Skip to content

daily_run_test

daily_run_test #287

Workflow file for this run

name: daily_run_test
on:
workflow_dispatch:
schedule:
- cron: '56 16 * * *'
concurrency:
group: ${{ github.workflow }}-${{ github.ref }}
cancel-in-progress: true
env:
CONDA_ENV: opencompass_regression
PIP_CACHE_PATH: /cpfs01/user/qa-llm-cicd/.cache/pip
USERSPACE_PREFIX: /cpfs01/user/qa-llm-cicd
HF_CACHE_PATH: /cpfs01/shared/public/public_hdd/llmeval/model_weights/hf_hub
DATEASET_CACHE_PATH: /cpfs01/shared/public/public_hdd/llmeval/llm-evaluation-datasets
HF_DATASETS_OFFLINE: 1
TRANSFORMERS_OFFLINE: 1
HF_HUB_OFFLINE: 1
TRITON_PTXAS_PATH: /usr/local/cuda/bin/ptxas
jobs:
daily_run_test:
runs-on: self-hosted
environment: 'prod'
timeout-minutes: 420 #7hours
steps:
- name: Clone repository
uses: actions/checkout@v2
- name: Prepare - create conda env and install torch
run: |
. /cpfs01/shared/public/qa-llm-cicd/miniconda3/bin/activate
conda create -y --name ${{env.CONDA_ENV}} python=3.10
conda activate ${{env.CONDA_ENV}}
pip install /cpfs01/user/qa-llm-cicd/packages/lmdeploy-0.5.0+cu118-cp310-cp310-manylinux2014_x86_64.whl --cache-dir ${{env.PIP_CACHE_PATH}}
pip install /cpfs01/user/qa-llm-cicd/packages/vllm-0.5.5+cu118-cp310-cp310-manylinux1_x86_64.whl --cache-dir ${{env.PIP_CACHE_PATH}}
pip uninstall torch torchvision torchaudio -y
pip install torch==2.4.0 torchvision==0.19.0 torchaudio==2.4.0 --cache-dir ${{env.PIP_CACHE_PATH}} --index-url https://download.pytorch.org/whl/cu118
FLASH_ATTENTION_FORCE_BUILD=TRUE pip install /cpfs01/user/qa-llm-cicd/packages/flash_attn-2.6.3+cu118torch2.4cxx11abiFALSE-cp310-cp310-linux_x86_64.whl
pip install /cpfs01/user/qa-llm-cicd/packages/xformers-0.0.27.post2+cu118-cp310-cp310-manylinux2014_x86_64.whl --cache-dir ${{env.PIP_CACHE_PATH}}
conda info --envs
- name: Prepare - Pip install code
run: |
. /cpfs01/shared/public/qa-llm-cicd/miniconda3/bin/activate
conda activate ${{env.CONDA_ENV}}
pip install -e . --cache-dir ${{env.PIP_CACHE_PATH}}
pip install human_eval transformers protobuf pytest gguf msgspec librosa vllm_flash_attn bitsandbytes--cache-dir ${{env.PIP_CACHE_PATH}}
conda info --envs
- name: Prepare - prepare data and hf model
run: |
ln -s ${{env.DATEASET_CACHE_PATH}} data
rm -rf ~/.cache/huggingface/hub -f && mkdir ~/.cache -p && mkdir ~/.cache/huggingface -p
ln -s ${{env.HF_CACHE_PATH}} ~/.cache/huggingface/hub
- name: Run chat model test
run: |
. /cpfs01/shared/public/qa-llm-cicd/miniconda3/bin/activate
conda activate ${{env.CONDA_ENV}}
conda info --envs
sed -i 's/judgemodel/'$(tail -n 1 /cpfs01/shared/public/llmeval/share_info/compassjuder_ip.txt)'/g' .github/scripts/eval_regression_chat.py
python3 run.py .github/scripts/eval_regression_chat.py --work-dir /cpfs01/user/qa-llm-cicd/report/${{ github.run_id }}/chat --reuse
rm regression_result_daily -f && ln -s /cpfs01/user/qa-llm-cicd/report/${{ github.run_id }}/chat/*/summary regression_result_daily
python -m pytest -m chat -s -v --color=yes .github/scripts/oc_score_assert.py
- name: Run base model test
run: |
