Understanding R1-Zero-Like Training: A Critical Perspective

Train transformer language models with reinforcement learning.

[ICLR 2026] SoFlow: Solution Flow Models for One-Step Generative Modeling

verl-agent is an extension of veRL, designed for training LLM/VLM agents via RL. verl-agent is also the official code for paper "Group-in-Group Policy Optimization for LLM Agent Training"

Implementation of Hindsight Differentiable Policy Optimization, as described in the paper Deep Reinforcement Learning for Inventory Networks: Toward Reliable Policy Optimization

Official codebase for the paper "How to Guide Your Flow: Few-Step Alignment via Flow Map Reward Guidance" (ICML 2026).

Parallel Token Prediction for Language Models (ICLR 2026)

Diffinity is a tool for constraining the output of continuous diffusion models to satisfy regular expressions. Companion artifact for ICML 2026 paper "Continous Diffusion Models can Obey Formal Syn…

Official Code Repo for Paper: Posterior Refinement

Official Implementation of MARS

Code for NeurIPS'24 paper 'Grokked Transformers are Implicit Reasoners: A Mechanistic Journey to the Edge of Generalization'

Code for the paper "Lessons from Studying Two-Hop Latent Reasoning"

A toy eval suite for tracing generalization dynamics of LM pre-training

Implementation of rewriting ensembles for the paper "What are the Right Symmetries for Formal Theorem Proving?"

Code of Training-free Detection of AI-generated images via Cropping Robustness

Official Pytorch Reimplementation of XFactor: True Self-Supervised Novel View Synthesis is Transferable (ICLR 2026, Oral)

Official implementation of "Infinite Mask Diffusion for Few-Step Distillation" (ICML 2026)

An LLM-agent framework that acts as a data scientist for relational learning.

Official implementation of Gumbel Distillation for Parallel Text Generation

Dreamer 4 jax implementation

A ~9M parameter LLM that talks like a small fish.