AnalysisAI ModelsSeptember 15, 2026

Nunchux AI releases VC-Attention low-bit kernel for video DiTs

VC-Attention is a training-free low-bit attention kernel for video Diffusion Transformers that targets value quantization error and a slow softmax stage. It addresses accuracy loss from outliers, where a block's quantization scale is set by its largest value.

2 sources

More stories today

Open the live feed