Skip to content
Unit Tests

Unit Tests #1300

Sign in to view logs

Unit Tests

Unit Tests #1300

Workflow file for this run

.github/workflows/unit_tests.yml at 5da5115
name: Unit Tests
run-name: "${{ github.event.inputs.title }}"
defaults:
run:
shell: bash -le {0}
on:
repository_dispatch:
workflow_dispatch:
inputs:
title:
description: 'set a title for this run'
required: false
default: ''
repo:
description: 'GitHub repo {owner}/{repo}'
required: false
default: ''
ref:
description: 'GitHub ref: Branch, Tag or Commit SHA'
required: false
default: ''
pr_number:
description: 'PR Number'
required: false
type: number
test_names:
description: 'Input Test(s) to Run (default all)'
required: false
default: ''
test_regex:
description: 'Regex to filter test files'
required: false
default: ''
artifact_id:
description: 'Run id for artifact to be downloaded'
required: false
default: ''
max-parallel:
description: 'Parallel jobs'
required: false
default: '20'
exclusive-gpu:
description: 'One Test Per GPU'
type: boolean
required: false
default: true
server:
description: 'Wheel Build Server'
type: choice
options:
- '["self-hosted", "zen4"]'
- '["self-hosted", "xeon5"]'
env:
CUDA_DEVICE_ORDER: PCI_BUS_ID
CUDA_VISIBLE_DEVICES: 0
TORCH_CUDA_ARCH_LIST: '8.6 8.9'
PYTORCH_CUDA_ALLOC_CONF: 'expandable_segments:True'
MAX_JOBS: 8
RUNNER: 10.0.13.31
XEON5: 10.0.14.248
LEGACY_TESTS: "models/test_internlm.py,models/test_internlm2_5.py,models/test_xverse.py"
IGNORED_TEST_FILES: "test_tgi.py,test_gptneox.py,models/test_mixtral.py,models/test_phi_3_moe.py,test_bits_new.py"
GPTQMODEL_FORCE_BUILD: 1
repo: ${{ github.event.inputs.repo || github.repository }}
ref: ${{ github.event.inputs.ref || github.ref }}
concurrency:
group: ${{ github.event.inputs.ref || github.ref }}-workflow-unit-tests-${{ github.event.inputs.test_names }}
cancel-in-progress: true
jobs:
check-vm:
runs-on: ubuntu-24.04
outputs:
ip: ${{ steps.get_ip.outputs.ip }}
run_id: ${{ steps.get_ip.outputs.run_id }}
max-parallel: ${{ steps.get_ip.outputs.max-parallel }}
steps:
- name: Print env
run: |
echo "repo: ${{ env.repo }}"
echo "ref: ${{ env.ref }}"
echo "artifact_id: ${{ github.event.inputs.artifact_id }}"
echo "test_names: ${{ github.event.inputs.test_names }}"
echo "exclusive-gpu: ${{ github.event.inputs.exclusive-gpu }}"
echo "selected server: ${{ github.event.inputs.server }}"
- name: Select server
id: get_ip
run: |
echo "ip=$RUNNER" >> "$GITHUB_OUTPUT"
echo "ip: $ip"
if [ -n "${{ github.event.inputs.artifact_id }}" ]; then
run_id="${{ github.event.inputs.artifact_id }}"
else
run_id="${{ github.run_id }}"
fi
echo "run_id=$run_id" >> "$GITHUB_OUTPUT"
echo "artifact_id=$run_id"
max_p=${{ github.event.inputs.max-parallel }}
max_p="{\"size\": ${max_p:-20}}"
echo "max-parallel=$max_p" >> "$GITHUB_OUTPUT"
echo "max-parallel=$max_p"
list-test-files:
runs-on: ubuntu-24.04
outputs:
torch-files: ${{ steps.files.outputs.torch-files }}
transformers-files: ${{ steps.files.outputs.transformers-files }}
m4-files: ${{ steps.files.outputs.m4-files }}
