feat: CI - add model converter #1
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: llama.cpp GGUF | |
| on: | |
| workflow_dispatch: | |
| inputs: | |
| hf_original_model_id: | |
| description: "HuggingFace original model ID" | |
| required: true | |
| type: string | |
| model_size: | |
| description: "The model size" | |
| required: true | |
| type: string | |
| hf_target_model_id: | |
| description: "HuggingFace target model ID" | |
| required: true | |
| type: string | |
| push: | |
| branches: | |
| - feat/model_converter_ci | |
| env: | |
| USER_NAME: cortexhub | |
| MODEL_ID: meta-llama/Meta-Llama-3-8B-Instruct # ${{ inputs.hf_model_id }} | |
| MODEL_SIZE: 8B #${{ inputs.model_size }} | |
| TARGET_MODEL_ID: llama3_test #${{ inputs.hf_target_model_id }} | |
| QUANTIZATION_METHOD: "Q4_K_M" | |
| HF_CACHE_DIR_LINUX: | |
| jobs: | |
| converter: | |
| runs-on: linux-cpu | |
| steps: | |
| - name: Clone | |
| id: checkout | |
| uses: actions/checkout@v3 | |
| with: | |
| submodules: recursive | |
| - uses: actions/setup-python@v4 | |
| with: | |
| python-version: "3.9" | |
| - name: Install python dependencies | |
| shell: bash | |
| run: | | |
| pip install huggingface_hub hf-transfer fire | |
| - name: Misc. env vars | |
| shell: bash | |
| run: | | |
| echo "Model ID: ${{ env.MODEL_ID }}" | |
| echo "Quantization: ${{ env.QUANTIZATION_METHOD }}" | |
| MODEL_ID=${{ env.MODEL_ID }} | |
| IFS='/' read -ra ADDR <<< "$MODEL_ID" | |
| ADDR_LENGTH=${#ADDR[@]} | |
| MODEL_NAME="${ADDR[$ADDR_LENGTH-1]}" | |
| if [[ "$MODEL_NAME" != "${MODEL_NAME,,}" ]]; then | |
| lowercase_model_name=$(echo "$MODEL_NAME" | tr '[:upper:]' '[:lower:]') | |
| else | |
| lowercase_model_name="$MODEL_NAME" | |
| fi | |
| echo "MODEL_NAME=$lowercase_model_name" >> $GITHUB_ENV | |
| bf16="${lowercase_model_name}.bf16.bin" | |
| echo "bf16=$bf16" >> $GITHUB_ENV | |
| - name: Install llama.cpp dependencies | |
| shell: bash | |
| run: | | |
| cd llama.cpp/ | |
| make clean | |
| make | |
| pip install -r requirements.txt | |
| - name: Download HF model | |
| shell: bash | |
| run: | | |
| huggingface-cli login --token ${{ secrets.HUGGINGFACE_TOKEN_READ }} --add-to-git-credential | |
| HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download --repo-type model --local-dir ./models ${{ env.MODEL_ID }} | |
| huggingface-cli logout | |
| - name: Convert to bf16 | |
| shell: bash | |
| run: | | |
| mkdir "${{ env.MODEL_NAME }}" | |
| python llama.cpp/convert-hf-to-gguf.py models --outtype bf16 --outfile "${{ env.MODEL_NAME }}/${{ env.bf16 }}" | |
| - name: GGUF quantization and push | |
| shell: bash | |
| run: | | |
| huggingface-cli login --token ${{ secrets.HUGGINGFACE_TOKEN_WRITE }} --add-to-git-credential | |
| ./llama.cpp/quantize "${{ env.MODEL_NAME }}/${{ env.bf16 }}" "${{ env.MODEL_NAME }}/model.gguf" "${{ env.QUANTIZATION_METHOD }}" | |
| rm "${{ env.MODEL_NAME }}/${{ env.bf16 }}" | |
| huggingface-cli upload ${{ env.USER_NAME }}/${{ env.TARGET_MODEL_ID }} ${{ env.MODEL_NAME }}/model.gguf model.gguf --revision "${{ env.MODEL_SIZE }}-gguf" | |
| huggingface-cli logout | |
| # - name: Generate Model metadata | |
| # shell: bash | |
| # run: | | |
| # cp ./models/README.md ./${{ env.MODEL_NAME }}/ | |
| # python modelCardGen.py --modelId=${{ env.MODEL_ID }} | |
| - name: Cleanup | |
| if: always() | |
| shell: bash | |
| run: | | |
| rm -rf ./models | |
| rm -rf "${{ env.MODEL_NAME }}" | |
| # rm -rf "/home/${USER}/.cache/huggingface/" |