Skip to content
Hugging Face Logo

Hugging Face Provider

Hugging Face provides access to thousands of open-source models that can run locally or via their inference API.

Setup

Prerequisites

Hugging Face models require additional dependencies:

# Install as PRE_COMMAND in .mega-linter.yml
pip install langchain-huggingface transformers torch

Configuration

  1. Optional: Get API Token (for private models or hosted inference):

  2. Set Environment Variable (optional):

    bash export HUGGINGFACE_API_TOKEN=hf_your-token

  3. Configure MegaLinter:

    yaml LLM_ADVISOR_ENABLED: true LLM_PROVIDER: huggingface LLM_MODEL_NAME: microsoft/DialoGPT-medium LLM_MAX_TOKENS: 1000 LLM_TEMPERATURE: 0.1

Official Model List

For the most up-to-date list of Hugging Face models and their capabilities, see the official Hugging Face model hub:

Configuration Options

Basic Configuration

LLM_PROVIDER: huggingface
LLM_MODEL_NAME: microsoft/DialoGPT-medium

Advanced Configuration

# Model task type
HUGGINGFACE_TASK: text-generation

# Device settings
HUGGINGFACE_DEVICE: -1              # -1 for CPU, 0+ for GPU

# For hosted inference API
HUGGINGFACE_USE_API: true

Troubleshooting

Common Issues

  1. "Model not found"

    • Verify model name and repository exists
    • Check if model requires authentication
    • Ensure model supports the specified task

  2. "Out of memory"

    • Use smaller models
    • Enable CPU-only mode: HUGGINGFACE_DEVICE: -1
    • Close other applications

  3. "Import errors"

    • Install required dependencies: bash pip install langchain-huggingface transformers torch

  4. "Slow inference"

    • Use GPU if available: HUGGINGFACE_DEVICE: 0
    • Consider smaller models
    • Use hosted API for large models

Debug Mode

LOG_LEVEL: DEBUG
HUGGINGFACE_VERBOSE: true