THUKEG

INFTY Engine: An Optimization Toolkit to Support Continual AI

Scaling Agentic Reinforcement Learning with a Multi-Turn, Multi-Task Framework

DeepDive: Advancing Deep Search Agents with Knowledge Graphs and Multi-Turn RL

IndexCache: Accelerating Sparse Attention via Cross-Layer Index Reuse

Finetuning and inference tools for the CogView4 and CogVideoX model series.

TreeRL: LLM Reinforcement Learning with On-Policy Tree Search in ACL'25

This repository contains the code and data for the paper "Chaining the Evidence: Robust Reinforcement Learning for Deep Search Agents with Citation-Aware Rubric Rewards".

[ACL25' Findings] SWE-Dev is an SWE agent with a scalable test case construction pipeline.

Open-source framework for computer use agents: VeriGen verifiable task synthesis, online RL training (AgentRL), and OSWorld/ScienceBoard evaluation.

Reinforcing LLM Reasoning through Self-Training and Value-Guided Decoding

Typescript SDK for Z.ai - Not yet released.

[Findings of ACL 2026] CodeRM-NT: Reward Model for Code RL without Unit Tests