RAGEN
RAGEN leverages reinforcement learning to train LLM reasoning agents in interactive, stochastic environments.
Stars: 1054
RAGEN is a reinforcement learning framework designed to train reasoning-capable large language model (LLM) agents in interactive, stochastic environments. It addresses challenges such as multi-turn interactions and stochastic environments through a Markov Decision Process (MDP) formulation, Reason-Interaction Chain Optimization (RICO) algorithm, and progressive reward normalization strategies. The framework enables LLMs to reason and interact with the environment, optimizing entire trajectories for long-horizon reasoning while maintaining computational efficiency.
README:
RAGEN leverages reinforcement learning to train LLM reasoning agents in interactive, stochastic environments.
We strongly believe in the future of RL + LLM + Agents. The release is a minimally viable leap forward.
2025.3.8 Update:
- In previous veRL implementation, there is a KL term issue, which has been fixed in recent versions.
- We find evidence from multiple sources that PPO could be more stable than GRPO training in Open-Reasoner-Zero, TinyZero, and Zhihu. We have changed the default advantage estimator to GAE (using PPO) and aim to find more stable while efficient RL optimization methods in later versions.
Reinforcement Learning (RL) with rule-based rewards has shown promise in enhancing reasoning capabilities of large language models (LLMs). However, existing approaches have primarily focused on static, single-turn tasks like math reasoning and coding. Extending these methods to agent scenarios introduces two fundamental challenges:
- Multi-turn Interactions: Agents must perform sequential decision-making and react to environment feedback
- Stochastic Environments: Uncertainty where identical actions can lead to different outcomes
RAGEN addresses these challenges through:
- A Markov Decision Process (MDP) formulation for agent tasks
- Reason-Interaction Chain Optimization (RICO) algorithm that optimizes entire trajectory distributions
- Progressive reward normalization strategies to handle diverse, complex environments
RAGEN introduces a reinforcement learning framework to train reasoning-capable LLM agents that can operate in interactive, stochastic environments.
The Reasoning-Interaction Chain Optimization (RICO) framework with two interleaved stages: rollout stage and update stage. LLM iteratively generates reasoning-guided actions to interact with the environment to obtain trajectory-level rewards, normalized for LLM update to jointly optimize reasoning and action strategies.
RAGEN introduces a reinforcement learning framework to train reasoning-capable LLM agents that can operate in interactive, stochastic environments. The framework consists of two key components:
We formulate agent-environment interactions as Markov Decision Processes (MDPs) where states and actions are token sequences, allowing LLMs to reason over environment dynamics. At time t, state $s_t$ transitions to the next state through action $a_t$ following a transition function. The policy generates actions given the trajectory history. The objective is to maximize expected cumulative rewards across multiple interaction turns.
RICO enables LLMs to jointly optimize reasoning and action strategies over entire trajectories. The algorithm alternates between two phases:
Given an initial state, the LLM generates multiple trajectories. At each step, the model receives the trajectory history and generates a reasoning-guided action: <think>...</think><ans> action </ans>. The environment receives the action and returns feedback (reward and next state).
After generating trajectories, we train LLMs to optimize expected rewards. Instead of step-by-step optimization, RICO optimizes entire trajectories using importance sampling. This approach enables long-horizon reasoning while maintaining computational efficiency.
We implement three progressive normalization strategies to stabilize training:
- ARPO: Preserves raw rewards directly
- BRPO: Normalizes rewards across each training batch using batch statistics
- GRPO: Normalizes within prompt groups to balance learning across varying task difficulties
For detailed setup instructions, please check our documentation. Here's a quick start guide:
# Setup environment and download data (7MB)
bash scripts/setup_ragen.sh
python scripts/download_data.pyIf this fails, you can follow the manual setup instructions in scripts/setup_ragen.md.
Here's how to train models with RAGEN:
We provide 10k first-round-observation data for both Sokoban and FrozenLake tasks.