steps:
- name: Checkout Codes
uses: actions/checkout@v4
with:
repository: ${{ env.repo }}
ref: ${{ env.ref }}
- name: Fetch PR by number
if: ${{ github.event.inputs.pr_number != 0 }}
run: |
PR_NUMBER=${{ github.event.inputs.pr_number }}
echo "pr number $PR_NUMBER"
git config --global --add safe.directory $(pwd)
git fetch origin pull/${PR_NUMBER}/head:pr-${PR_NUMBER}
git checkout pr-${PR_NUMBER}
- name: List files
id: files
run: |
script="
import json
import os
import re
LEGACY_TESTS = '${LEGACY_TESTS}'
IGNORED_TEST_FILES = '${IGNORED_TEST_FILES}'
TEST_NAMES='${{ github.event.inputs.test_names }}'
TEST_REGEX='${{ github.event.inputs.test_regex }}'
input_test_files_list = [f.strip().removesuffix('.py') for f in TEST_NAMES.split(',') if f.strip()]
transformers_test_files = [f.strip().removesuffix('.py') for f in f'{LEGACY_TESTS}'.split(',') if f.strip()]
transformers_test_files = [f for f in transformers_test_files if not input_test_files_list or f in input_test_files_list]
all_tests = [f.removesuffix('.py') for f in os.listdir('tests/') if f.startswith('test_') and f.endswith('.py') and f.strip().removesuffix('py') not in f'{IGNORED_TEST_FILES}']
all_tests_models = ['models/'+f.removesuffix('.py') for f in os.listdir('tests/models') if f.startswith('test_') and f.endswith('.py') and f.strip().removesuffix('py') not in f'{IGNORED_TEST_FILES}']
torch_test_files = [f for f in all_tests+all_tests_models if (not input_test_files_list or f in input_test_files_list) and f not in transformers_test_files and 'mlx' not in f]
torch_test_files = [test for test in torch_test_files if re.match(f'{TEST_REGEX}', test)]
transformers_test_files = [test for test in transformers_test_files if re.match(f'{TEST_REGEX}', test)]
m4_test_files = [f for f in all_tests if ('mlx' in f or 'apple' in f) and (f.strip().removesuffix('.py') in input_test_files_list if input_test_files_list else True)]
print(f'{json.dumps(torch_test_files)}|{json.dumps(transformers_test_files)}|{json.dumps(m4_test_files)}')
"
test_files=$(python3 -c "$script")
IFS='|' read -r torch_test_files transformers_test_files mlx_files <<< "$test_files"
echo "torch-files=$torch_test_files" >> "$GITHUB_OUTPUT"
echo "transformers-files=$transformers_test_files" >> "$GITHUB_OUTPUT"
echo "m4-files=$mlx_files" >> "$GITHUB_OUTPUT"
echo "Test files: $test_files"
echo "Torch Test files: $torch_test_files"
echo "Transformers Test files: $transformers_test_files"
echo "MLX Test files: $mlx_files"
echo "Ignored Test files: $IGNORED_TEST_FILES"
- name: Print conditions and parameters
run: |
echo "Parameters:"
echo "needs.list-test-files.outputs.torch-files: ${{ needs.list-test-files.outputs.torch-files }}"
echo "needs.list-test-files.outputs.transformers-files: ${{ needs.list-test-files.outputs.transformers-files }}"
echo "needs.list-test-files.outputs.m4-files: ${{ needs.list-test-files.outputs.m4-files }}"
echo "github.event.inputs.artifact_id: ${{ github.event.inputs.artifact_id }}"
echo "needs.build.result: ${{ needs.build.result }}"
