nunchaku-ai/nunchaku
SVDQuant-based inference engine enabling efficient 4-bit diffusion models for image generation.
Not currently ranked — collecting fresh signals.
star history
Nunchaku is a high-performance inference engine for 4-bit quantized neural networks as introduced in the SVDQuant paper (ICLR 2025 Spotlight). It absorbs outliers using low-rank components to enable aggressive quantization of diffusion models including Flux. The project provides ComfyUI integration and quantized model weights for practical deployment of highly compressed generative models.
Frequently asked
- What is nunchaku-ai/nunchaku?
- SVDQuant-based inference engine enabling efficient 4-bit diffusion models for image generation.
- Is nunchaku open source?
- Yes — nunchaku-ai/nunchaku is open source, released under the Apache-2.0 license.
- What language is nunchaku written in?
- nunchaku-ai/nunchaku is primarily written in Python.
- How popular is nunchaku?
- nunchaku-ai/nunchaku has 3.9k stars on GitHub.
- Where can I find nunchaku?
- nunchaku-ai/nunchaku is on GitHub at https://github.com/nunchaku-ai/nunchaku.