yangzhr/EvoDS
EvoDS by yangzhr is an 8 billion parameter language model, initialized from Qwen3-8B, specifically fine-tuned for autonomous data science tasks. It integrates Autonomous Skill Acquisition (ASA) for expanding its action space and Adaptive Context Compression (ACC) for efficient long-horizon reasoning. This model is designed to continuously improve its capabilities in data science through self-evolving mechanisms.
Loading preview...
EvoDS: Self-Evolving Autonomous Data Science Agent
EvoDS is an 8 billion parameter language model, based on Qwen3-8B, developed by yangzhr. It is specifically designed to function as a self-evolving autonomous data science agent, as detailed in the paper "EvoDS: Self-Evolving Autonomous Data Science Agent with Skill Learning and Context Management". The model's training involves multi-agent supervised fine-tuning and reinforcement learning to enhance its capabilities in data science tasks.
Key Capabilities
- Autonomous Skill Acquisition (ASA): Enables the agent to autonomously synthesize, validate, cache, and reuse executable skills, allowing its action space to expand over time.
- Adaptive Context Compression (ACC): Facilitates efficient long-horizon reasoning by dynamically compressing interaction history and preserving critical information within limited context budgets.
Good for
- Automated data science workflows.
- Applications requiring agents that can learn and adapt new skills.
- Tasks demanding efficient long-context reasoning in data analysis.