Create HuggingFace Model Repository #17
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: Create HuggingFace Model Repository | |
| on: | |
| workflow_dispatch: | |
| inputs: | |
| model_name: | |
| description: "Name of the model to create (will be used in repo name and files)" | |
| required: true | |
| type: string | |
| prompt_template: | |
| description: "Prompt template for the model" | |
| required: true | |
| type: string | |
| default: "<|im_start|>system\n{system_message}<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant" | |
| stop_tokens: | |
| description: "Stop tokens for the model (comma-separated, e.g., <|im_end|>,</s>)" | |
| required: true | |
| type: string | |
| default: "<|im_end|>" | |
| engine: | |
| description: "Engine to run the model (e.g., llama-cpp)" | |
| required: true | |
| type: string | |
| default: "llama-cpp" | |
| env: | |
| USER_NAME: cortexso | |
| MODEL_NAME: ${{ inputs.model_name }} | |
| PROMPT_TEMPLATE: ${{ inputs.prompt_template }} | |
| STOP_TOKENS: ${{ inputs.stop_tokens }} | |
| ENGINE: ${{ inputs.engine }} | |
| jobs: | |
| create-repo: | |
| runs-on: ubuntu-20-04-gguf | |
| timeout-minutes: 7200 | |
| steps: | |
| - name: Checkout | |
| uses: actions/checkout@v4 | |
| - name: Set up Python | |
| uses: actions/setup-python@v5 | |
| with: | |
| python-version: '3.10' | |
| - name: Cache Python packages | |
| uses: actions/cache@v4 | |
| with: | |
| path: | | |
| ~/.cache/pip | |
| ~/.local/share/pip | |
| .venv | |
| key: ${{ runner.os }}-pip-${{ github.sha }} | |
| restore-keys: | | |
| ${{ runner.os }}-pip- | |
| - name: Install dependencies | |
| run: | | |
| pip install huggingface_hub PyYAML | |
| git lfs install | |
| - name: Create YAML files | |
| run: | | |
| python3 - << EOF | |
| import yaml | |
| import os | |
| from yaml.representer import SafeRepresenter | |
| class CustomDumper(yaml.SafeDumper): | |
| pass | |
| def custom_str_representer(dumper, data): | |
| if isinstance(data, str): | |
| return dumper.represent_scalar('tag:yaml.org,2002:str', data, style='') | |
| return SafeRepresenter.represent_str(dumper, data) | |
| CustomDumper.add_representer(str, custom_str_representer) | |
| def dict_representer(dumper, data): | |
| return dumper.represent_dict(data.items()) | |
| CustomDumper.add_representer(dict, dict_representer) | |
| # Process stop tokens | |
| stop_tokens = os.environ['STOP_TOKENS'].split(',') | |
| stop_tokens = [token.strip() for token in stop_tokens if token.strip()] | |
| # Create model.yml content with sections | |
| def write_section(file, content, section_name='', subsection=''): | |
| if section_name: | |
| file.write(f"# BEGIN {section_name}\n") | |
| if subsection: | |
| file.write(f"# BEGIN {subsection}\n") | |
| # Special handling for prompt_template to ensure it's a single line with preserved \n | |
| if 'prompt_template' in content: | |
| # Create a copy of the content to modify | |
| modified_content = content.copy() | |
| # Convert the prompt template to a literal string that preserves \n | |
| prompt_template = modified_content['prompt_template'] | |
| # Remove the prompt_template from the content to handle it separately | |
| del modified_content['prompt_template'] | |
| # Dump the rest of the content | |
| yaml_content = yaml.dump(modified_content, Dumper=CustomDumper, default_flow_style=False, sort_keys=False) | |
| file.write(yaml_content) | |
