LLM Performance Test #56
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
name: LLM Performance Test | |
# Cancel previous runs in the PR when you push new commits | |
concurrency: | |
group: ${{ github.workflow }}-llm-performance-tests-${{ github.event.pull_request.number || github.run_id }} | |
cancel-in-progress: true | |
# Controls when the action will run. | |
on: | |
schedule: | |
- cron: "00 13 * * *" # GMT time, 13:00 GMT == 21:00 China | |
pull_request: | |
branches: [main] | |
paths: | |
- ".github/workflows/llm_performance_tests.yml" | |
- ".github/workflows/llm-binary-build.yml" | |
- ".github/actions/llm/setup-llm-env/action.yml" | |
- ".github/actions/llm/remove-llm-env/action.yml" | |
- ".github/actions/llm/download-llm-binary/action.yml" | |
- "python/llm/test/benchmark/**" | |
workflow_dispatch: | |
workflow_call: | |
# 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-performance-test: | |
needs: llm-cpp-build | |
strategy: | |
fail-fast: false | |
matrix: | |
python-version: ["3.9"] | |
instruction: ["AVX512"] | |
runs-on: [self-hosted, llm, perf] | |
env: | |
THREAD_NUM: 24 | |
steps: | |
- uses: actions/checkout@v2 | |
- name: Set up Python ${{ matrix.python-version }} | |
uses: actions/setup-python@v2 | |
with: | |
python-version: ${{ matrix.python-version }} | |
- name: Install dependencies | |
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 | |
env: | |
ANALYTICS_ZOO_ROOT: ${{ github.workspace }} | |
- name: Run LLM Performance test | |
env: | |
ANALYTICS_ZOO_ROOT: ${{ github.workspace }} | |
run: bash python/llm/dev/benchmark/run-benchmark-tests.sh | |
# - name: Clean up test environment | |
# uses: ./.github/actions/llm/remove-llm-env | |
# env: | |
# ANALYTICS_ZOO_ROOT: ${{ github.workspace }} | |
llm-performance-test-on-arc: | |
needs: llm-cpp-build | |
strategy: | |
fail-fast: false | |
matrix: | |
python-version: ["3.9"] | |
runs-on: [self-hosted, llm, perf] | |
env: | |
OMP_NUM_THREADS: 16 | |
THREAD_NUM: 16 | |
ANALYTICS_ZOO_ROOT: ${{ github.workspace }} | |
steps: | |
- name: Set model directories | |
shell: bash | |
run: | | |
echo "ORIGIN_DIR=/mnt/disk1/models" >> "$GITHUB_ENV" | |
- name: Set environment variables | |
shell: bash | |
run: | | |
echo "LLAMA2_7B_ORIGIN_PATH=${ORIGIN_DIR}/Llama-2-7b-chat-hf" >> "$GITHUB_ENV" | |
echo "LLAMA2_13B_ORIGIN_PATH=${ORIGIN_DIR}/Llama-2-13b-chat-hf" >> "$GITHUB_ENV" | |
echo "CHATGLM2_6B_ORIGIN_PATH=${ORIGIN_DIR}/chatglm2-6b" >> "$GITHUB_ENV" | |
echo "WHISPER_MEDIUM_ORIGIN_PATH=${ORIGIN_DIR}/whisper-medium" >> "$GITHUB_ENV" | |
- uses: actions/checkout@v3 | |
- name: Set up Python ${{ matrix.python-version }} | |
uses: actions/setup-python@v4 | |
with: | |
python-version: ${{ matrix.python-version }} | |
- name: Install dependencies | |
shell: bash | |
run: | | |
python -m pip install --upgrade pip | |
python -m pip install --upgrade setuptools | |
python -m pip install --upgrade wheel | |
- name: Download llm binary | |
uses: ./.github/actions/llm/download-llm-binary | |
- name: Run LLM install (all) test | |
uses: ./.github/actions/llm/setup-llm-env | |
with: | |
extra-dependency: "xpu" | |
- name: Test installed xpu version | |
shell: bash | |
run: | | |
source /opt/intel/oneapi/setvars.sh | |
bash python/llm/test/run-llm-install-tests.sh | |
- name: Test on xpu | |
shell: bash | |
run: | | |
source /opt/intel/oneapi/setvars.sh | |
export USE_XETLA=OFF | |
export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1 | |
cd python/llm/test/benchmark/gpu | |
export http_proxy=${HTTP_PROXY} | |
export https_proxy=${HTTPS_PROXY} | |
rm -rf test-result || true | |
mkdir test-result | |
taskset -c 0-$((THREAD_NUM - 1)) python llama2.py --model-dir="${LLAMA2_7B_ORIGIN_PATH}" --input-tokens=32 --max-new-tokens=32 > test-result/llama2_7b-32-32.log | |
taskset -c 0-$((THREAD_NUM - 1)) python llama2.py --model-dir="${LLAMA2_7B_ORIGIN_PATH}" --input-tokens=1024 --max-new-tokens=1024 > test-result/llama2_7b-1024-1024.log | |
taskset -c 0-$((THREAD_NUM - 1)) python llama2.py --model-dir="${LLAMA2_13B_ORIGIN_PATH}" --input-tokens=32 --max-new-tokens=32 > test-result/llama2_13b-32-32.log | |
taskset -c 0-$((THREAD_NUM - 1)) python llama2.py --model-dir="${LLAMA2_13B_ORIGIN_PATH}" --input-tokens=1024 --max-new-tokens=1024 > test-result/llama2_13b-1024-1024.log | |
taskset -c 0-$((THREAD_NUM - 1)) python chatglm2.py --model-dir="${CHATGLM2_6B_ORIGIN_PATH}" --input-tokens=32 --max-new-tokens=32 > test-result/chatglm2_6b-32-32.log | |
taskset -c 0-$((THREAD_NUM - 1)) python chatglm2.py --model-dir="${CHATGLM2_6B_ORIGIN_PATH}" --input-tokens=1024 --max-new-tokens=1024 > test-result/chatglm2_6b-1024-1024.log | |
taskset -c 0-$((THREAD_NUM - 1)) python whisper.py --model-dir="${WHISPER_MEDIUM_ORIGIN_PATH}" > test-result/whisper_medium-default-default.log | |
python ../analyze_log_dir.py --log-dir=./test-result --output-path=./xpu_latency.csv | |
timestamp=`date '+%Y%m%d'` | |
curl -T ./xpu_latency.csv ${LLM_FTP_URL}/llm/ggml-actions/perf/xpu_lantency_$timestamp.csv |