Search Results for "art"

Agent Reinforcement Trainer, or ART is an open-source reinforcement learning framework tailored to training large language model agents through experience, making them more reliable and performant on multi-turn, multi-step tasks. Instead of just manually crafting prompts or relying on supervised fine-tuning, ART uses techniques like Group Relative Policy Optimization (GRPO) to let agents learn from environmental feedback and reward signals.

...GPT-4) into software engineering agents that can resolve issues in real GitHub repositories. On the SWE-bench, the SWE-agent resolves 12.47% of issues, achieving state-of-the-art performance on the full test set. We accomplish our results by designing simple LM-centric commands and feedback formats to make it easier for the LM to browse the repository, and view, edit, and execute code files. We call this an Agent-Computer Interface (ACI).

...Built to operate graphical user interfaces like a human, it allows AI agents to perceive screens, reason about tasks, and execute actions across macOS, Windows, and Linux systems. The latest version, Agent S3, surpasses human-level performance on the OSWorld benchmark, demonstrating state-of-the-art results in complex multi-step computer tasks. Agent S combines powerful foundation models (such as GPT-5) with grounding models like UI-TARS to translate visual inputs into precise executable actions. It supports flexible deployment via CLI, SDK, or cloud, and integrates with multiple model providers including OpenAI, Anthropic, Gemini, Azure, and Hugging Face endpoints. ...

...Welcome to npcpy, the core library of the NPC Toolkit that supercharges natural language processing pipelines and agent tooling. npcpy is a flexible framework for building state-of-the-art applications and conducting novel research with LLMs. The structure of npcpy also allows one to pass an npc to get_llm_response in addition to using the NPC's wrapped method, allowing you to be flexible in your implementation and testing.

zeta is a deep learning library focused on providing cutting-edge AI and neural network models with a strong emphasis on research-grade architectures. It includes state-of-the-art implementations for rapid experimentation and model building.

...The project is designed to support large-scale language models (including support for big models via integrated training backends), making it relevant for state-of-the-art research and production use. The framework includes tools for defining workflows, specifying objectives or reward functions, and managing training/policy updates across possibly distributed settings.