EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

Enable pytest and yaml style accuracy test (#2073)

Browse Source 
### What this PR does / why we need it?

This PR enabled pytest and yaml style accuracy test, users now can
enable accuracy test by running:

```bash
cd ~/vllm-ascend
pytest -sv ./tests/e2e/singlecard/models/test_lm_eval_correctness.py \
          --config ./tests/e2e/singlecard/models/configs/Qwen3-8B-Base.yaml \
          --report_output ./benchmarks/accuracy/Qwen3-8B-Base.md

pytest -sv ./tests/e2e/singlecard/models/test_lm_eval_correctness.py \
          --config-list-file ./tests/e2e/singlecard/models/configs/accuracy.txt
```

Closes: https://github.com/vllm-project/vllm-ascend/issues/1970

### Does this PR introduce _any_ user-facing change?
no

### How was this patch tested?


- vLLM version: v0.10.0
- vLLM main:
2836dd73f1

---------

Signed-off-by: Icey <1790571317@qq.com>
This commit is contained in:
Icey
2025-07-31 21:39:13 +08:00
committed by GitHub
parent 9c9a7cd90b
commit 86bdde1ca8
10 changed files with 336 additions and 446 deletions
Download Patch File Download Diff File Expand all files Collapse all files

180
.github/workflows/accuracy_test.yaml vendored
View File

@@ -29,35 +29,15 @@ on:
types: [ labeled ]
workflow_dispatch:
inputs:
vllm-version:
description: 'vllm version:'
vllm-ascend-version:
description: 'vllm-ascend:'
required: true
type: choice
# Please also update this when bump matched version
# Current supported vLLM versions
options:
- latest
- main
- v0.10.0
- v0.9.1
- v0.7.3
vllm-ascend-version:
description: 'vllm-ascend version:'
required: true
type: choice
options:
- main
- v0.9.1-dev
- v0.7.3-dev
models:
description: 'model:'
required: true
type: choice
options:
- all
- Qwen/Qwen2.5-VL-7B-Instruct
- Qwen/Qwen3-8B-Base
- Qwen/Qwen3-30B-A3B
default: 'all'
default: main
# Bash shells do not use ~/.profile or ~/.bashrc so these shells need to be explicitly
# declared as "shell: bash -el {0}" on steps that need to be properly activated.
@@ -76,58 +56,27 @@ jobs:
# test will be triggered when tag '*-accuracy-test' & 'ready-for-test' or workflow_dispatch job
if: >-
${{
(contains(github.event.pull_request.labels.*.name, 'accuracy-test') ||
contains(github.event.pull_request.labels.*.name, 'vl-accuracy-test') ||
contains(github.event.pull_request.labels.*.name, 'moe-accuracy-test') ||
contains(github.event.pull_request.labels.*.name, 'dense-accuracy-test')) &&
contains(github.event.pull_request.labels.*.name, 'accuracy-test') &&
contains(github.event.pull_request.labels.*.name, 'ready-for-test') ||
github.event_name == 'workflow_dispatch' || github.event_name == 'schedule'
}}
runs-on: >-
${{
(matrix.model_name == 'Qwen/Qwen3-30B-A3B' && 'linux-aarch64-a2-2') ||
'linux-aarch64-a2-1'
}}
runs-on: ${{ matrix.runner }}
strategy:
matrix:
