying to get them from ATL
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:40:  117 |CCL_WARN| could not get local_idx/count from environment variables, trying to get them from ATL
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:41:  223:[1] |CCL_WARN| no membind support for NUMA node 0, skip thread membind
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:41:  224:[0] |CCL_WARN| no membind support for NUMA node 0, skip thread membind
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:41:  118:[1] |CCL_WARN| topology recognition shows PCIe connection between devices. If this is not correct, you can disa
ble topology recognition, with CCL_TOPO_FABRIC_VERTEX_CONNECTION_CHECK=0. This will assume XeLinks across devices
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:41:  117:[0] |CCL_WARN| topology recognition shows PCIe connection between devices. If this is not correct, you can disa
ble topology recognition, with CCL_TOPO_FABRIC_VERTEX_CONNECTION_CHECK=0. This will assume XeLinks across devices
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:41:  118:[1] |CCL_WARN| pidfd is not supported, fallbacks to drmfd exchange mode
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:41:  117:[0] |CCL_WARN| pidfd is not supported, fallbacks to drmfd exchange mode
</pre>


<quickmmd name="vllm-inst">
== CCL Optimization ==
flowchart LR
  A("Ubuntu 24.04")
  B("Install pipx")
  C("Install uv<br>pipx install uv")


  A --> B
TODO
  B --> C
</quickmmd>


== TODO ==
== TODO ==


* https://docs.vllm.ai/en/stable/getting_started/installation/gpu/#requirements
* [https://hub.docker.com/r/vllm/vllm-openai-xpu/tags Official Images for XPU]
* [https://docs.vllm.ai/en/stable/cli/serve/ vllm serve arguments]
* [https://huggingface.co/docs/huggingface_hub/package_reference/environment_variables HF_* Environments]
* [https://github.com/intel/llvm/blob/sycl/sycl/doc/EnvironmentVariables.md ONEAPI_* SYCL_* Environments]
* [https://www.intel.com/content/www/us/en/docs/oneccl/developer-guide-reference/2021-16/environment-variables.html CCL_* Environments]
 
== References ==
 
<references/>

Latest revision as of 19:04, 30 September 2026

compose

services:
  vllm-xpu:
    image: vllm/vllm-openai-xpu:v0.30.0
    container_name: vllm-xpu-qwen38-exp

    devices:
      - /dev/dri:/dev/dri

    volumes:
      - ${HOME}/.cache/huggingface:/root/.cache/huggingface

    environment:
      - HF_HOME=/root/.cache/huggingface
      - HF_TOKEN=${HF_TOKEN:-}
      - no_proxy=localhost,127.0.0.1
      - ZE_FLAT_DEVICE_HIERARCHY=COMPOSITE
      - ZE_AFFINITY_MASK=0
      - VLLM_WORKER_MULTIPROC_METHOD=spawn
      - PYTORCH_ALLOC_CONF=expandable_segments:True
      - VLLM_XPU_ENABLE_XPU_GRAPH=1

    ports:
      - "9000:9000"

    entrypoint: ["vllm", "serve"]
    command:
      - "SergiioB/Qwen3.8-27B-GPTQ-Int4-sym-G128-MTP-BF16"
      - "--served-model-name=qwen38"
      - "--port=9000"
      - "--host=0.0.0.0"
      - "--quantization=gptq"
      - "--dtype=float16"
      - "--max-model-len=131072"
      - "--gpu-memory-utilization=0.88"
      - "--kv-cache-dtype=fp8"
      - "--max-num-seqs=1"
      - "--max-num-batched-tokens=8192"
      - "--trust-remote-code"
      - "--enable-auto-tool-choice"
      - "--tool-call-parser=qwen3_xml"
      - "--reasoning-parser=qwen3"
      - "--limit-mm-per-prompt={\"image\":10}"
      - "--speculative-config={\"method\":\"mtp\",\"num_speculative_tokens\":4}"

    restart: no

Verify environments in container

Command Results in vllm v0.30.0 
docker exec vllm-xpu-qwen38-prod bash -c \
  "pip list | awk 'NR<3 || tolower(\$0) ~ /torch|intel|triton|vllm|oneccl|mkl|dpc|level/'"

