May2222/Fisher-R1-7B
Fisher-R1-7B by May2222 is a 7.6 billion parameter open-weight LLM agent post-trained from Qwen2.5-Coder-7B-Instruct. It is specifically designed for reliable hypothesis testing, inspecting data, selecting and executing statistical tests, reporting p-values, and drawing conclusions. The model was fine-tuned using supervised learning and reinforcement learning with verified statistical rewards. It excels at automating statistical analysis tasks, offering a specialized tool for data scientists and researchers.
Loading preview...
Overview
May2222/Fisher-R1-7B is a 7.6 billion parameter open-weight LLM agent developed by May2222, specifically engineered for reliable hypothesis testing. It is built upon the Qwen2.5-Coder-7B-Instruct architecture and has undergone post-training through supervised fine-tuning (SFT) followed by reinforcement learning (RL) using verified statistical rewards. This specialized training enables the model to perform complex statistical analysis tasks autonomously.
Key Capabilities
- Automated Hypothesis Testing: Designed to inspect datasets, identify appropriate statistical tests, and execute them.
- P-value Reporting: Accurately reports p-values derived from statistical tests.
- Conclusion Drawing: Formulates conclusions based on the statistical analysis.
- Reliability: Training incorporates reinforcement learning with verified statistical rewards to enhance the reliability of its outputs.
- Evaluation: Evaluated on the P-Bench dataset, a benchmark for statistical task performance.
Good For
- Developers and researchers requiring an automated agent for statistical hypothesis testing.
- Applications involving data inspection and the selection of appropriate statistical methods.
- Use cases where reliable reporting of p-values and data-driven conclusions are critical.
For detailed information on its training setup, evaluation protocols, and results, refer to the associated research paper: Fisher-R1: Training LLM Agents for Reliable Hypothesis Testing.