cb228c70ea
* modify output_path as a directory * schedule nightly at 21 on Friday * add tasks and models for nightly * add accuracy regression * comment out if to test * mixed fp4 * for test * add missing delimiter * remove comma * fixed golden results * add mixed 4 golden result * add more options * add mistral results * get golden result of stable lm * move nightly scripts and results to test folder * add license * add fp8 stable lm golden * run on all available devices * trigger only when ready for review * fix new line * update golden * add mistral
115 lines
4.1 KiB
YAML
115 lines
4.1 KiB
YAML
name: LLM Harness Evalution
|
|
|
|
# Cancel previous runs in the PR when you push new commits
|
|
concurrency:
|
|
group: ${{ github.workflow }}-llm-nightly-test-${{ github.event.pull_request.number || github.run_id }}
|
|
cancel-in-progress: true
|
|
|
|
# Controls when the action will run.
|
|
on:
|
|
schedule:
|
|
- cron: "00 13 * * 5" # GMT time, 13:00 GMT == 21:00 China
|
|
pull_request:
|
|
types: ready_for_review
|
|
branches: [main]
|
|
paths:
|
|
- ".github/workflows/llm-harness-evaluation.yml"
|
|
# Allows you to run this workflow manually from the Actions tab
|
|
workflow_dispatch:
|
|
|
|
# A workflow run is made up of one or more jobs that can run sequentially or in parallel
|
|
jobs:
|
|
llm-cpp-build:
|
|
uses: ./.github/workflows/llm-binary-build.yml
|
|
llm-harness-evalution:
|
|
timeout-minutes: 1000
|
|
needs: llm-cpp-build
|
|
strategy:
|
|
fail-fast: false
|
|
matrix:
|
|
# include:
|
|
# python-version: "3.9"
|
|
# model_name: "stablelm-3b-4e1t"
|
|
# task: "arc"
|
|
# precision: "sym_int4" #options: sym_int4, fp4, mixed_fp4, sym_int8, fp8, mixed_fp8
|
|
python-version: ["3.9"]
|
|
model_name: [stablelm-3b-4e1t,Mistral-7B-v0.1]
|
|
task: [truthfulqa, arc]
|
|
precision: [mixed_fp4, fp8]
|
|
device: [xpu]
|
|
|
|
runs-on: [self-hosted, llm, accuracy]
|
|
env:
|
|
ANALYTICS_ZOO_ROOT: ${{ github.workspace }}
|
|
ORIGIN_DIR: /mnt/disk1/models
|
|
HARNESS_HF_HOME: /mnt/disk1/harness_home
|
|
steps:
|
|
- uses: actions/checkout@v3
|
|
- name: Set up Python ${{ matrix.python-version }}
|
|
uses: actions/setup-python@v4
|
|
with:
|
|
python-version: ${{ matrix.python-version }}
|
|
- name: Install dependencies
|
|
shell: bash
|
|
run: |
|
|
python -m pip install --upgrade pip
|
|
python -m pip install --upgrade setuptools==58.0.4
|
|
python -m pip install --upgrade wheel
|
|
|
|
- name: Download llm binary
|
|
uses: ./.github/actions/llm/download-llm-binary
|
|
|
|
- name: Run LLM install (all) test
|
|
uses: ./.github/actions/llm/setup-llm-env
|
|
with:
|
|
extra-dependency: "xpu"
|
|
|
|
- name: Install harness
|
|
working-directory: ${{ github.workspace }}/python/llm/dev/benchmark/harness/
|
|
shell: bash
|
|
run: |
|
|
git clone https://github.com/EleutherAI/lm-evaluation-harness.git
|
|
cd lm-evaluation-harness
|
|
git checkout e81d3cc
|
|
pip install -e .
|
|
|
|
- name: Download models and datasets
|
|
shell: bash
|
|
run: |
|
|
echo "MODEL_PATH=${ORIGIN_DIR}/${{ matrix.model_name }}/" >> "$GITHUB_ENV"
|
|
MODEL_PATH=${ORIGIN_DIR}/${{ matrix.model_name }}/
|
|
if [ ! -d $HARNESS_HF_HOME ]; then
|
|
mkdir -p $HARNESS_HF_HOME
|
|
fi
|
|
wget -r -nH -nc -l inf --no-verbose --cut-dirs=2 ${LLM_FTP_URL}/llm/LeaderBoard_Datasets/ -P $HARNESS_HF_HOME/
|
|
wget -r -nH -nc --no-verbose --cut-dirs=1 ${LLM_FTP_URL}/llm/${{ matrix.model_name }} -P ${ORIGIN_DIR}
|
|
|
|
- name: Upgrade packages
|
|
shell: bash
|
|
run: |
|
|
pip install --upgrade transformers==4.34.0
|
|
|
|
- name: Run harness
|
|
shell: bash
|
|
working-directory: ${{ github.workspace }}/python/llm/dev/benchmark/harness
|
|
env:
|
|
USE_XETLA: OFF
|
|
# SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS: 1
|
|
run: |
|
|
export HF_HOME=${HARNESS_HF_HOME}
|
|
export HF_DATASETS=$HARNESS_HF_HOME/datasets
|
|
export HF_DATASETS_CACHE=$HARNESS_HF_HOME/datasets
|
|
source /opt/intel/oneapi/setvars.sh
|
|
python run_llb.py --model bigdl-llm --pretrained ${MODEL_PATH} --precision ${{ matrix.precision }} --device ${{ matrix.device }} --tasks ${{ matrix.task }} --batch_size 1 --no_cache --output_path results
|
|
|
|
|
|
- name: Compare with golden accuracy
|
|
shell: bash
|
|
|
|
working-directory: ${{ github.workspace }}/python/llm
|
|
run: |
|
|
python test/benchmark/harness_nightly/accuracy_regression.py dev/benchmark/harness/results/${{ matrix.model_name }}/${{ matrix.device }}/${{ matrix.precision }}/${{ matrix.task }}/result.json test/benchmark/harness_nightly/golden_results.json
|
|
|
|
|
|
|
|
|