# Basic data creation
bash scripts/create_data.sh
# Or for research purposes, create more comprehensive data
bash scripts/create_data_full.shWe provide default configuration in verl/trainer/config/ppo_trainer.yaml. To train:
bash train.sh sokoban \
model.experiment_name=new_test
# Override config parameters as needed
bash train.sh sokoban \
model.experiment_name=new_test_debug \
training.train_batch_size=128 \
training.ppo_batch_size=64For supervised finetuning with LoRA:
- Create supervised finetuning data:
bash sft/generate_data.sh <env_type>- Finetune the model:
bash sft/finetune_lora.sh <env_type> <num_gpus> <save_path>- Merge LoRA weights with the base model:
python sft/utils/merge_lora.py \
--base_model_name <base_model_name> \
--lora_model_path <lora_model_path> \
--output_path <output_path>To visualize agent trajectories:
- Set visualization parameters in
train.sh:
logging.log_images=True
logging.log_image_dir=log/trajectory
logging.log_image_step_size=4
logging.log_n_image_per_batch=32- View the visualizations:
cd log/trajectory
python -m http.server 8000
# Access at http://localhost:8000/[EXP_NAME]/step_[STEP_NUM]/trajectory_data_[ID].html- For proper font rendering:
sudo apt-get install fonts-noto-cjk- Download visualization data from wandb:
from ragen.utils.wandb import download_wandb
download_wandb("RUN_ID") # e.g., 9o465jqjWe evaluate RAGEN across multiple model sizes and configurations. Below are results from our Sokoban experiments using Qwen-2.5-{0.5B, 3B}-{Instruct, None} and DeepSeek-R1-Distill-Qwen-1.5B.
NOTE: The loss shows the reward curve, where the KL term is considered.
Key observations:
- Instruct-finetuned models show early advantages but the gap narrows as training progresses
- Larger models (3B) generally outperform smaller models (0.5B), though the advantage is not dramatic
- The R1-distilled 1.5B model initially underperforms compared to 0.5B models
- Training has not yet converged in these experiments
Our analysis reveals two key aspects of LLM agent training with RL:
- Prompt diversity: Balancing observation variety and effective response comparison
- Online rollout frequency: Mediating between training stability and data recency
Visualization of agent reasoning on the Sokoban task:
The visualizations show how the agent reasons through sequential steps to solve the puzzle.
We provide several case studies showing the model's behavior:
More case studies will be added to showcase both successful reasoning patterns and failure modes.
We welcome all forms of feedback! Please raise an issue for bugs, questions, or suggestions. This helps our team address common problems efficiently and builds a more productive community.
Zihan Wang*, Kangrui Wang*, Qineng Wang*, Pingyue Zhang*, Linjie Li*, Zhengyuan Yang, Kefan Yu, Minh Nhat Nguyen, Monica Lam, Yiping Lu, Kyunghyun Cho, Jiajun Wu, Li Fei-Fei, Lijuan Wang, Yejin Choi, Manling Li
*:Equal Contribution.
We thank DeepSeek for providing the DeepSeek-R1 model and ideas. We thank the veRL team for their infrastructure. We thank the TinyZero team for their discoveries that inspired our early exploration. We thank Han Liu, Xinyu Xing, Li Erran Li, Akari Asai, Eiso Kant, Lu Lu, Runxin Xu, Huajian Xin, Zijun Liu, Weiyi Liu, Weimin Wu, Yibo Wen, Jiarui Liu, Lorenzo Xiao, Ishan Mukherjee, Anabella Isaro, Haosen Sun, How-Yeh Wan, Lester Xue, Weiyi Liu for insightful discussions.
