HuggingFace Blog · 1d ago · 8 · benchmark agent reinforcement learning open source

EcomRLVE-Gym is a new open-source benchmark providing 400 multi-turn, tool-augmented environments for training agentic LLMs with reinforcement learning and verifiable rewards. The framework addresses a critical gap in deploying LLMs as shopping assistants by enabling algorithmic reward verification (no LLM-as-judge) across e-commerce tasks like constrained product search, cart management, and multi-turn conversations.