yingfan-bot/lotus

Code for COLM '26 paper 'Loop, Think, & Generalize: Implicit Reasoning in Recurrent-Depth Transformers'

Code for Base Models Can Reason by Taking a Cue From Training Data

[NeurIPS 2025 Spotlight] LLM post-training suite — featuring ReasonFlux, ReasonFlux-PRM, and ReasonFlux-Coder.

The Truth Is In There: Improving Reasoning in Language Models with Layer-Selective Rank Reduction

Official implementation of Latent-SFT: teaching LLMs to reason with vocabulary-space latent chains.

Implement a reasoning LLM in PyTorch from scratch, step by step