Official repository of 'Visual-RFT: Visual Reinforcement Fine-Tuning' & 'Visual-ARFT: Visual Agentic Reinforcement Fine-Tuning'’

AgentCPM-GUI: An on-device GUI agent for operating Android apps, enhancing reasoning ability with reinforcement fine-tuning for efficient task execution.

Fine-tune LLM agents with online reinforcement learning

Official Repo for Fine-Tuning Large Vision-Language Models as Decision-Making Agents via Reinforcement Learning

Attention-based Deep Reinforcement Learning framework for portfolio allocation on S&P 500 equities. Includes custom environment, policy architecture with cross-sectional attention, PPO/A2C/REINFORCE agents, training/evaluation pipeline, and fine-tuning grid search.

PantheonRL is a package for training and testing multi-agent reinforcement learning environments. PantheonRL supports cross-play, fine-tuning, ad-hoc coordination, and more.

A trading agent AI is an artificial intelligence system that uses computational intelligence methods such as machine learning and deep reinforcement learning to automatically discover, implement, and fine-tune strategies for autonomous adaptive automated trading in financial markets

Eval Protocol (EP) is an open solution for doing reinforcement learning fine-tuning on existing agents — across any language, container, or framework.

No description available

This repository contains an AI agent for playing Tetris using the Deep Q-Learning (DQL) algorithm with fine-tuned rewards. It focuses on optimizing decision-making and performance through reinforcement learning techniques.

🧪 Advanced LLM fine-tuning framework with Reinforcement Learning (GRPO/DPO), Multi-Agent Swarm Training, Adaptive Optimization, and Unsloth integration (2x faster, 70% less VRAM). Train 1.5B-70B+ models on 8GB+ GPUs with QLoRA, PEFT, LUFFY off-policy reasoning, and Search-R1. RAG-enabled with smart chunking.

RLFTSim: Realistic and Controllable Multi-Agent Traffic Simulation via Reinforcement Learning Fine-Tuning [Under Review]]

official repo for SEEA-R1: Tree-Structured Reinforcement Fine-Tuning for Self-Evolving Embodied Agents

Distilling Multi-Agent Social Reasoning: Compressing MetaMind's Cognitive Architecture via Reinforcement Learning and Supervised Fine-Tuning

Benchmark and improve VLMs capabilities in complex, evolving markets as sequential traders - LLM post training, Reinforcement Learning Fine Tuning and Agentic AI

Led the R&D of the first closed-loop intelligent agent framework for visual content generation that integrates "underlying model capability perception" with "high-level reinforcement fine-tuning," significantly boosting complex semantic alignment and execution success rates.

Reinforcement Learning Agents to fine tune telecom networks.

Workshop on Agentic finetuning for AOAI models

Official Tuning Visual Fine Reinforcement Fine Reinforcement of ARFT Tuning Visual repository Agentic Visual Visual RFT

healthcare claim adjudication reinforcement fine-tuning agent

Fine-Tuning AI Agents with Reinforcement Learning

We present Query-MARFT, a query-guided multi-agent reinforcement fine-tuning framework.

official repo for SEEA-R1: Tree-Structured Reinforcement Fine-Tuning for Self-Evolving Embodied Agents

Unlocking the Power of Generative AI: In-Context Learning, Instruction Fine-Tuning, Reinforcement Learning Fine-Tuning, Retrieval Augmented Generation and LangGraph Workflows for AI Agents.

Core Idea Train a reinforcement learning agent that improves reasoning prompts for an LLM. Instead of fine-tuning the LLM directly, agent learns to optimize the reasoning process.

Multi-agent debate meets reinforcement learning through game theory. Four specialized LLM agents (Researcher→Reasoning→Critic→Refiner) debate math problems; Nash equilibrium convergence and debate quality metrics provide rich reward signals for GRPO fine-tuning. Achieves +30% accuracy over baseline on GSM8K with Qwen2.5-3B-Instruct and LoRA.

This project focuses on creating a realistic, enterprise-scale seed dataset simulating how a large B2B SaaS organization uses Asana for project management. The generated dataset is designed to serve as seed data for a reinforcement learning (RL) environment, enabling evaluation and fine-tuning of AI agents.

This project demonstrates how Agentic AI systems can be aligned using Reinforcement Learning from Human Feedback (RLHF) for aerospace engineering tasks. We build a browser-based aerospace dataset, create human preference pairs, and fine-tune a small language model using Direct Preference Optimization (DPO) to improve domain alignment.