Skip to content

Unit Tests

Unit Tests #563

Workflow file for this run

name: Unit Tests
defaults:
run:
shell: bash -le {0}
on:
repository_dispatch:
workflow_dispatch:
inputs:
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: ''
artifact_id:
description: 'Run id for artifact to be downloaded'
required: false
default: ''
max-parallel:
description: 'max parallel jobs'
required: false
default: '6'
env:
CUDA_DEVICE_ORDER: PCI_BUS_ID
CUDA_VISIBLE_DEVICES: 0
MAX_JOBS: 4
ZEN4_SERVER: 10.0.14.248
TRANSFORMERS_DIFF_TESTS: "models/test_internlm,models/test_internlm2_5,models/test_xverse"
TORCH_2_5_TESTS: "test_q4_ipex.py,test_save_loaded_quantized_model,test_quant_formats"
IGNORED_TEST_FILES: "test_tgi.py,test_gptneox.py,models/test_mixtral"
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: self-hosted
container:
image: modelcloud/gptqmodel:alpine-ci-v1
outputs:
ip: ${{ steps.get_ip.outputs.ip }}
tag: ${{ steps.get_ip.outputs.tag }}
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: ${{ env.ref }}"
- name: Select server
id: get_ip
run: |
echo "ip=$ZEN4_SERVER" >> "$GITHUB_OUTPUT"
echo "tag=zen4" >> "$GITHUB_OUTPUT"
echo "GPU_IP=$ZEN4_SERVER" >> $GITHUB_ENV
echo "tag: $tag, 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\": ${max_p:-6}}"
echo "max-parallel=$max_p" >> "$GITHUB_OUTPUT"
echo "max-parallel=$max_p"
list-test-files:
runs-on: ubuntu-latest
outputs:
torch-2-5-files: ${{ steps.files.outputs.torch-2-5-files }}
gpu-files: ${{ steps.files.outputs.gpu-files }}
transformers-files: ${{ steps.files.outputs.transformers-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
cpu_file_list = [f.strip().removesuffix('.py') for f in '${TORCH_2_5_TESTS}'.split(',') if f.strip()]
test_files_list = [f.strip().removesuffix('.py') for f in '${{ github.event.inputs.test_names }}'.split(',') if f.strip()]
cpu_test_files = [f for f in cpu_file_list if not test_files_list or f in test_files_list]
transformers_diff_list = [f.strip().removesuffix('.py') for f in '${TRANSFORMERS_DIFF_TESTS}'.split(',') if f.strip()]
transformers_test_files = [f for f in transformers_diff_list if not test_files_list or f in test_files_list]
all_tests = [f.removesuffix('.py') for f in os.listdir('tests/') if f.startswith('test_') and f.endswith('.py') and f.strip().removesuffix('.py') not in '${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 '${IGNORED_TEST_FILES}']
all_tests = [item for item in all_tests+all_tests_models if item.strip().removesuffix('.py') not in '${TORCH_2_5_TESTS}']
gpu_test_files = [f for f in all_tests if f not in cpu_file_list and (not test_files_list or f in test_files_list) and f not in transformers_diff_list]
print(f'{json.dumps(cpu_test_files)}|{json.dumps(gpu_test_files)}|{json.dumps(transformers_test_files)}')
"
test_files=$(python3 -c "$script")
IFS='|' read -r cpu_test_files gpu_test_files transformers_test_files <<< "$test_files"
echo "torch-2-5-files=$cpu_test_files" >> "$GITHUB_OUTPUT"
echo "gpu-files=$gpu_test_files" >> "$GITHUB_OUTPUT"
echo "transformers-files=$transformers_test_files" >> "$GITHUB_OUTPUT"
echo "Test files: $test_files"
echo "CPU Test files: $cpu_test_files"
echo "GPU Test files: $gpu_test_files"
echo "Transformers Test files: $transformers_test_files"
echo "Ignored Test files: $IGNORED_TEST_FILES"
build-zen4:
runs-on: [self-hosted, zen4]
needs: check-vm