echo "github.event.inputs.test_names: ${{ github.event.inputs.test_names }}"
echo ""
echo "Conditions:"
echo "will build run: ${{ needs.list-test-files.outputs.torch-files != '[]' && needs.list-test-files.outputs.transformers-files != '[]' && !(needs.list-test-files.outputs.m4-files == '[]' && needs.list-test-files.outputs.m4-files == '[]') }}"
echo "will legacy run: ${{ (needs.build.result == 'success' || github.event.inputs.artifact_id != '') && needs.list-test-files.outputs.transformers-files != '[]' }}"
echo "will torch run: ${{ (needs.build.result == 'success' || github.event.inputs.artifact_id != '') && needs.list-test-files.outputs.torch-files != '[]' }}"
echo "will m4 run: ${{ (github.event.inputs.test_names == '' || contains(github.event.inputs.test_names, 'apple') || contains(github.event.inputs.test_names, 'mlx') ) && (needs.list-test-files.outputs.m4-files != '' || needs.list-test-files.outputs.m4-files != '[]') }}"
build:
runs-on: ${{ fromJSON(github.event.inputs.server ) }}
needs:
- check-vm
- list-test-files
if: needs.list-test-files.outputs.torch-files != '[]' || needs.list-test-files.outputs.transformers-files != '[]'
container:
image: ${{ needs.check-vm.outputs.ip }}:5000/modelcloud/gptqmodel:github-ci-v7
options: --device /dev/dri --ipc=host --runtime=nvidia --gpus all
volumes:
- /dev/dri/by-path:/dev/dri/by-path
- /home/ci/models:/monster/data/model
- /home/ci/models/huggingface:/github/home/.cache/huggingface
steps:
- name: Checkout Codes
uses: actions/checkout@v4
with:
repository: ${{ env.repo }}
ref: ${{ env.ref }}
- name: Fetch PR by number
if: ${{ github.event.inputs.pr_number != 0 }}
run: |
PR_NUMBER=${{ github.event.inputs.pr_number }}
echo "pr number $PR_NUMBER"
git config --global --add safe.directory $(pwd)
git fetch origin pull/${PR_NUMBER}/head:pr-${PR_NUMBER}
git checkout pr-${PR_NUMBER}
- name: Print Env
run: |
echo "== pyenv =="
pyenv versions
echo "== python =="
python --version
echo "== nvcc =="
nvcc --version
echo "== torch =="
pip show torch
echo "##### pip list #####"
pip list
- name: Compress dir
run: |
mkdir dist || true
rm -rf dist/* || true
tar -zcf ../gptqmodel_source.tar.gz ./
mv ../gptqmodel_source.tar.gz dist/
sha256=$(sha256sum dist/gptqmodel_source.tar.gz)
echo "hash=$sha256"
echo "SOURCE_HASH=$sha256" >> $GITHUB_ENV
- name: Upload source to local
continue-on-error: true
run: curl -s -F "runid=${{ github.run_id }}" -F "repo=${{ env.repo }}" -F "ref=${{ env.ref }}" -F "sha256=${{ env.SOURCE_HASH }}" -F "file=@dist/gptqmodel_source.tar.gz" http://$RUNNER/gpu/whl/upload
- name: Upload source to github artifact
uses: actions/upload-artifact@v4
with:
name: source
path: dist/gptqmodel_source.tar.gz
- name: Compile
if: github.event.inputs.artifact_id == '' && !cancelled()
timeout-minutes: 35
run: python setup.py bdist_wheel
- name: Test install
if: github.event.inputs.artifact_id == '' && !cancelled()
run: |
ls -ahl dist
whl=$(ls -t dist/*.whl | head -n 1 | xargs basename)
sha256=$(sha256sum dist/$whl)
echo "hash=$sha256"
