DoubtLess: Training-Free Contrastive Decoding Shortens Reasoning Chains Without Losing Accuracy

DoubtLess, a training-free contrastive decoding method from CUHK and EPFL, uses a small model to read a large model's chain-of-thought twice and contrast the outputs, shortening Qwen3-32B's reasoning chains by 31% with no loss in accuracy.

2026-10-14 ~ 2026-10-15 · 2 related posts

• DoubtLess cuts Qwen3-32B chain-of-thought by 31% training-free with no accuracy loss — Minghao Luo · 2026-10-14
• Training-free contrastive decoding shortens reasoning chains without retraining — lulzxdxdxd · 2026-10-15