This code implements the algorithm of FIPO, a value-free RL recipe for eliciting deeper reasoning from a clean base model.

The official GitHub repo for the survey paper "A Survey on Diffusion Language Models".

slime is an LLM post-training framework for RL Scaling.

Official implementation of GDPO: Group reward-Decoupled Normalization Policy Optimization for Multi-reward RL Optimization

Build compute kernels and load them from the Hub.

Scalable toolkit for efficient model reinforcement

Evaluate and improve models and agents using environments

✨✨Latest Papers and Benchmarks in Reasoning with Foundation Models

[ICLR2026] Laser: Learn to Reason Efficiently with Adaptive Length-based Reward Shaping

[NeurIPS 2025] VeriThinker: Learning to Verify Makes Reasoning Model Efficient

[TMLR 2025] Efficient Reasoning Models: A Survey

Fully open data curation for reasoning models

Attribute statements generated by LLMs to preceding tokens using attention weights.

s1: Simple test-time scaling

Work in progress.

[ICLRW'26] EoRA: Fine-tuning-free Compensation for Compressed LLM with Eigenspace Low-Rank Approximation

Simple RL training for reasoning

Train transformer language models with reinforcement learning.

Fully open reproduction of DeepSeek-R1

YuE: Open Full-song Music Generation Foundation Model, something similar to Suno.ai but open

A suite of image and video neural tokenizers

LLM model quantization (compression) toolkit with HW acceleration support for Nvidia, AMD, Intel GPU and Intel/AMD/Apple CPU via HF, vLLM, and SGLang.

PyTorch native post-training library

Code for the ICLR 2023 paper "GPTQ: Accurate Post-training Quantization of Generative Pretrained Transformers".

Solutions to all questions of the book Introduction to the Theory of Computation, 3rd edition by Michael Sipser

Solve puzzles. Learn CUDA.