echo "WHL_HASH=$sha256" >> $GITHUB_ENV
echo "WHL_NAME=$whl" >> $GITHUB_ENV
twine check dist/$whl
uv pip install dist/$whl
- name: Upload wheel to local
if: github.event.inputs.artifact_id == '' && !cancelled()
continue-on-error: true
run: curl -s -F "runid=${{ github.run_id }}" -F "repo=${{ env.repo }}" -F "ref=${{ env.ref }}" -F "sha256=${{ env.WHL_HASH }}" -F "file=@dist/${{ env.WHL_NAME }}" http://$RUNNER/gpu/whl/upload
- name: Upload wheel to github artifact
if: github.event.inputs.artifact_id == '' && !cancelled()
uses: actions/upload-artifact@v4
with:
name: whl
path: dist/${{ env.WHL_NAME }}
- name: Clean cache
if: always()
run: pip cache purge && uv cache clean && rm -rf ./* ./.*
legacy:
needs:
- build
- list-test-files
- check-vm
runs-on: [ self-hosted, xeon5 ]
if: always() && !cancelled() && (needs.build.result == 'success' || github.event.inputs.artifact_id != '') && needs.list-test-files.outputs.transformers-files != '[]'
container:
image: ${{ needs.check-vm.outputs.ip }}:5000/modelcloud/gptqmodel:github-ci-v7
volumes:
- /home/ci/models:/monster/data/model
- /home/ci/models/huggingface:/github/home/.cache/huggingface
strategy:
fail-fast: false
max-parallel: ${{ fromJson(needs.check-vm.outputs.max-parallel).size || 20 }}
matrix:
test_script: ${{ fromJSON(needs.list-test-files.outputs.transformers-files) }}
steps:
- name: Checkout Codes
uses: actions/checkout@v4
with:
repository: ${{ github.event.inputs.repo }}
ref: ${{ github.event.inputs.ref }}
- name: Fetch PR by number
if: ${{ github.event.inputs.pr_number != 0 }}
run: |
PR_NUMBER=${{ github.event.inputs.pr_number }}
echo "pr number $PR_NUMBER"
git config --global --add safe.directory $(pwd)
git fetch origin pull/${PR_NUMBER}/head:pr-${PR_NUMBER}
git checkout pr-${PR_NUMBER}
- name: Print Env
run: |
echo "== pyenv =="
pyenv versions
echo "== python =="
python --version
echo "== nvcc =="
nvcc --version
echo "== torch =="
pip show torch
echo "== pip list =="
pip list
- name: Download source from local
continue-on-error: true
run: |
curl -s -O http://$RUNNER/whl/${{ env.repo }}/${{ github.run_id }}/gptqmodel_source.tar.gz
ls -ahl .
sha256=$(sha256sum $file_name)
echo "sha256=$sha256"
echo "SOURCE_DOWNLOADED=1" >> $GITHUB_ENV
- name: Download source from github
if: env.SOURCE_DOWNLOADED == '' && !cancelled()
uses: actions/download-artifact@v4
with:
name: source
path: dist
run-id: ${{ github.run_id }}
- name: Uncompress source
continue-on-error: true
run: |
find . -mindepth 1 ! -name "gptqmodel_source.tar.gz" -exec rm -rf {} +
ls -ahl .
tar -zxf gptqmodel_source.tar.gz
- name: Download wheel from local
continue-on-error: true
run: |
file_name=$(curl -s -F "runid=${{ needs.check-vm.outputs.run_id }}" -F "repo=${{ env.repo }}" -F "ref=${{ env.ref }}" -F "fuzz=1" "http://$RUNNER/gpu/whl/download")
echo "file_name=$file_name"
if echo "$file_name" | grep -q "gptqmodel"; then
mkdir dist || true
cd dist
curl -s -O http://$RUNNER/whl/${{ env.repo }}/${{ needs.check-vm.outputs.run_id }}/$file_name
ls -ahl .