@misc{RAGEN,
author = {Zihan Wang* and Kangrui Wang* and Qineng Wang* and Pingyue Zhang* and Linjie Li* and Zhengyuan Yang and Kefan Yu and Minh Nhat Nguyen and Monica Lam and Yiping Lu and Kyunghyun Cho and Jiajun Wu and Li Fei-Fei and Lijuan Wang and Yejin Choi and Manling Li},
title = {Training Agents by Reinforcing Reasoning},
year = {2025},
organization = {GitHub},
url = {https://github.com/ZihanWang314/ragen},
}For Tasks:
Click tags to check more tools for each tasksFor Jobs:
Alternative AI tools for RAGEN
Similar Open Source Tools
RAGEN
RAGEN is a reinforcement learning framework designed to train reasoning-capable large language model (LLM) agents in interactive, stochastic environments. It addresses challenges such as multi-turn interactions and stochastic environments through a Markov Decision Process (MDP) formulation, Reason-Interaction Chain Optimization (RICO) algorithm, and progressive reward normalization strategies. The framework enables LLMs to reason and interact with the environment, optimizing entire trajectories for long-horizon reasoning while maintaining computational efficiency.
RAGEN
RAGEN is a reinforcement learning framework designed to train reasoning-capable large language model (LLM) agents in interactive, stochastic environments. It addresses challenges such as multi-turn interactions and stochastic environments through a Markov Decision Process (MDP) formulation, Reason-Interaction Chain Optimization (RICO) algorithm, and progressive reward normalization strategies. The framework consists of MDP formulation, RICO algorithm with rollout and update stages, and reward normalization strategies to stabilize training. RAGEN aims to optimize reasoning and action strategies for LLM agents operating in complex environments.
motia
Motia is an AI agent framework designed for software engineers to create, test, and deploy production-ready AI agents quickly. It provides a code-first approach, allowing developers to write agent logic in familiar languages and visualize execution in real-time. With Motia, developers can focus on business logic rather than infrastructure, offering zero infrastructure headaches, multi-language support, composable steps, built-in observability, instant APIs, and full control over AI logic. Ideal for building sophisticated agents and intelligent automations, Motia's event-driven architecture and modular steps enable the creation of GenAI-powered workflows, decision-making systems, and data processing pipelines.
MARBLE
MARBLE (Multi-Agent Coordination Backbone with LLM Engine) is a modular framework for developing, testing, and evaluating multi-agent systems leveraging Large Language Models. It provides a structured environment for agents to interact in simulated environments, utilizing cognitive abilities and communication mechanisms for collaborative or competitive tasks. The framework features modular design, multi-agent support, LLM integration, shared memory, flexible environments, metrics and evaluation, industrial coding standards, and Docker support.
Mooncake
Mooncake is a serving platform for Kimi, a leading LLM service provided by Moonshot AI. It features a KVCache-centric disaggregated architecture that separates prefill and decoding clusters, leveraging underutilized CPU, DRAM, and SSD resources of the GPU cluster. Mooncake's scheduler balances throughput and latency-related SLOs, with a prediction-based early rejection policy for highly overloaded scenarios. It excels in long-context scenarios, achieving up to a 525% increase in throughput while handling 75% more requests under real workloads.
MathCoder
MathCoder is a repository focused on enhancing mathematical reasoning by fine-tuning open-source language models to use code for modeling and deriving math equations. It introduces MathCodeInstruct dataset with solutions interleaving natural language, code, and execution results. The repository provides MathCoder models capable of generating code-based solutions for challenging math problems, achieving state-of-the-art scores on MATH and GSM8K datasets. It offers tools for model deployment, inference, and evaluation, along with a citation for referencing the work.
