LoRA Inspector

A comprehensive tool for inspecting LoRA (Low-Rank Adaptation) files for machine learning models, particularly Stable Diffusion.

See the different blocks of the different networks. Ideally uses the more common format but↵ still good to see how the weights are effectively. We use Frobenius norm for the magnitude↵ and a vector norm for the strength.

See the different settings for the LoRA file. What model it was trained on. Any VAE. Networ↵ k Dim/Rank and Alpha. Learning rates. Optimizer settings, learning rate schedulers.

Dataset with buckets. Bucket resolutions.

Subsets showing the different subset datasets, image augments, captions.

Features

• Browser-based LoRA file inspection
• Detailed metadata analysis
• Weight and network characteristic visualization
• WebAssembly-powered performance
• No external dependencies (like torch or python)

Project Structure

• crates/: Project sub-modules
  • inspector/: Core Rust library
  • lora-inspector-wasm/: WebAssembly web frontend
  • lora-inspector/: CLI application

Technologies

• Rust
• WebAssembly
• React
• Vite
• Candle (machine learning library)

Testing Strategy

Test Frameworks

• lora-inspector-wasm:
  • AVA (JavaScript unit tests)
  • Playwright (E2E testing)
• inspector: Rust built-in test module
• lora-inspector: Limited test coverage

Test Coverage

• Unit tests for key parsing
• E2E browser testing
• Metadata validation
• Cross-browser compatibility

Getting Started

Prerequisites

• Rust toolchain
• Node.js
• Yarn 4.9.1+
• wasm-pack

Installation

# Clone the repository
git clone https://github.com/rockerBOO/lora-inspector-rs.git

# Navigate to the project directory
cd lora-inspector-rs

# Install dependencies for web frontend
cd crates/lora-inspector-wasm
yarn install --immutable

Development

# Start development server
make dev-wasm

# Build the project
make build

# Run tests
make test

Crate-Specific Documentation

• LoRA Inspector WASM Frontend
• LoRA Inspector CLI

Contributing

Contributions are welcome! Please check individual crate READMEs for specific guidelines.

Current Roadmap

• WebAssembly frontend
• Add LR Warmup metadata support
• Improve weight visualization
• Expand metadata parsing capabilities
• Increase test coverage

License

MIT License - See LICENSE file

Acknowledgements

• Inspired by the need for better LoRA model inspection tools
• Built with passion for machine learning and open-source development