Fix wrong token latency when batch size is greater than 1 (#1244) #2042
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
| # This workflow will install Python dependencies, run tests and lint with a single version of Python | |
| # For more information see: https://docs.github.com/en/actions/automating-builds-and-tests/building-and-testing-python | |
| name: llm_bench Python Test | |
| on: | |
| push: | |
| branches: [ "master" ] | |
| paths: | |
| - tools/llm_bench/** | |
| - tools/who_what_benchmark/** | |
| pull_request: | |
| paths: | |
| - tools/llm_bench/** | |
| - tools/who_what_benchmark/** | |
| - .github/workflows/llm_bench-python.yml | |
| permissions: read-all # Required by https://github.com/ossf/scorecard/blob/e23b8ad91fd6a64a0a971ca4fc0a4d1650725615/docs/checks.md#token-permissions | |
| concurrency: | |
| # github.ref is not unique in post-commit | |
| group: ${{ github.event_name == 'push' && github.run_id || github.ref }}-llm-bench-python | |
| cancel-in-progress: true | |
| env: | |
| LLM_BENCH_PYPATH: tools/llm_bench | |
| WWB_PATH: tools/who_what_benchmark | |
| jobs: | |
| build: | |
| runs-on: ubuntu-latest | |
| strategy: | |
| fail-fast: false | |
| matrix: | |
| python-version: ["3.10"] | |
| steps: | |
| - uses: actions/checkout@v4 | |
| - name: Set up Python ${{ matrix.python-version }} | |
| uses: actions/setup-python@v3 | |
| with: | |
| python-version: ${{ matrix.python-version }} | |
| - name: Install dependencies | |
| run: | | |
| python -m pip install --upgrade pip | |
| python -m pip install flake8 pytest black | |
| GIT_CLONE_PROTECTION_ACTIVE=false pip install -r ${{ env.LLM_BENCH_PYPATH }}/requirements.txt | |
| python -m pip install -U --pre openvino openvino-tokenizers openvino-genai --extra-index-url https://storage.openvinotoolkit.org/simple/wheels/nightly | |
| - name: Lint with flake8 | |
| run: | | |
| # stop the build if there are Python syntax errors or undefined names | |
| python -m flake8 ${{ env.LLM_BENCH_PYPATH }} --config=${{ env.LLM_BENCH_PYPATH }}/setup.cfg | |
| python -m flake8 ${{ env.WWB_PATH }} --config=${{ env.WWB_PATH }}/setup.cfg | |
| - name: Create code style diff for samples | |
| if: failure() | |
| run: | | |
| python -m black -l 160 -S ${{ env.LLM_BENCH_PYPATH }}/ | |
| git diff > llm.bench_diff.diff | |
| - uses: actions/upload-artifact@v3 | |
| if: failure() | |
| with: | |
| name: llm.bench_diff | |
| path: llm.bench_diff.diff | |
| - name: Test native pytorch model on Linux | |
| run: | | |
| git clone --depth 1 https://huggingface.co/katuni4ka/tiny-random-qwen | |
| python ./tools/llm_bench/benchmark.py -m tiny-random-qwen -d cpu -n 1 -f pt | |
| env: | |
| GIT_LFS_SKIP_SMUDGE: 0 | |
| - name: Test tiny-random-baichuan2 on Linux | |
| run: | | |
| optimum-cli export openvino --model katuni4ka/tiny-random-baichuan2 --trust-remote-code --weight-format fp16 ./ov_models/tiny-random-baichuan2/pytorch/dldt/FP16 | |
| python ./tools/llm_bench/benchmark.py -m ./ov_models/tiny-random-baichuan2/pytorch/dldt/FP16/ -d cpu -n 1 | |
| - name: Test tiny-stable-diffusion on Linux | |
| run: | | |
| optimum-cli export openvino --model segmind/tiny-sd --trust-remote-code --weight-format fp16 ./ov_models/tiny-sd/pytorch/dldt/FP16/ | |
| python ./tools/llm_bench/benchmark.py -m ./ov_models/tiny-sd/pytorch/dldt/FP16/ -pf ./tools/llm_bench/prompts/stable-diffusion.jsonl -d cpu -n 1 | |
| - name: Test dreamlike-anime on Linux with GenAI | |
| run: | | |
| optimum-cli export openvino --model dreamlike-art/dreamlike-anime-1.0 --task stable-diffusion --weight-format fp16 ov_models/dreamlike-art-dreamlike-anime-1.0/FP16 | |
| python ./tools/llm_bench/benchmark.py -m ./ov_models/dreamlike-art-dreamlike-anime-1.0/FP16/ -pf ./tools/llm_bench/prompts/stable-diffusion.jsonl -d cpu -n 1 --genai | |
| - name: Test dreamlike-anime on Linux with GenAI and LoRA | |
| run: | | |