# the accuracy test will run:
# 1. workflow_dispatch with models input
# - all: Qwen/Qwen3-30B-A3B, Qwen/Qwen2.5-VL-7B-Instruct, Qwen/Qwen3-8B-Base
# - specified but not all: Qwen/Qwen3-30B-A3B, Qwen/Qwen2.5-VL-7B-Instruct, Qwen/Qwen3-8B-Base
# 2. PR labeled with "*-accuracy-test"
# - accuracy-test: Qwen/Qwen3-8B-Base, Qwen/Qwen2.5-VL-7B-Instruct, Qwen/Qwen3-30B-A3B
# - dense-accuracy-test: Qwen/Qwen3-8B-Base
# - vl-accuracy-test: Qwen/Qwen2.5-VL-7B-Instruct
# - moe-accuracy-test: Qwen/Qwen3-30B-A3B
model_name: ${{ fromJSON(
(github.event_name == 'schedule' &&
'["Qwen/Qwen3-30B-A3B","Qwen/Qwen2.5-VL-7B-Instruct","Qwen/Qwen3-8B-Base"]') ||
(github.event.inputs.models == 'all' &&
'["Qwen/Qwen3-30B-A3B","Qwen/Qwen2.5-VL-7B-Instruct","Qwen/Qwen3-8B-Base"]') ||
(github.event.inputs.models == 'Qwen/Qwen3-30B-A3B' &&
'["Qwen/Qwen3-30B-A3B"]') ||
(github.event.inputs.models == 'Qwen/Qwen2.5-VL-7B-Instruct' &&
'["Qwen/Qwen2.5-VL-7B-Instruct"]') ||
(github.event.inputs.models == 'Qwen/Qwen3-8B-Base' &&
'["Qwen/Qwen3-8B-Base"]') ||
contains(github.event.pull_request.labels.*.name, 'accuracy-test') &&
'["Qwen/Qwen3-8B-Base","Qwen/Qwen2.5-VL-7B-Instruct", "Qwen/Qwen3-30B-A3B"]' ||
contains(github.event.pull_request.labels.*.name, 'dense-accuracy-test') &&
'["Qwen/Qwen3-8B-Base"]' ||
contains(github.event.pull_request.labels.*.name, 'vl-accuracy-test') &&
'["Qwen/Qwen2.5-VL-7B-Instruct"]' ||
contains(github.event.pull_request.labels.*.name, 'moe-accuracy-test') &&
'["Qwen/Qwen3-30B-A3B"]'
) }}
include:
- model_name: Qwen3-8B-Base
runner: linux-aarch64-a2-1
- model_name: Qwen2.5-VL-7B-Instruct
runner: linux-aarch64-a2-1
- model_name: Qwen3-30B-A3B
runner: linux-aarch64-a2-2
fail-fast: false
name: ${{ matrix.model_name }} accuracy
container:
image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.2.rc1-910b-ubuntu22.04-py3.11
env:
DATASET_SOURCE: ModelScope
VLLM_USE_MODELSCOPE: True
USE_MODELSCOPE_HUB: 1
# 1. If version specified (work_dispatch), do specified branch accuracy test
# 2. If no version (labeled PR), do accuracy test by default ref:
# The branch, tag or SHA to checkout. When checking out the repository that
@@ -139,10 +88,10 @@ jobs:
- name: Checkout repository
uses: actions/checkout@v4
- name: Check npu and CANN info
- name: Set model name as output
id: set_output
run: |
npu-smi info
cat /usr/local/Ascend/ascend-toolkit/latest/"$(uname -i)"-linux/ascend_toolkit_install.info
echo "model_name=${{ matrix.model_name }}" >> $GITHUB_OUTPUT
- name: Config mirrors
run: |
@@ -161,19 +110,19 @@ jobs:
uses: actions/checkout@v4
with:
repository: vllm-project/vllm
ref: v0.10.0
path: ./vllm-empty
# Please also update this when bump matched version
ref: ${{ github.event.inputs.vllm-version || 'v0.10.0' }}
- name: Install vllm-project/vllm from source
working-directory: ./vllm-empty
run: VLLM_TARGET_DEVICE=empty pip install -e .
run: |
VLLM_TARGET_DEVICE=empty pip install -e .
- name: Resolve vllm-ascend version
run: |
