Theory-optimal Quantization Based on Flatness

In three linesNew post-training quantization method for LLMs called Bidirectional Diagonal Quantization (BDQ). Introduces Flatness metric to quantify activation outlier distribution. BDQ achieves <1% accuracy drop in W4A4 on LLaMA-3-8B and reduces performance gap by 39.1% in W2A4KV16 on DeepSeek-R1-Distill-LLaMA-70B.

Summary generated by Claude — human-verified