AgenticArXiv-RL
What it is
A reinforcement learning pipeline for training ReAct agents to perform arXiv paper retrieval tasks. It implements a staged training approach (SFT, DPO, GRPO) using a deterministic snapshot-replay environment and rule-based verifiable rewards.
The problem it solves
Training LLM agents for complex tool-use tasks typically requires expensive human annotation for preference data or complex reward models; this project solves that by using rule-based, verifiable rewards and a deterministic offline environment.
What you could use it for
- Training LLMs to autonomously search and download academic papers
- Researching agentic reinforcement learning with verifiable rewards
- Developing tool-use agents for scientific literature workflows
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.