if: needs.check-vm.outputs.tag == 'zen4' && github.event.inputs.artifact_id == '' && !cancelled()
container:
image: ${{ needs.check-vm.outputs.ip }}:5000/modelcloud/gptqmodel:github-ci-v2-torch2.4.1
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
- name: Install requirements
run: |
bash -c "$(curl -L http://$ZEN4_SERVER/scripts/compiler/init_env.sh)" @ 12.4 2.4.1 3.11
uv pip install transformers -U
- name: Compile
timeout-minutes: 35
run: python setup.py bdist_wheel
- name: Test install
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/$whl
- name: Upload wheel
continue-on-error: true
run: |
curl -s -F "runid=${{ github.run_id }}" -F "repo=${{ env.repo }}" -F "ref=${{ env.ref }}" -F "sha256=${{ env.WHL_HASH }}" -F "file=@dist/${{ env.WHL_NAME }}" http://${{ needs.check-vm.outputs.ip }}/gpu/whl/upload
- name: Upload to artifact
uses: actions/upload-artifact@v4
with:
name: dist
path: dist
torch2_4:
needs:
- build-zen4
- list-test-files
- check-vm
runs-on: self-hosted
if: always() && !cancelled() && (needs.build-zen4.result == 'success' || github.event.inputs.artifact_id != '')
container:
image: ${{ needs.check-vm.outputs.ip }}:5000/modelcloud/gptqmodel:github-ci-v2-torch2.4.1
volumes:
- /home/ci/models:/monster/data/model
strategy:
fail-fast: false
max-parallel: ${{ fromJson(needs.check-vm.outputs.max-parallel).size || 6 }}
matrix:
test_script: ${{ fromJSON(needs.list-test-files.outputs.gpu-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
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" >> $GITHUB_ENV
echo "RUN_ID=${run_id}"
if grep -q "bitblas" tests/${{ matrix.test_script }}.py; then
echo "BITBLAS=1" >> $GITHUB_ENV
fi
- name: Download wheel
continue-on-error: true
run: |
file_name=$(curl -s -F "runid=${{ needs.check-vm.outputs.run_id }}" -F "repo=${{ env.repo }}" -F "ref=${{ env.ref }}" -F "fuzz=1" "http://${{ needs.check-vm.outputs.ip }}/gpu/whl/download")
if echo "$file_name" | grep -q "gptqmodel"; then
mkdir dist || true
cd dist
curl -s -O http://${{ needs.check-vm.outputs.ip }}/whl/${{ env.repo }}/${{ needs.check-vm.outputs.run_id }}/$file_name
ls -ahl .
sha256=$(sha256sum $file_name)
echo "sha256=$sha256"
echo "DOWNLOADED=1" >> $GITHUB_ENV
fi
- name: Download artifact
if: env.DOWNLOADED == '' && !cancelled()
uses: actions/download-artifact@v4
with:
name: dist
path: dist
run-id: ${{ needs.check-vm.outputs.run_id }}
- name: Install wheel
run: |
uv pip install auto_round optimum bitblas==0.0.1.dev13 parameterized uvicorn -i http://${{ needs.check-vm.outputs.ip }}/simple/ --trusted-host ${{ needs.check-vm.outputs.ip }}
uv pip install dist/*.whl
bash -c "$(curl -L http://$ZEN4_SERVER/scripts/compiler/init_env.sh)" @ 12.4 2.4.1 3.11
uv pip install transformers numpy==1.26.4 -U -i http://${{ needs.check-vm.outputs.ip }}/simple/ --trusted-host ${{ needs.check-vm.outputs.ip }}
- name: Check platform
run: |
ip=${ZEN4_SERVER}
echo "GPU_IP=$ip" >> $GITHUB_ENV
echo "-----------"
pip show torch
echo "-----------"
nvcc --version
- name: Find suitable GPU
run: |
timestamp=$(date +%s%3N)
gpu_id=-1
while [ "$gpu_id" -lt 0 ]; do
gpu_id=$(curl -s "http://${{ env.GPU_IP }}/gpu/get?id=${{ github.run_id }}&timestamp=$timestamp")
if [ "$gpu_id" -lt 0 ]; then
echo "http://${{ env.GPU_IP }}/gpu/get?id=${{ github.run_id }}&timestamp=$timestamp returned $gpu_id"
echo "No available GPU, waiting 5 seconds..."