Package                                  Version         Editable project location
---------------------------------------- --------------- -------------------------------------
dpcpp-cpp-rt                             2026.0.0
intel-cmplr-lib-rt                       2026.0.0
intel-cmplr-lib-ur                       2026.0.0
intel-cmplr-lic-rt                       2026.0.0
intel-opencl-rt                          2026.0.0
intel-openmp                             2026.0.0
intel-pti                                0.17.0
intel-sycl-rt                            2026.0.0
mkl                                      2026.0.0
oneccl                                   2022.0.0
oneccl-devel                             2022.0.0
onemkl-license                           2026.0.0
onemkl-sycl-blas                         2026.0.0
onemkl-sycl-dft                          2026.0.0
onemkl-sycl-lapack                       2026.0.0
onemkl-sycl-rng                          2026.0.0
onemkl-sycl-sparse                       2026.0.0
open_clip_torch                          2.32.0
torch                                    2.13.0+xpu
torchaudio                               2.11.0+xpu
torchcodec                               0.16.0+cpu
torchvision                              0.28.0+xpu
triton                                   3.7.2+xpu
triton-xpu                               3.7.2
vllm                                     0.30.0+xpu
vllm_test_utils                          0.1             /workspace/vllm/tests/vllm_test_utils
vllm-xpu-kernels                         0.1.14.1

docker exec vllm-xpu-qwen38-prod bash -c 'pip show vllm-xpu-kernels'

Name: vllm-xpu-kernels
Version: 0.1.14.1
Summary: A high-throughput and memory-efficient inference and serving engine for LLMs
Home-page: https://github.com/vllm-project/vllm-xpu-kernels
Author: vLLM Team
Author-email:
License-Expression: Apache-2.0
Location: /opt/venv/lib/python3.12/site-packages
Requires:
Required-by: vllm

docker exec vllm-xpu-qwen38-prod bash -c 'lsb_release -a'

Distributor ID: Ubuntu
Description:    Ubuntu 24.04.5 LTS
Release:        24.04
Codename:       noble

docker exec -it vllm-xpu-qwen38-prod bash

About CCL_SYCL_* variables

vllm-xpu-qwen38-prod  | 2026:09:30-07:47:40:  118 |CCL_WARN| value of CCL_ATL_TRANSPORT changed to be ofi (default:mpi)
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:40:  118 |CCL_WARN| value of CCL_SYCL_ALLREDUCE_SIMPLE_THRESHOLD changed to be 4294967296 (default:4194304)
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:40:  118 |CCL_WARN| value of CCL_SYCL_REDUCE_SCATTER_SIMPLE_THRESHOLD changed to be 4294967296 (default:8388608)
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:40:  118 |CCL_WARN| value of CCL_SYCL_ALLGATHERV_SIMPLE_THRESHOLD changed to be 4294967296 (default:1048576)
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:40:  118 |CCL_WARN| value of CCL_SYCL_ALLTOALL_TMP_BUF changed to be 1 (default:0)
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:40:  117 |CCL_WARN| value of CCL_ATL_TRANSPORT changed to be ofi (default:mpi)
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:40:  117 |CCL_WARN| value of CCL_SYCL_ALLREDUCE_SIMPLE_THRESHOLD changed to be 4294967296 (default:4194304)
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:40:  117 |CCL_WARN| value of CCL_SYCL_REDUCE_SCATTER_SIMPLE_THRESHOLD changed to be 4294967296 (default:8388608)
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:40:  117 |CCL_WARN| value of CCL_SYCL_ALLGATHERV_SIMPLE_THRESHOLD changed to be 4294967296 (default:1048576)
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:40:  117 |CCL_WARN| value of CCL_SYCL_ALLTOALL_TMP_BUF changed to be 1 (default:0)
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:40:  118 |CCL_WARN| could not get local_idx/count from environment variables, trying to get them from ATL
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:40:  117 |CCL_WARN| could not get local_idx/count from environment variables, trying to get them from ATL
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:41:  223:[1] |CCL_WARN| no membind support for NUMA node 0, skip thread membind
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:41:  224:[0] |CCL_WARN| no membind support for NUMA node 0, skip thread membind
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:41:  118:[1] |CCL_WARN| topology recognition shows PCIe connection between devices. If this is not correct, you can disa
ble topology recognition, with CCL_TOPO_FABRIC_VERTEX_CONNECTION_CHECK=0. This will assume XeLinks across devices
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:41:  117:[0] |CCL_WARN| topology recognition shows PCIe connection between devices. If this is not correct, you can disa
ble topology recognition, with CCL_TOPO_FABRIC_VERTEX_CONNECTION_CHECK=0. This will assume XeLinks across devices
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:41:  118:[1] |CCL_WARN| pidfd is not supported, fallbacks to drmfd exchange mode
vllm-xpu-qwen38-prod  | 2026:09:30-07:47:41:  117:[0] |CCL_WARN| pidfd is not supported, fallbacks to drmfd exchange mode

CCL Optimization

TODO

TODO

References

Retrieved from "http://192.168.25.90:53000/index.php?title=LM/vLLM&oldid=3766"