VERSION_INPUT="${{ github.event.inputs.vllm-ascend-version }}"
if [[ "$VERSION_INPUT" == "main" ]]; then
if [[ "$VERSION_INPUT" == "latest" ]]; then
TAGS=$(git ls-remote --tags --sort=-v:refname https://github.com/vllm-project/vllm-ascend "v*" | cut -f2 | sed 's|refs/tags/||')
LATEST_TAG=$(echo "$TAGS" | head -n1)
if [[ -z "$LATEST_TAG" ]]; then
@@ -199,8 +148,8 @@ jobs:
PIP_EXTRA_INDEX_URL: https://mirrors.huaweicloud.com/ascend/repos/pypi
run: |
pip install -r requirements-dev.txt
pip install -v -e .
pip install -v -e .
- name: Get vLLM commit hash and URL
working-directory: ./vllm-empty
run: |
@@ -213,15 +162,6 @@ jobs:
VLLM_ASCEND_COMMIT=$(git rev-parse --short=7 HEAD)
echo "VLLM_ASCEND_COMMIT=$VLLM_ASCEND_COMMIT" >> $GITHUB_ENV
- name: Print resolved hashes
run: |
echo "vLLM : ${{ env.VLLM_COMMIT }}"
echo "vLLM-Ascend: ${{ env.VLLM_ASCEND_COMMIT }}"
- name: Install lm-eval, ray, and datasets
run: |
pip install lm-eval==0.4.8
- name: Collect version info
run: |
for dir in /usr/local/Ascend/ascend-toolkit/*; do
@@ -242,37 +182,27 @@ jobs:
pip show torch_npu | grep "Version:" | awk '{print "GHA_TORCH_NPU_VERSION="$2}'
pip show vllm | grep "Version:" | awk '{print "GHA_VLLM_VERSION="$2}' | sed 's/+.*//'
} >> "$GITHUB_ENV"
- name: Print versions
run: |
echo "CANN: ${{ env.GHA_CANN_VERSION }}"
echo "Torch NPU: ${{ env.GHA_TORCH_NPU_VERSION }}"
echo "Torch: ${{ env.GHA_TORCH_VERSION }}"
echo "vLLM: ${{ env.GHA_VLLM_VERSION }}"
echo "vLLM Ascend: ${{ env.GHA_VLLM_ASCEND_VERSION }}"
- name: Run Accuracy Test
- name: Run accuracy test
id: report
working-directory: ./benchmarks
env:
PYTORCH_NPU_ALLOC_CONF: max_split_size_mb:256
VLLM_WORKER_MULTIPROC_METHOD: spawn
VLLM_USE_MODELSCOPE: True
VLLM_VERSION: ${{ env.GHA_VLLM_VERSION }}
VLLM_COMMIT: ${{ env.VLLM_COMMIT }}
VLLM_ASCEND_VERSION: ${{ env.GHA_VLLM_ASCEND_VERSION || github.ref }}
VLLM_ASCEND_COMMIT: ${{ env.VLLM_ASCEND_COMMIT }}
CANN_VERSION: ${{ env.GHA_CANN_VERSION }}
TORCH_VERSION: ${{ env.GHA_TORCH_VERSION }}
TORCH_NPU_VERSION: ${{ env.GHA_TORCH_NPU_VERSION }}
run: |
model_base_name=$(basename ${{ matrix.model_name }})
markdown_name="${model_base_name}"
echo "markdown_name=$markdown_name"
echo "markdown_name=$markdown_name" >> $GITHUB_OUTPUT
mkdir -p ./accuracy
python ./scripts/run_accuracy.py \
--model "${{ matrix.model_name }}" \
--output "./accuracy/${markdown_name}.md" \
--vllm_ascend_version "${{ env.GHA_VLLM_ASCEND_VERSION || github.ref }}" \
--cann_version "${{ env.GHA_CANN_VERSION }}" \
--torch_npu_version "${{ env.GHA_TORCH_NPU_VERSION }}" \
--torch_version "${{ env.GHA_TORCH_VERSION }}" \
--vllm_version "${{ env.GHA_VLLM_VERSION }}" \
--vllm_commit "${{ env.VLLM_COMMIT }}" \
--vllm_ascend_commit "${{ env.VLLM_ASCEND_COMMIT }}" \
mkdir -p ./benchmarks/accuracy
pytest -sv ./tests/e2e/singlecard/models/test_lm_eval_correctness.py \
--config ./tests/e2e/singlecard/models/configs/${{ matrix.model_name }}.yaml \