APOLLO
APOLLO is a memory-efficient optimizer designed for large language model (LLM) pre-training and full-parameter fine-tuning. It offers SGD-like memory cost with AdamW-level performance. The optimizer integrates low-rank approximation and optimizer state redundancy reduction to achieve significant memory savings while maintaining or surpassing the performance of Adam(W). Key contributions include structured learning rate updates for LLM training, approximated channel-wise gradient scaling in a low-rank auxiliary space, and minimal-rank tensor-wise gradient scaling. APOLLO aims to optimize memory efficiency during training large language models.
eole
EOLE is an open language modeling toolkit based on PyTorch. It aims to provide a research-friendly approach with a comprehensive yet compact and modular codebase for experimenting with various types of language models. The toolkit includes features such as versatile training and inference, dynamic data transforms, comprehensive large language model support, advanced quantization, efficient finetuning, flexible inference, and tensor parallelism. EOLE is a work in progress with ongoing enhancements in configuration management, command line entry points, reproducible recipes, core API simplification, and plans for further simplification, refactoring, inference server development, additional recipes, documentation enhancement, test coverage improvement, logging enhancements, and broader model support.
sec-parser
The `sec-parser` project simplifies extracting meaningful information from SEC EDGAR HTML documents by organizing them into semantic elements and a tree structure. It helps in parsing SEC filings for financial and regulatory analysis, analytics and data science, AI and machine learning, causal AI, and large language models. The tool is especially beneficial for AI, ML, and LLM applications by streamlining data pre-processing and feature extraction.
AirSLAM
AirSLAM is an efficient visual SLAM system designed to tackle short-term and long-term illumination challenges. It combines deep learning techniques with traditional optimization methods, featuring a unified CNN for keypoint and structural line extraction. The system includes a relocalization pipeline for map reuse, accelerated using C++ and NVIDIA TensorRT. Outperforming other SLAM systems in challenging environments, it runs at 73Hz on PC and 40Hz on embedded platforms.
inngest
Inngest is a platform that offers durable functions to replace queues, state management, and scheduling for developers. It allows writing reliable step functions faster without dealing with infrastructure. Developers can create durable functions using various language SDKs, run a local development server, deploy functions to their infrastructure, sync functions with the Inngest Platform, and securely trigger functions via HTTPS. Inngest Functions support retrying, scheduling, and coordinating operations through triggers, flow control, and steps, enabling developers to build reliable workflows with robust support for various operations.
petals
Petals is a tool that allows users to run large language models at home in a BitTorrent-style manner. It enables fine-tuning and inference up to 10x faster than offloading. Users can generate text with distributed models like Llama 2, Falcon, and BLOOM, and fine-tune them for specific tasks directly from their desktop computer or Google Colab. Petals is a community-run system that relies on people sharing their GPUs to increase its capacity and offer a distributed network for hosting model layers.
mindnlp
MindNLP is an open-source NLP library based on MindSpore. It provides a platform for solving natural language processing tasks, containing many common approaches in NLP. It can help researchers and developers to construct and train models more conveniently and rapidly. Key features of MindNLP include: * Comprehensive data processing: Several classical NLP datasets are packaged into a friendly module for easy use, such as Multi30k, SQuAD, CoNLL, etc. * Friendly NLP model toolset: MindNLP provides various configurable components. It is friendly to customize models using MindNLP. * Easy-to-use engine: MindNLP simplified complicated training process in MindSpore. It supports Trainer and Evaluator interfaces to train and evaluate models easily. MindNLP supports a wide range of NLP tasks, including: * Language modeling * Machine translation * Question answering * Sentiment analysis * Sequence labeling * Summarization MindNLP also supports industry-leading Large Language Models (LLMs), including Llama, GLM, RWKV, etc. For support related to large language models, including pre-training, fine-tuning, and inference demo examples, you can find them in the "llm" directory. To install MindNLP, you can either install it from Pypi, download the daily build wheel, or install it from source. The installation instructions are provided in the documentation. MindNLP is released under the Apache 2.0 license. If you find this project useful in your research, please consider citing the following paper: @misc{mindnlp2022, title={{MindNLP}: a MindSpore NLP library}, author={MindNLP Contributors}, howpublished = {\url{https://github.com/mindlab-ai/mindnlp}}, year={2022} }
UFO
UFO is a UI-focused dual-agent framework to fulfill user requests on Windows OS by seamlessly navigating and operating within individual or spanning multiple applications.
postgresml
PostgresML is a powerful Postgres extension that seamlessly combines data storage and machine learning inference within your database. It enables running machine learning and AI operations directly within PostgreSQL, leveraging GPU acceleration for faster computations, integrating state-of-the-art large language models, providing built-in functions for text processing, enabling efficient similarity search, offering diverse ML algorithms, ensuring high performance, scalability, and security, supporting a wide range of NLP tasks, and seamlessly integrating with existing PostgreSQL tools and client libraries.
