Article

SACQ: Structure-Aware Codebook Quantization for Ultra-Low-Bit Vision-Language-Action Models

SACQ is an ultra-low-bit weight quantization framework for Vision-Language-Action (VLA) models, combining structure-aware codebook fitting, action-aware mixed-bit allocation, and two-stage post-quantization refinement.

Part of the event: SACQ Ultra-Low-Bit VLA Quantization Framework Released

This page provides a summary and source link only.