--report_output ./benchmarks/accuracy/${model_base_name}.md
- name: Generate step summary
if: ${{ always() }}
@@ -284,19 +214,7 @@ jobs:
SAFE_VLLM_ASCEND_VERSION="${GHA_VLLM_ASCEND_VERSION//\//-}"
echo "SAFE_VLLM_ASCEND_VERSION=$SAFE_VLLM_ASCEND_VERSION" >> "$GITHUB_ENV"
- name: Check report first line for failure
id: check_report
run: |
REPORT_PATH="./benchmarks/accuracy/${{ steps.report.outputs.markdown_name }}.md"
echo "Scanning $REPORT_PATH for ❌ …"
if grep -q '❌' "$REPORT_PATH"; then
echo "contains_fail=true" >> $GITHUB_OUTPUT
else
echo "contains_fail=false" >> $GITHUB_OUTPUT
fi
- name: Upload Report
if: ${{ github.event_name == 'workflow_dispatch' && steps.check_report.outputs.contains_fail == 'false' }}
uses: actions/upload-artifact@v4
with:
name: "report-${{ env.SAFE_VLLM_ASCEND_VERSION }}-${{ steps.report.outputs.markdown_name }}"
@@ -305,12 +223,16 @@ jobs:
retention-days: 90
overwrite: true
outputs:
model_name: ${{ steps.set_output.outputs.model_name }}
create_pr:
runs-on: ubuntu-latest
needs: accuracy_tests
if: ${{ github.event_name == 'workflow_dispatch' }}
if: ${{ github.event_name == 'workflow_dispatch' && github.event.inputs.vllm-ascend-version == 'latest' }}
env:
UPSTREAM_REPO: vllm-project/vllm-ascend
steps:
- name: Checkout repository
uses: actions/checkout@v4
@@ -318,7 +240,7 @@ jobs:
repository: vllm-ascend-ci/vllm-ascend
token: ${{ secrets.PAT_TOKEN }}
ref: main
- name: Add upstream remote
run: |
git remote add upstream https://github.com/${{ env.UPSTREAM_REPO }}.git
@@ -350,7 +272,7 @@ jobs:
find ./docs/source/developer_guide/evaluation/accuracy_report -maxdepth 1 -type f -name '*.md' ! -name 'index.md' -delete
find ./docs/source/developer_guide/evaluation/accuracy_report -mindepth 2 -type f -name '*.md' -exec mv -f {} ./docs/source/developer_guide/evaluation/accuracy_report \;
find ./docs/source/developer_guide/evaluation/accuracy_report -mindepth 1 -type d -empty -delete
- name: Update accuracy_report/index.md
run: |
REPORT_DIR="./docs/source/developer_guide/evaluation/accuracy_report"
@@ -390,16 +312,10 @@ jobs:
head: `vllm-ascend-ci:${{ env.BRANCH_NAME }}`,
base: '${{ github.event.inputs.vllm-ascend-version }}',
title: `[Doc] Update accuracy reports for ${{ github.event.inputs.vllm-ascend-version }}`,
body: `The accuracy results running on NPU Altlas A2 have changed, updating reports for:
${{
github.event.inputs.models == 'all'
&& 'All models (Qwen/Qwen3-30B-A3B, Qwen2.5-VL-7B-Instruct, Qwen3-8B-Base)'
|| github.event.inputs.models
}}
body: `The accuracy results running on NPU Altlas A2 have changed, updating reports for: All models (Qwen/Qwen3-30B-A3B, Qwen2.5-VL-7B-Instruct, Qwen3-8B-Base)
- [Workflow run][1]
[1]: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}`
- [Workflow run][1]
[1]: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}`
});
core.info(`Created PR #${pr.data.number}`);