YuE
YuE (乐) is an open-source foundation model designed for music generation, specifically transforming lyrics into full songs. It can generate complete songs in various genres and vocal styles, ensuring a polished and cohesive result. The model requires significant GPU memory for generating long sequences and recommends specific configurations for optimal performance. Users can customize the number of sessions for memory usage. The tool provides a quickstart guide for generating music using Transformers and includes tips for execution time and tag selection. The project is licensed under Creative Commons Attribution Non Commercial 4.0.
For similar tasks
lighteval
LightEval is a lightweight LLM evaluation suite that Hugging Face has been using internally with the recently released LLM data processing library datatrove and LLM training library nanotron. We're releasing it with the community in the spirit of building in the open. Note that it is still very much early so don't expect 100% stability ^^' In case of problems or question, feel free to open an issue!
Firefly
Firefly is an open-source large model training project that supports pre-training, fine-tuning, and DPO of mainstream large models. It includes models like Llama3, Gemma, Qwen1.5, MiniCPM, Llama, InternLM, Baichuan, ChatGLM, Yi, Deepseek, Qwen, Orion, Ziya, Xverse, Mistral, Mixtral-8x7B, Zephyr, Vicuna, Bloom, etc. The project supports full-parameter training, LoRA, QLoRA efficient training, and various tasks such as pre-training, SFT, and DPO. Suitable for users with limited training resources, QLoRA is recommended for fine-tuning instructions. The project has achieved good results on the Open LLM Leaderboard with QLoRA training process validation. The latest version has significant updates and adaptations for different chat model templates.
Awesome-Text2SQL
Awesome Text2SQL is a curated repository containing tutorials and resources for Large Language Models, Text2SQL, Text2DSL, Text2API, Text2Vis, and more. It provides guidelines on converting natural language questions into structured SQL queries, with a focus on NL2SQL. The repository includes information on various models, datasets, evaluation metrics, fine-tuning methods, libraries, and practice projects related to Text2SQL. It serves as a comprehensive resource for individuals interested in working with Text2SQL and related technologies.
create-million-parameter-llm-from-scratch
The 'create-million-parameter-llm-from-scratch' repository provides a detailed guide on creating a Large Language Model (LLM) with 2.3 million parameters from scratch. The blog replicates the LLaMA approach, incorporating concepts like RMSNorm for pre-normalization, SwiGLU activation function, and Rotary Embeddings. The model is trained on a basic dataset to demonstrate the ease of creating a million-parameter LLM without the need for a high-end GPU.
StableToolBench
StableToolBench is a new benchmark developed to address the instability of Tool Learning benchmarks. It aims to balance stability and reality by introducing features such as a Virtual API System with caching and API simulators, a new set of solvable queries determined by LLMs, and a Stable Evaluation System using GPT-4. The Virtual API Server can be set up either by building from source or using a prebuilt Docker image. Users can test the server using provided scripts and evaluate models with Solvable Pass Rate and Solvable Win Rate metrics. The tool also includes model experiments results comparing different models' performance.
BetaML.jl
The Beta Machine Learning Toolkit is a package containing various algorithms and utilities for implementing machine learning workflows in multiple languages, including Julia, Python, and R. It offers a range of supervised and unsupervised models, data transformers, and assessment tools. The models are implemented entirely in Julia and are not wrappers for third-party models. Users can easily contribute new models or request implementations. The focus is on user-friendliness rather than computational efficiency, making it suitable for educational and research purposes.
