hf-model-tool 0.2.4

A CLI tool to manage Hugging Face models.

HF-MODEL-TOOL

A CLI tool for managing your locally downloaded Huggingface models and datasets

Disclaimer: This tool is not affiliated with or endorsed by Hugging Face. It is an independent, community-developed utility.

Screenshots

Welcome Screen

List All Assets

Features

Core Functionality

• Smart Asset Detection: Detect HuggingFace models, datasets, LoRA adapters, fine-tuned models, and custom formats
• Asset Listing: View all your AI assets with size information and metadata
• Duplicate Detection: Find and clean duplicate downloads to save disk space
• Asset Details: View model configurations and dataset documentation with rich formatting
• Directory Management: Add and manage custom directories containing your AI assets
• Manifest System: Customize model names, publishers, and metadata with JSON manifests

Supported Asset Types

• HuggingFace Models & Datasets: Standard cached downloads from Hugging Face Hub
• LoRA Adapters: Fine-tuned adapters from training frameworks like Unsloth
• Custom Models: Fine-tuned models, merged models, and other custom formats

Installation

From PyPI (Recommended)

pip install hf-model-tool

From Source

git clone https://github.com/Chen-zexi/hf-model-tool.git
cd hf-model-tool
pip install -e .

Usage

Interactive Mode

hf-model-tool

Launches the interactive CLI with:

• System status showing assets across all configured directories
• Asset management tools for all supported formats
• Easy directory configuration and management

Integrating in vLLM-CLI

The tool provides API specifically designed for vLLM-CLI for model discovery and management.

Also can be launched directly from vLLM-CLI

Python API Usage

from hf_model_tool import get_downloaded_models
from hf_model_tool.api import HFModelAPI

# Quick access to models
models = get_downloaded_models()

# Full API access
api = HFModelAPI()
api.add_directory("/path/to/models", "custom")
assets = api.list_assets()

See API Reference for complete documentation.

Command Line Usage

The tool provides comprehensive command-line options for direct operations:

Basic Commands

# Launch interactive mode
hf-model-tool

# List all detected assets
hf-model-tool -l
hf-model-tool --list

# Enter asset management mode
hf-model-tool -m
hf-model-tool --manage

# View detailed asset information
hf-model-tool -v
hf-model-tool --view
hf-model-tool --details

# Show version
hf-model-tool --version

# Show help
hf-model-tool -h
hf-model-tool --help

Directory Management

# Add a directory containing LoRA adapters
hf-model-tool -path ~/my-lora-models
hf-model-tool --add-path ~/my-lora-models

# Add a custom model directory
hf-model-tool -path /data/custom-models

# Add current working directory
hf-model-tool -path .

# Add with absolute path
hf-model-tool -path /home/user/ai-projects/models

Sorting Options

# List assets sorted by size (default)
hf-model-tool -l --sort size

# List assets sorted by name
hf-model-tool -l --sort name

# List assets sorted by date
hf-model-tool -l --sort date

Interactive Navigation

• ↑/↓ arrows: Navigate menu options
• Enter: Select current option
• Back: Select to return to previous menu
• Config: Select to access settings and directory management
• Main Menu: Select to return to main menu from anywhere
• Exit: Select to clean application shutdown
• Ctrl+C: Force exit

Key Workflows

1. Directory Setup: Add directories containing your AI assets (HuggingFace cache, LoRA adapters, custom models)
2. List Assets: View all detected assets with size information across all directories
3. Manage Assets: Delete unwanted files and deduplicate identical assets
4. View Details: Inspect model configurations and dataset documentation
5. Configuration: Manage directories, change sorting preferences, and access help

Documentation

Quick Links

• Manifest System Guide - Learn how to customize model metadata with JSON manifests
• Custom Directories Guide - Configure and manage custom model directories
• API Reference - Complete API documentation for programmatic usage

Configuration

Directory Management

Add custom directories containing your AI assets:

• HuggingFace Cache: Standard HF cache with models--publisher--name structure
• Custom Directory: LoRA adapters, fine-tuned models, or other custom formats
• Auto-detect: Let the tool automatically determine the directory type

Interactive Configuration

Access via "Config" from any screen:

• Directory Management: Add, remove, and test directories
• Sort Options: Size (default), Date, or Name
• Help System: Navigation and usage guide

Manifest System

Automatic Generation: When you add a custom directory, the tool automatically generates a models_manifest.json file that:

• Becomes the primary source for model information
• Is always read first for classification
• Can be edited to ensure accurate display in vLLM-CLI

Customize model metadata using JSON manifests:

• Define custom names for your models
• Specify publishers and organizations
• Add notes and documentation
• See Manifest System Guide for details

Important: Review and edit auto-generated manifests to ensure model names and publishers are accurate for your use case.

Project Structure

hf_model_tool/
├── __main__.py       # Application entry point with welcome screen
├── cache.py          # Multi-directory asset scanning
├── ui.py             # Rich terminal interface components
├── utils.py          # Asset grouping and duplicate detection
├── navigation.py     # Menu navigation
├── config.py         # Configuration and directory management
└── asset_detector.py # Asset detection (LoRA, custom models, etc.)

Development

Requirements

• Python ≥ 3.7
• Dependencies: rich, inquirer, html2text

Logging

Application logs are written to ~/.hf-model-tool.log for debugging and monitoring.

Configuration Storage

Settings and directory configurations are stored in ~/.config/hf-model-tool/config.json

Contributing

We welcome contributions from the community! Please feel free to:

1. Open an issue at GitHub Issues
2. Submit a pull request with your improvements
3. Share feedback about your experience using the tool

License

This project is licensed under the MIT License - see the LICENSE file for details.