| wget -O ./ov_models/soulcard.safetensors https://civitai.com/api/download/models/72591 | |
| python ./tools/llm_bench/benchmark.py -m ./ov_models/dreamlike-art-dreamlike-anime-1.0/FP16/ -pf ./tools/llm_bench/prompts/stable-diffusion.jsonl -d cpu -n 1 --genai --lora ./ov_models/soulcard.safetensors --lora_alphas 0.7 | |
| - name: Test TinyLlama-1.1B-Chat-v1.0 in Speculative Deconding mode on Linux | |
| run: | | |
| optimum-cli export openvino --model TinyLlama/TinyLlama-1.1B-Chat-v1.0 --trust-remote-code --weight-format fp16 ov_models/TinyLlama-1.1B-Chat-v1.0/FP16 | |
| optimum-cli export openvino --model TinyLlama/TinyLlama-1.1B-Chat-v1.0 --trust-remote-code --weight-format int8 ov_models/TinyLlama-1.1B-Chat-v1.0/INT8 | |
| python ./tools/llm_bench/benchmark.py -m ./ov_models/TinyLlama-1.1B-Chat-v1.0/FP16/ --draft_model ./ov_models/TinyLlama-1.1B-Chat-v1.0/INT8/ -p "Why is the Sun yellow?" -d cpu --draft_device cpu -n 1 --genai --assistant_confidence_threshold 0.4 | |
| python ./tools/llm_bench/benchmark.py -m ./ov_models/TinyLlama-1.1B-Chat-v1.0/FP16/ --draft_model ./ov_models/TinyLlama-1.1B-Chat-v1.0/INT8/ -p "Why is the Sun yellow?" -d cpu --draft_device cpu -n 1 --genai --num_assistant_tokens 5 | |
| - name: Test whisper-tiny on Linux | |
| run: | | |
| GIT_LFS_SKIP_SMUDGE=1 git clone --depth 1 --branch main --single-branch https://huggingface.co/datasets/facebook/multilingual_librispeech | |
| cd multilingual_librispeech | |
| git lfs pull -I /data/mls_polish/train/audio/3283_1447_000.tar.gz | |
| mkdir data/mls_polish/train/audio/3283_1447_000 | |
| tar zxvf data/mls_polish/train/audio/3283_1447_000.tar.gz -C data/mls_polish/train/audio/3283_1447_000/ | |
| cd .. | |
| optimum-cli export openvino --trust-remote-code --model openai/whisper-tiny ./ov_models/whisper-tiny | |
| python ./tools/llm_bench/benchmark.py -m ./ov_models/whisper-tiny --media multilingual_librispeech/data/mls_polish/train/audio/3283_1447_000/3283_1447_000000.flac -d cpu -n 1 | |
| python ./tools/llm_bench/benchmark.py -m ./ov_models/whisper-tiny --media multilingual_librispeech/data/mls_polish/train/audio/3283_1447_000/3283_1447_000000.flac -d cpu -n 1 --genai | |
| - name: WWB Tests | |
| run: | | |
| GIT_CLONE_PROTECTION_ACTIVE=false pip install -r ${{ env.WWB_PATH }}/requirements.txt | |
| pip install git+https://github.com/huggingface/optimum-intel.git | |
| GIT_CLONE_PROTECTION_ACTIVE=false pip install ${{ env.WWB_PATH }} | |
| python -m pip install -U --pre openvino openvino-tokenizers openvino-genai --extra-index-url https://storage.openvinotoolkit.org/simple/wheels/nightly --force-reinstall | |
| python -m pytest -v tools/who_what_benchmark/tests | |
| stateful: | |
| runs-on: ubuntu-20.04 | |
| steps: | |
| - uses: actions/checkout@v4 | |
| - uses: actions/setup-python@v4 | |
| with: | |
| python-version: "3.10" | |
| - name: Test stateful | |
| run: | | |
| GIT_CLONE_PROTECTION_ACTIVE=false python -m pip install -r tools/llm_bench/requirements.txt | |
| python -m pip uninstall --yes openvino | |
| python -m pip install -U --pre openvino openvino-tokenizers openvino-genai --extra-index-url https://storage.openvinotoolkit.org/simple/wheels/nightly | |
| python tools/llm_bench/convert.py --model_id TinyLlama/TinyLlama-1.1B-Chat-v1.0 --output_dir . --stateful | |
| grep beam_idx pytorch/dldt/FP32/openvino_model.xml | |
| - name: WWB Tests | |
| run: | | |
| GIT_CLONE_PROTECTION_ACTIVE=false pip install -r tools/who_what_benchmark/requirements.txt | |
| pip install git+https://github.com/huggingface/optimum-intel.git | |
| GIT_CLONE_PROTECTION_ACTIVE=false pip install tools/who_what_benchmark/ | |
| pip install pytest | |
| python -m pip install -U --pre openvino openvino-tokenizers openvino-genai --extra-index-url https://storage.openvinotoolkit.org/simple/wheels/nightly --force-reinstall | |
| python -m pytest -v tools/who_what_benchmark/tests |