sha256=$(sha256sum $file_name)
echo "sha256=$sha256"
echo "WHL_DOWNLOADED=1" >> $GITHUB_ENV
fi
- name: Download artifact from github
if: env.WHL_DOWNLOADED == '' && !cancelled()
uses: actions/download-artifact@v4
with:
name: whl
path: dist
run-id: ${{ needs.check-vm.outputs.run_id }}
- name: Install wheel
run: |
uv pip install colorlog logbar git+https://github.com/ModelCloud/Tokenicer -U
echo "===== install optimum bitblas parameterized uvicorn ====="
uv pip install optimum bitblas==0.0.1.dev13 parameterized uvicorn -i http://$RUNNER/simple/ --trusted-host $RUNNER --extra-index-url https://pypi.org/simple
echo "===== install dist/whl ====="
uv pip install dist/*.whl -i http://$RUNNER/simple/ --trusted-host $RUNNER --extra-index-url https://pypi.org/simple
echo "===== init test env ====="
echo "===== install transformers==4.38.2 typing-extensions numpy==1.26.4 peft==0.13.2 ====="
uv pip install transformers==4.38.2 typing-extensions numpy==1.26.4 peft==0.13.2 -U -i http://$RUNNER/simple/ --trusted-host $RUNNER --extra-index-url https://pypi.org/simple
if [ "${{ matrix.test_script }}" == "models/test_xverse" ]; then
echo "===== install tokenizers==0.15.2 ====="
uv pip install tokenizers==0.15.2 -i http://$RUNNER/simple/ --trusted-host $RUNNER --extra-index-url https://pypi.org/simple
fi
if [ "${{ matrix.test_script }}" == "test_quant_formats" ] || [ "${{ matrix.test_script }}" == "test_perplexity" ]; then
echo "===== install auto_round ====="
uv pip install auto_round -i http://$RUNNER/simple/ --trusted-host $RUNNER --extra-index-url https://pypi.org/simple
fi
- name: Find suitable GPU
if: ${{ !contains(matrix.test_script, 'ipex') && !cancelled() }}
run: |
timestamp=$(date +%s%3N)
gpu_id=-1
while [ "$gpu_id" -lt 0 ]; do
gpu_id=$(curl -s "http://$RUNNER/gpu/get?id=${{ github.run_id }}&timestamp=$timestamp&test=${{ matrix.test_script }}&runner=${RUNNER_NAME}&exclusive=${{ github.event.inputs.exclusive-gpu }}")
if [ "$gpu_id" -lt 0 ]; then
echo "http://$RUNNER/gpu/get?id=${{ github.run_id }}&timestamp=$timestamp&test=${{ matrix.test_script }}&runner=${RUNNER_NAME}&exclusive=${{ github.event.inputs.exclusive-gpu }} returned $gpu_id"
echo "No available GPU, waiting 5 seconds..."
curl http://$XEON5/gpu/status2
sleep 5
else
echo "Allocated GPU ID: $gpu_id"
fi
done
if [[ ! "$gpu_id" =~ ^[0-9]+$ ]]; then
echo "gpu_id: $gpu_id is not a number"
fi
echo "CUDA_VISIBLE_DEVICES=$gpu_id" >> $GITHUB_ENV
echo "STEP_TIMESTAMP=$timestamp" >> $GITHUB_ENV
echo "CUDA_VISIBLE_DEVICES set to $gpu_id, timestamp=$timestamp"
curl http://$XEON5/gpu/status2
- name: Run tests
if: ${{ (!github.event.inputs.test_names || contains(github.event.inputs.test_names, matrix.test_script)) && !cancelled() }}
run: |
start_time=$(date +%s)
pytest --durations=0 tests/${{ matrix.test_script }}.py || { echo "ERROR=1" >> $GITHUB_ENV; exit 1; }
execution_time=$(( $(date +%s) - start_time ))
echo "$((execution_time / 60))m $((execution_time % 60))s"
curl "http://$RUNNER/gpu/log_test_vram?id=${{ github.run_id }}&gpu=${{ env.CUDA_VISIBLE_DEVICES }}&range=$execution_time&unit=second&name=${{ matrix.test_script }}"
- name: Release GPU