sleep 5
else
echo "Allocated GPU ID: $gpu_id"
fi
done
echo "CUDA_VISIBLE_DEVICES=$gpu_id" >> $GITHUB_ENV
echo "STEP_TIMESTAMP=$timestamp" >> $GITHUB_ENV
echo "CUDA_VISIBLE_DEVICES set to $gpu_id, timestamp=$timestamp"
- name: Run tests
if: ${{ (!github.event.inputs.test_names || contains(github.event.inputs.test_names, matrix.test_script)) && !cancelled() }}
run: pytest --durations=0 tests/${{ matrix.test_script }}.py || { echo "ERROR=1" >> $GITHUB_ENV; exit 1; }
- name: Clear cache
if: failure() && env.BITBLAS == '1' && env.ERROR == '1'
run: |
rm -rf ~/.cache/bitblas/nvidia/geforce-rtx-4090
echo "clear bitblas cache"
- name: Release GPU
if: always()
run: curl -X GET "http://${{ env.GPU_IP }}/gpu/release?id=${{ github.run_id }}&gpu=${{ env.CUDA_VISIBLE_DEVICES }}&timestamp=${{ env.STEP_TIMESTAMP }}"
transformers_diff:
needs:
- build-zen4
- list-test-files
- check-vm
runs-on: self-hosted
if: always() && !cancelled() && (needs.build-zen4.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-v2-torch2.4.1
volumes:
- /home/ci/models:/monster/data/model
strategy:
fail-fast: false
max-parallel: ${{ fromJson(needs.check-vm.outputs.max-parallel).size || 6 }}
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
ip=${{ needs.check-vm.outputs.ip }}
echo "GPU_IP=$ip" >> $GITHUB_ENV
- name: Download wheel
continue-on-error: true
run: |
file_name=$(curl -s -F "runid=${{ needs.check-vm.outputs.run_id }}" -F "repo=${{ env.repo }}" -F "ref=${{ env.ref }}" -F "fuzz=1" "http://${{ needs.check-vm.outputs.ip }}/gpu/whl/download")
if echo "$file_name" | grep -q "gptqmodel"; then
mkdir dist || true
cd dist
curl -s -O http://${{ needs.check-vm.outputs.ip }}/whl/${{ env.repo }}/${{ needs.check-vm.outputs.run_id }}/$file_name
ls -ahl .
sha256=$(sha256sum $file_name)
echo "sha256=$sha256"
echo "DOWNLOADED=1" >> $GITHUB_ENV
fi
- name: Download artifact
if: env.DOWNLOADED == '' && !cancelled()
uses: actions/download-artifact@v4
with:
name: dist
path: dist
run-id: ${{ needs.check-vm.outputs.run_id }}
- name: Install wheel
run: |
uv pip install auto_round optimum bitblas==0.0.1.dev13 parameterized uvicorn -i http://${{ needs.check-vm.outputs.ip }}/simple/ --trusted-host ${{ needs.check-vm.outputs.ip }}
uv pip install dist/*.whl
bash -c "$(curl -L http://$ZEN4_SERVER/scripts/compiler/init_env.sh)" @ 12.4 2.4.1 3.11
uv pip install transformers==4.38.2 numpy==1.26.4 -U -i http://${{ needs.check-vm.outputs.ip }}/simple/ --trusted-host ${{ needs.check-vm.outputs.ip }}
if [ "${{ matrix.test_script }}" = "test_xverse" ]; then
uv pip install tokenizers==0.15.2
fi
- name: Find suitable GPU
run: |
timestamp=$(date +%s%3N)
gpu_id=-1
while [ "$gpu_id" -lt 0 ]; do
gpu_id=$(curl -s "http://${{ env.GPU_IP }}/gpu/get?id=${{ github.run_id }}&timestamp=$timestamp")
if [ "$gpu_id" -lt 0 ]; then
echo "http://${{ env.GPU_IP }}/gpu/get?id=${{ github.run_id }}&timestamp=$timestamp returned $gpu_id"
echo "No available GPU, waiting 5 seconds..."
