Institution

Xiaohongshu

The Chinese lifestyle and social commerce platform, whose research team works on LLM reasoning and reinforcement learning.

Language Models · Xiaohongshu

NITP: Predict the Next Token's Meaning, Not Just Its ID

NITP adds a dense target to next-token prediction: forecast a shallow-layer embedding of the next token. On a 9B MoE it lifts MMLU-Pro by 5.71 points for about 2 percent extra training FLOPs and zero inference cost.

LLM Reasoning · Xiaohongshu

Anti-Self-Distillation (AntiSD): A PMI Reward That Speeds Up Reasoning RL

AntiSD inverts self-distillation — it rewards tokens where a privileged context disagrees with the base model — reaching GRPO's accuracy in 2 to 10x fewer steps and ending up to 11.5 points higher on 4B-30B models.