if: always() && !contains(matrix.test_script, 'ipex')
run: curl -X GET "http://$RUNNER/gpu/release?id=${{ github.run_id }}&gpu=${{ env.CUDA_VISIBLE_DEVICES }}&timestamp=${{ env.STEP_TIMESTAMP }}&test=${{ matrix.test_script }}&runner=${RUNNER_NAME}"
- name: Clean cache
if: always()
run: pip cache purge && uv cache clean && rm -rf ./* ./.*
torch:
needs:
- build
- list-test-files
- check-vm
runs-on: [ self-hosted, xeon5 ]
if: always() && !cancelled() && (needs.build.result == 'success' || github.event.inputs.artifact_id != '') && needs.list-test-files.outputs.torch-files != '[]'
container:
image: ${{ needs.check-vm.outputs.ip }}:5000/modelcloud/gptqmodel:github-ci-v7
options: --device /dev/dri --ipc=host --runtime=nvidia --gpus all
volumes:
- /dev/dri/by-path:/dev/dri/by-path
- /home/ci/models:/monster/data/model
- /home/ci/models/huggingface:/github/home/.cache/huggingface
strategy:
fail-fast: false
max-parallel: ${{ fromJson(needs.check-vm.outputs.max-parallel).size || 20 }}
matrix:
test_script: ${{ fromJSON(needs.list-test-files.outputs.torch-files) }}
steps:
- name: Checkout Codes
uses: actions/checkout@v4
with:
repository: ${{ env.repo }}
ref: ${{ env.ref }}
- name: Fetch PR by number
if: ${{ github.event.inputs.pr_number != 0 }}
run: |
PR_NUMBER=${{ github.event.inputs.pr_number }}
echo "pr number $PR_NUMBER"
git config --global --add safe.directory $(pwd)
git fetch origin pull/${PR_NUMBER}/head:pr-${PR_NUMBER}
git checkout pr-${PR_NUMBER}
- name: Print Env
run: |
echo "== pyenv =="
pyenv versions
echo "== python =="
python --version
echo "== nvcc =="
nvcc --version
echo "== torch =="
pip show torch
echo "== pip list =="
pip list
- name: Download source from local
continue-on-error: true
run: |
curl -s -O http://$RUNNER/whl/${{ env.repo }}/${{ github.run_id }}/gptqmodel_source.tar.gz
ls -ahl .
sha256=$(sha256sum $file_name)
echo "sha256=$sha256"
echo "SOURCE_DOWNLOADED=1" >> $GITHUB_ENV
- name: Download source from github
if: env.SOURCE_DOWNLOADED == '' && !cancelled()
uses: actions/download-artifact@v4
with:
name: source
path: dist
run-id: ${{ github.run_id }}
- name: Uncompress source
continue-on-error: true
run: |
find . -mindepth 1 ! -name "gptqmodel_source.tar.gz" -exec rm -rf {} +
ls -ahl .
tar -zxf gptqmodel_source.tar.gz
- name: Download wheel from local
continue-on-error: true
run: |
file_name=$(curl -s -F "runid=${{ needs.check-vm.outputs.run_id }}" -F "repo=${{ env.repo }}" -F "ref=${{ env.ref }}" -F "fuzz=1" "http://$RUNNER/gpu/whl/download")
echo "file_name=$file_name"
if echo "$file_name" | grep -q "gptqmodel"; then
mkdir dist || true
cd dist
curl -s -O http://$RUNNER/whl/${{ env.repo }}/${{ needs.check-vm.outputs.run_id }}/$file_name
ls -ahl .
sha256=$(sha256sum $file_name)
echo "sha256=$sha256"
echo "WHL_DOWNLOADED=1" >> $GITHUB_ENV
fi
- name: Download artifact from github
if: env.WHL_DOWNLOADED == '' && !cancelled()
uses: actions/download-artifact@v4
with:
name: whl
path: dist
run-id: ${{ needs.check-vm.outputs.run_id }}
- name: Install wheel
run: |
uv pip install -U transformers colorlog logbar
if [ "${{ matrix.test_script }}" == "test_perplexity" ] || \
[ "${{ matrix.test_script }}" == "test_q4_bitblas" ] || \
