Skip to content

Hugging Face Guide

New Model Setup

Uploading a Model

Create a Model Repository

Go to Hugging Face and create a New Model under your profile, which essentially creates a repository for your models. Refer to Hugging Face Repositories Getting Started for detailed instructions.

Once created, you can clone the repository and proceed with the next steps on your local machine.

Set Up Git LFS (Large File Storage)

Hugging Face uses Git LFS to manage large files, such as model weights. Install and set up Git LFS as follows:

sudo apt install git-lfs
git lfs install

Configure Hugging Face Interface

Install the Hugging Face Hub library and set up authentication:

pip install huggingface_hub

# Configure Git to store credentials
git config --global credential.helper store

# Log in to Hugging Face CLI
huggingface-cli login

# Your token will be saved in the following location:
cat $HOME/.cache/huggingface/token

# Enable LFS support for large files
huggingface-cli lfs-enable-largefiles

Push a Large File

Once Git LFS is configured, you can push large files just like a normal Git commit:

git add your_model
git commit -m "Add a large file using Git LFS"
git push

For more details on uploading models, refer to the Hugging Face Model Uploading Guide.

Downloading a Model

To download a model from Hugging Face, you can use the following Python code:

from huggingface_hub import hf_hub_download

model_path = hf_hub_download(repo_id="fkariminejadasl/bird", filename="45_best.pth")

For more downloading options, refer to the Hugging Face Models Downloading Guide.

Creating a New Space on Hugging Face

To start, navigate to Hugging Face and create a New Space under your profile. This will generate a repository for your application. For detailed guidance, refer to the Gradio Spaces documentation.

Once your space is created, you can clone the repository to your local machine, add your app.py and requirements.txt files, and push the changes back to the repository. The application will launch automatically.

Application Code and Dependencies

  • app.py: Your Gradio application code must be saved in a file named app.py.

  • requirements.txt: This file should list all the Python dependencies your app requires, including any custom libraries. An example requirements.txt might look like this:

torch
gradio
huggingface_hub
git+https://github.com/username/repository_name.git@branch_name
  • Note: The branch_name can also be replaced by tag_name or commit_hash depending on your need.

If the pyproject.toml file for a custom library is located in a subdirectory within the repository, you can specify the subdirectory like this:

git+https://github.com/username/repository_name.git@branch_name#subdirectory=subdirectory_name

For further details on how to use pip install with various options, refer to the Pip Install Guide.