sleep 5
else
echo "Allocated GPU ID: $gpu_id"
fi
done
echo "CUDA_VISIBLE_DEVICES=$gpu_id" >> $GITHUB_ENV
echo "STEP_TIMESTAMP=$timestamp" >> $GITHUB_ENV
echo "CUDA_VISIBLE_DEVICES set to $gpu_id, timestamp=$timestamp"
- name: Run tests
if: ${{ (!github.event.inputs.test_names || contains(github.event.inputs.test_names, matrix.test_script)) && !cancelled() }}
run: pytest --durations=0 tests/${{ matrix.test_script }}.py || { echo "ERROR=1" >> $GITHUB_ENV; exit 1; }
- name: Release GPU
if: always()
run: curl -X GET "http://${{ env.GPU_IP }}/gpu/release?id=${{ github.run_id }}&gpu=${{ env.CUDA_VISIBLE_DEVICES }}&timestamp=${{ env.STEP_TIMESTAMP }}"
torch2_5:
needs:
- build-zen4
- list-test-files
- check-vm
runs-on: self-hosted
if: always() && !cancelled() && (needs.build-zen4.result == 'success' || github.event.inputs.artifact_id != '') && needs.list-test-files.outputs.torch-2-5-files != '[]'
container:
image: ${{ needs.check-vm.outputs.ip }}:5000/modelcloud/gptqmodel:github-ci-v2-torch2.5.1
volumes:
- /home/ci/models:/monster/data/model
strategy:
fail-fast: false
max-parallel: ${{ fromJson(needs.check-vm.outputs.max-parallel).size || 6 }}
matrix:
test_script: ${{ fromJSON(needs.list-test-files.outputs.torch-2-5-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
ip=${{ needs.check-vm.outputs.ip }}
echo "GPU_IP=$ip" >> $GITHUB_ENV
- name: Download wheel
continue-on-error: true
run: |
file_name=$(curl -s -F "runid=${{ needs.check-vm.outputs.run_id }}" -F "repo=${{ env.repo }}" -F "ref=${{ env.ref }}" -F "fuzz=1" "http://${{ needs.check-vm.outputs.ip }}/gpu/whl/download")
if echo "$file_name" | grep -q "gptqmodel"; then
mkdir dist || true
cd dist
curl -s -O http://${{ needs.check-vm.outputs.ip }}/whl/${{ env.repo }}/${{ needs.check-vm.outputs.run_id }}/$file_name
ls -ahl .
sha256=$(sha256sum $file_name)
echo "sha256=$sha256"
echo "DOWNLOADED=1" >> $GITHUB_ENV
fi
- name: Download artifact
if: env.DOWNLOADED == '' && !cancelled()
uses: actions/download-artifact@v4
with:
name: dist
path: dist
run-id: ${{ needs.check-vm.outputs.run_id }}
- name: Install wheel
run: |
uv pip install device-smi -i https://pypi.org/simple
uv pip install intel_extension_for_pytorch auto_round bitblas==0.0.1.dev13 dist/*.whl -i http://${{ needs.check-vm.outputs.ip }}/simple/ --trusted-host ${{ needs.check-vm.outputs.ip }}
- name: Run tests
if: ${{ (!github.event.inputs.test_names || contains(github.event.inputs.test_names, matrix.test_script)) && !cancelled() }}
run: pytest --durations=0 tests/${{ matrix.test_script }}.py
- name: Release GPU
if: always()
run: curl -X GET "http://${{ env.GPU_IP }}/gpu/release?id=${{ github.run_id }}&gpu=${{ env.CUDA_VISIBLE_DEVICES }}&timestamp=${{ env.STEP_TIMESTAMP }}"