[ "${{ matrix.test_script }}" == "test_save_loaded_quantized_model" ]; then
echo "===== install bitblas==0.0.1.dev13 ====="
uv pip install bitblas==0.0.1.dev13 -i http://$RUNNER/simple/ --trusted-host $RUNNER --extra-index-url https://pypi.org/simple
fi
if [[ "${{ matrix.test_script }}" == *auto_round* ]]; then
uv pip install auto_round
fi
if [ "${{ matrix.test_script }}" == "models/test_cohere2" ] || [ "${{ matrix.test_script }}" == "models/test_gemma" ]; then
echo "===== install transformers from git ====="
uv pip install -U transformers -i http://$RUNNER/simple/ --trusted-host $RUNNER --extra-index-url https://pypi.org/simple
fi
if [[ "${{ matrix.test_script }}" == *xpu* ]]; then
echo "===== switching to xpu env ====="
source /etc/profile.d/pyenv.sh && pyenv activate xpu
uv pip install colorlog
fi
if [[ "${{ matrix.test_script }}" == *ipex* ]] && [[ "${{ matrix.test_script }}" != *xpu* ]]; then
uv pip uninstall torchvision torch flash_attn # fix ipex can't be used with torch+cu126
uv pip install torchvision torch
uv pip install -U intel_extension_for_pytorch -i http://$RUNNER/simple/ --trusted-host $RUNNER --extra-index-url https://pypi.org/simple
fi
if [[ "${{ matrix.test_script }}" == *"mlx"* ]]; then
uv pip install mlx_lm --no-build-isolation -i http://$RUNNER/simple/ --trusted-host $RUNNER --extra-index-url https://pypi.org/simple
fi
if [[ "${{ matrix.test_script }}" == "test_modelscope" ]]; then
echo "===== installing modelscope ====="
uv pip install modelscope --no-build-isolation -i http://$RUNNER/simple/ --trusted-host $RUNNER --extra-index-url https://pypi.org/simple
fi
uv pip install git+https://github.com/ModelCloud/Tokenicer -U
# ipex doesn't need to compile kernels. xpu can't install cuda package
if [[ "${{ matrix.test_script }}" != *ipex* && "${{ matrix.test_script }}" != *xpu* ]]; then
echo "===== install dist/whl ====="
uv pip install dist/*.whl -i http://$RUNNER/simple/ --trusted-host $RUNNER --extra-index-url https://pypi.org/simple
else
echo "===== install with local files for xpu env ====="
export CUDA_VISIBLE_DEVICES=""
unset TORCH_CUDA_ARCH_LIST
uv pip install . --no-build-isolation
fi
if [ "${{ matrix.test_script }}" == "test_transformers" ]; then
echo "===== install optimum from git ====="
uv pip install -U git+https://github.com/huggingface/optimum.git -i http://$RUNNER/simple/ --trusted-host $RUNNER
fi
if [[ "${{ matrix.test_script }}" == "test_sglang" ]]; then
uv pip install transformers==4.48.3
uv pip install numpy==1.26.3
fi
- name: Find suitable GPU
if: ${{ !contains(matrix.test_script, 'ipex') && !contains(matrix.test_script, 'xpu') && !cancelled() }}
run: |
timestamp=$(date +%s%3N)
gpu_id=-1
while [ "$gpu_id" -lt 0 ]; do
gpu_id=$(curl -s "http://$RUNNER/gpu/get?id=${{ github.run_id }}&timestamp=$timestamp&test=${{ matrix.test_script }}&runner=${RUNNER_NAME}&exclusive=${{ github.event.inputs.exclusive-gpu }}")
if [ "$gpu_id" -lt 0 ]; then
echo "http://$RUNNER/gpu/get?id=${{ github.run_id }}&timestamp=$timestamp&test=${{ matrix.test_script }}&runner=${RUNNER_NAME}&exclusive=${{ github.event.inputs.exclusive-gpu }} returned $gpu_id"
echo "No available GPU, waiting 5 seconds..."