AI-TOD
AI-TOD is a dataset for tiny object detection in aerial images, containing 700,621 object instances across 28,036 images. Objects in AI-TOD are smaller with a mean size of 12.8 pixels compared to other aerial image datasets. To use AI-TOD, download xView training set and AI-TOD_wo_xview, then generate the complete dataset using the provided synthesis tool. The dataset is publicly available for academic and research purposes under CC BY-NC-SA 4.0 license.
UMOE-Scaling-Unified-Multimodal-LLMs
Uni-MoE is a MoE-based unified multimodal model that can handle diverse modalities including audio, speech, image, text, and video. The project focuses on scaling Unified Multimodal LLMs with a Mixture of Experts framework. It offers enhanced functionality for training across multiple nodes and GPUs, as well as parallel processing at both the expert and modality levels. The model architecture involves three training stages: building connectors for multimodal understanding, developing modality-specific experts, and incorporating multiple trained experts into LLMs using the LoRA technique on mixed multimodal data. The tool provides instructions for installation, weights organization, inference, training, and evaluation on various datasets.
For similar jobs
weave
Weave is a toolkit for developing Generative AI applications, built by Weights & Biases. With Weave, you can log and debug language model inputs, outputs, and traces; build rigorous, apples-to-apples evaluations for language model use cases; and organize all the information generated across the LLM workflow, from experimentation to evaluations to production. Weave aims to bring rigor, best-practices, and composability to the inherently experimental process of developing Generative AI software, without introducing cognitive overhead.
LLMStack
LLMStack is a no-code platform for building generative AI agents, workflows, and chatbots. It allows users to connect their own data, internal tools, and GPT-powered models without any coding experience. LLMStack can be deployed to the cloud or on-premise and can be accessed via HTTP API or triggered from Slack or Discord.
VisionCraft
The VisionCraft API is a free API for using over 100 different AI models. From images to sound.
kaito
Kaito is an operator that automates the AI/ML inference model deployment in a Kubernetes cluster. It manages large model files using container images, avoids tuning deployment parameters to fit GPU hardware by providing preset configurations, auto-provisions GPU nodes based on model requirements, and hosts large model images in the public Microsoft Container Registry (MCR) if the license allows. Using Kaito, the workflow of onboarding large AI inference models in Kubernetes is largely simplified.
PyRIT
PyRIT is an open access automation framework designed to empower security professionals and ML engineers to red team foundation models and their applications. It automates AI Red Teaming tasks to allow operators to focus on more complicated and time-consuming tasks and can also identify security harms such as misuse (e.g., malware generation, jailbreaking), and privacy harms (e.g., identity theft). The goal is to allow researchers to have a baseline of how well their model and entire inference pipeline is doing against different harm categories and to be able to compare that baseline to future iterations of their model. This allows them to have empirical data on how well their model is doing today, and detect any degradation of performance based on future improvements.
tabby
Tabby is a self-hosted AI coding assistant, offering an open-source and on-premises alternative to GitHub Copilot. It boasts several key features: * Self-contained, with no need for a DBMS or cloud service. * OpenAPI interface, easy to integrate with existing infrastructure (e.g Cloud IDE). * Supports consumer-grade GPUs.
spear
SPEAR (Simulator for Photorealistic Embodied AI Research) is a powerful tool for training embodied agents. It features 300 unique virtual indoor environments with 2,566 unique rooms and 17,234 unique objects that can be manipulated individually. Each environment is designed by a professional artist and features detailed geometry, photorealistic materials, and a unique floor plan and object layout. SPEAR is implemented as Unreal Engine assets and provides an OpenAI Gym interface for interacting with the environments via Python.
Magick
Magick is a groundbreaking visual AIDE (Artificial Intelligence Development Environment) for no-code data pipelines and multimodal agents. Magick can connect to other services and comes with nodes and templates well-suited for intelligent agents, chatbots, complex reasoning systems and realistic characters.