FAIR-Calib: Frontier-Aware Instability-Reweighted Calibration for Post-Training Quantization of Diffusion Large Language Models

In three linesFAIR-Calib introduces a post-training quantization (PTQ) method for diffusion large language models (dLLMs). The two-stage framework protects fragile frontier decisions by reweighting unstable hidden states, avoiding expensive diffusion rollouts. Results on LLaDA and Dream (W4A4) show reduced quantization errors and decision flips.

Summary generated by Claude — human-verified