curl http://$XEON5/gpu/status2
sleep 5
else
echo "Allocated GPU ID: $gpu_id"
fi
done
if [[ ! "$gpu_id" =~ ^[0-9]+$ ]]; then
echo "gpu_id: $gpu_id is not a number"
fi
echo "CUDA_VISIBLE_DEVICES=$gpu_id" >> $GITHUB_ENV
echo "STEP_TIMESTAMP=$timestamp" >> $GITHUB_ENV
echo "CUDA_VISIBLE_DEVICES set to $gpu_id, timestamp=$timestamp"
curl http://$XEON5/gpu/status2
- name: Run tests
if: ${{ (!github.event.inputs.test_names || contains(github.event.inputs.test_names, matrix.test_script)) && !cancelled() }}
run: |
if [[ "${{ matrix.test_script }}" == *ipex* ]]; then
export CUDA_VISIBLE_DEVICES=""
fi
if [[ "${{ matrix.test_script }}" == *xpu* ]]; then
export CUDA_VISIBLE_DEVICES=""
source /etc/profile.d/pyenv.sh && pyenv activate xpu
pip uninstall vllm -y
pip list
fi
start_time=$(date +%s)
pytest --durations=0 tests/${{ matrix.test_script }}.py || { echo "ERROR=1" >> $GITHUB_ENV; exit 1; }
execution_time=$(( $(date +%s) - start_time ))
echo "$((execution_time / 60))m $((execution_time % 60))s"
curl "http://$RUNNER/gpu/log_test_vram?id=${{ github.run_id }}&gpu=${{ env.CUDA_VISIBLE_DEVICES }}&range=$execution_time&unit=second&test=${{ matrix.test_script }}"
- name: Release GPU
if: always() && !contains(matrix.test_script, 'ipex') && !contains(matrix.test_script, 'xpu')
run: curl -X GET "http://$RUNNER/gpu/release?id=${{ github.run_id }}&gpu=${{ env.CUDA_VISIBLE_DEVICES }}&timestamp=${{ env.STEP_TIMESTAMP }}&test=${{ matrix.test_script }}&runner=${RUNNER_NAME}"
- name: Clean cache
if: always()
run: |
# rm ~/.cache/evalplus/*pkl || true
pip cache purge && uv cache clean && rm -rf ./* ./.*
show-statistics:
runs-on: [ self-hosted, xeon5 ]
if: github.event.inputs.exclusive-gpu
container:
image: modelcloud/gptqmodel:alpine-ci-v1
needs:
- legacy
- torch
steps:
- name: Print statistics
run: curl "http://$RUNNER/gpu/get_vram_logs?id=${{ github.run_id }}"
m4:
runs-on: [ self-hosted, m4 ]
needs:
- check-vm
- list-test-files
if: (github.event.inputs.test_names == '' || contains(github.event.inputs.test_names, 'apple') || contains(github.event.inputs.test_names, 'mlx') ) && (needs.list-test-files.outputs.m4-files != '' && needs.list-test-files.outputs.m4-files != '[]') && !cancelled()
strategy:
fail-fast: false
matrix:
test_script: ${{ fromJSON(needs.list-test-files.outputs.m4-files) }}
steps:
- name: Print Env
run: |
echo "repo: ${{ env.repo }}"
echo "ref: ${{ env.ref }}"
ls -ahl .
- name: Checkout Codes
uses: actions/checkout@v4
with:
repository: ${{ env.repo }}
ref: ${{ env.ref }}
- name: Run test
run: |
export PATH="/opt/homebrew/bin:$PATH" && eval "$(pyenv init -)"
rm -rf venv || true
echo "=== checking models dir is mounted"
ls ../../../monster
echo "=== activating venv"
pyenv global 3.11.11 && python -m venv venv
source venv/bin/activate
rm profile.sb || true
curl -O http://$RUNNER/scripts/m4/profile.sb
echo "=== installing uv setuptools build"
pip install uv setuptools build -U -i http://$RUNNER/simple --trusted-host $RUNNER --extra-index-url https://pypi.org/simple
echo "=== installing test tools"
uv pip install pytest parameterized vllm lm-eval device-smi mlx-lm -U -i http://$RUNNER/simple/ --trusted-host $RUNNER --extra-index-url https://pypi.org/simple
echo "=== installing gptqmodel"
uv pip install . --no-build-isolation -i http://$RUNNER/simple/ --trusted-host $RUNNER --extra-index-url https://pypi.org/simple
echo "replacing model path"
find tests -name "*.py" -exec sed -i '' 's/\/monster\/data\/model/..\/..\/..\/monster/g' {} +
TEST=${{ matrix.test_script }}
if [[ ! "$TEST" == *.py ]]; then
TEST="$TEST.py"
fi
echo "=== running test: $TEST"
pytest tests/$TEST
- name: Clean cache
if: always()
run: |
source venv/bin/activate && pip cache purge && uv cache clean || true
rm -rf ../GPTQModel && mkdir ../GPTQModel