Bringing Nunchaku 4-bit Diffusion Inference to Diffusers
Nunchaku is a quantization engine for diffusion transformers that reduces memory usage and improves denoising latency by quantizing both weights and activations, supporting architecture-agnostic quantization. It enables efficient execution on consumer hardware and provides a streamlined workflow for model developers, with pre-quantized checkpoints and tools like `diffuse-compressor` to facilitate quantizing new architectures. The engine is now natively supported in Diffusers, allowing developers to quantize and publish models, with detailed documentation available in integration PRs and the SVDQuant paper.
More from Huggingface
LFM2.5-VL-3B for Better and Faster Vision Capabilities for the Edge
Thinking of ACE? We Can Do It with Fewer Tokens
Meta is back with Muse Glimmer: local, agentic, multimodal, and open source