| # Write the prompt_template separately with proper escaping | |
| file.write(f'prompt_template: "{prompt_template}"\n') | |
| else: | |
| yaml_content = yaml.dump(content, Dumper=CustomDumper, default_flow_style=False, sort_keys=False) | |
| file.write(yaml_content) | |
| if subsection: | |
| file.write(f"# END {subsection}\n") | |
| if section_name: | |
| file.write(f"# END {section_name}\n") | |
| with open('model.yml', 'w') as f: | |
| # General metadata section | |
| general_metadata = { | |
| 'id': os.environ['MODEL_NAME'], | |
| 'model': os.environ['MODEL_NAME'], | |
| 'name': os.environ['MODEL_NAME'], | |
| 'version': 1 | |
| } | |
| write_section(f, general_metadata, 'GENERAL GGUF METADATA') | |
| f.write('\n') | |
| # Inference parameters section | |
| f.write("# BEGIN INFERENCE PARAMETERS\n") | |
| # Required subsection | |
| required_inference = {'stop': stop_tokens} | |
| write_section(f, required_inference, '', 'REQUIRED') | |
| f.write('\n') | |
| # Optional subsection | |
| optional_inference = { | |
| 'stream': True, | |
| 'top_p': 0.9, | |
| 'temperature': 0.7, | |
| 'frequency_penalty': 0, | |
| 'presence_penalty': 0, | |
| 'max_tokens': 4096, | |
| 'seed': -1, | |
| 'dynatemp_range': 0, | |
| 'dynatemp_exponent': 1, | |
| 'top_k': 40, | |
| 'min_p': 0.05, | |
| 'tfs_z': 1, | |
| 'typ_p': 1, | |
| 'repeat_last_n': 64, | |
| 'repeat_penalty': 1, | |
| 'mirostat': False, | |
| 'mirostat_tau': 5, | |
| 'mirostat_eta': 0.100000001, | |
| 'penalize_nl': False, | |
| 'ignore_eos': False, | |
| 'n_probs': 0, | |
| 'min_keep': 0 | |
| } | |
| write_section(f, optional_inference, '', 'OPTIONAL') | |
| f.write("# END INFERENCE PARAMETERS\n\n") | |
| # Model load parameters section | |
| f.write("# BEGIN MODEL LOAD PARAMETERS\n") | |
| required_load = { | |
| 'engine': os.environ['ENGINE'], | |
| 'prompt_template': os.environ['PROMPT_TEMPLATE'], | |
| 'ctx_len': 4096, | |
| 'ngl': 34 | |
| } | |
| write_section(f, required_load, '', 'REQUIRED') | |
| f.write("# END MODEL LOAD PARAMETERS\n") | |
| # Create metadata.yml | |
| metadata_content = { | |
| 'version': 1, | |
| 'name': os.environ['MODEL_NAME'], | |
| 'default': '8b-gguf-q4-km' | |
| } | |
| with open('metadata.yml', 'w') as f: | |
| write_section(f, metadata_content) | |
| EOF | |
| - name: Create HuggingFace Repository and Upload Files | |
| env: | |
| HF_TOKEN: ${{ secrets.HUGGINGFACE_TOKEN_WRITE }} | |
| run: | | |
| python3 - << EOF | |
| from huggingface_hub import HfApi, create_repo | |
| import os | |
| # Initialize the Hugging Face API | |
| api = HfApi(token=os.environ['HF_TOKEN']) | |
| # Create the repository | |
| repo_id = f"${{ env.USER_NAME }}/${{ env.MODEL_NAME }}" | |
| try: | |
| create_repo(repo_id, private=False, token=os.environ['HF_TOKEN']) | |
| print(f"Created repository: {repo_id}") | |
| except Exception as e: | |
| print(f"Repository might already exist or error occurred: {e}") | |
| # Upload the files | |
| api.upload_file( | |
| path_or_fileobj="model.yml", | |
| path_in_repo="model.yml", | |
| repo_id=repo_id, | |
| token=os.environ['HF_TOKEN'] | |
| ) | |
| api.upload_file( | |
| path_or_fileobj="metadata.yml", | |
| path_in_repo="metadata.yml", | |
| repo_id=repo_id, | |
| token=os.environ['HF_TOKEN'] | |
| ) | |
| print("Files uploaded successfully") | |
| EOF | |
| - name: Cleanup | |
| run: | | |
| rm -f model.yml metadata.yml |