. /cpfs01/shared/public/qa-llm-cicd/miniconda3/bin/activate
conda activate ${{env.CONDA_ENV}}
conda info --envs
python3 run.py .github/scripts/eval_regression_base.py --work-dir /cpfs01/user/qa-llm-cicd/report/${{ github.run_id }}/base --reuse
rm regression_result_daily -f && ln -s /cpfs01/user/qa-llm-cicd/report/${{ github.run_id }}/base/*/summary regression_result_daily
python -m pytest -m base -s -v --color=yes .github/scripts/oc_score_assert.py
- name: Run command testcase
run: |
. /cpfs01/shared/public/qa-llm-cicd/miniconda3/bin/activate
conda activate ${{env.CONDA_ENV}}
conda info --envs
export from_tf=TRUE
python tools/list_configs.py internlm2_5 mmlu
python run.py --models hf_internlm2_5_7b --datasets race_ppl --work-dir /cpfs01/user/qa-llm-cicd/report/${{ github.run_id }}/cmd1 --reuse
rm regression_result_daily -f && ln -s /cpfs01/user/qa-llm-cicd/report/${{ github.run_id }}/cmd1/*/summary regression_result_daily
python -m pytest -m case1 -s -v --color=yes .github/scripts/oc_score_assert.py
python run.py --models hf_internlm2_5_7b_chat --datasets race_gen -a lmdeploy --work-dir /cpfs01/user/qa-llm-cicd/report/${{ github.run_id }}/cmd2 --reuse
rm regression_result_daily -f && ln -s /cpfs01/user/qa-llm-cicd/report/${{ github.run_id }}/cmd2/*/summary regression_result_daily
python -m pytest -m case2 -s -v --color=yes .github/scripts/oc_score_assert.py
python run.py --datasets race_ppl --hf-type base --hf-path internlm/internlm2_5-7b --work-dir /cpfs01/user/qa-llm-cicd/report/${{ github.run_id }}/cmd3 --reuse
rm regression_result_daily -f && ln -s /cpfs01/user/qa-llm-cicd/report/${{ github.run_id }}/cmd3/*/summary regression_result_daily
python -m pytest -m case3 -s -v --color=yes .github/scripts/oc_score_assert.py
python run.py --datasets race_gen --hf-type chat --hf-path internlm/internlm2_5-7b-chat --work-dir /cpfs01/user/qa-llm-cicd/report/${{ github.run_id }}/cmd4 --reuse
rm regression_result_daily -f && ln -s /cpfs01/user/qa-llm-cicd/report/${{ github.run_id }}/cmd4/*/summary regression_result_daily
python -m pytest -m case4 -s -v --color=yes .github/scripts/oc_score_assert.py
- name: Remove Conda Env
if: always()
run: |
rm -rf regression_result_daily
. /cpfs01/shared/public/qa-llm-cicd/miniconda3/bin/activate
conda env remove -y --name ${{env.CONDA_ENV}}
conda info --envs
notify_to_feishu:
if: ${{ always() && !cancelled() && contains(needs.*.result, 'failure') && (github.ref_name == 'develop' || github.ref_name == 'main') }}
needs: [daily_run_test]
environment: 'prod'
timeout-minutes: 5
runs-on: self-hosted
steps:
- name: notify
run: |
curl -X POST -H "Content-Type: application/json" -d '{"msg_type":"post","content":{"post":{"zh_cn":{"title":"Opencompass- Daily test failed","content":[[{"tag":"text","text":"branch: ${{github.ref_name}}, run action: ${{github.workflow}} failed. "},{"tag":"a","text":"Please click here for details ","href":"https://github.com/'${{ github.repository }}'/actions/runs/'${GITHUB_RUN_ID}'"},{"tag":"at","user_id":"'${{ secrets.USER_ID }}'"}]]}}}}' ${{ secrets.WEBHOOK_URL }}