Research
ProAS proposes progress-aware advantage shaping for multi-turn agentic reinforcement learning
ProAS uses an Agentic Reference Graph to assign finer credit to actions in LLM agent training. Reported peak success rates reach 91.9% and 94.3% on ALFWorld and 66.7% and 70.8% on WebShop for Qwen2.5 1.5B and 7B, with convergence up to 2.44 times faster.
Every outlet that ran it 1 article
Hugging Face PapersResearchLead
ProAS: A Progress-Aware Advantage Shaping Framework for Multi-Turn Agentic Reinforcement Learning