May2222/Fisher-R1-14B
Fisher-R1-14B is a 14.8 billion parameter open-weight LLM agent developed by May2222, post-trained from Qwen2.5-Coder-14B-Instruct. It specializes in reliable hypothesis testing by inspecting data, selecting and executing statistical tests, reporting p-values, and drawing conclusions. This model is uniquely designed for automated statistical analysis tasks, leveraging supervised fine-tuning and reinforcement learning with verified statistical rewards.
Loading preview...
Fisher-R1-14B: LLM Agent for Reliable Hypothesis Testing
Fisher-R1-14B is a 14.8 billion parameter open-weight language model agent developed by May2222, specifically engineered for reliable hypothesis testing. It is built upon the Qwen2.5-Coder-14B-Instruct architecture and has undergone a specialized post-training regimen.
Key Capabilities
- Automated Statistical Analysis: The model is designed to function as an agent that can inspect datasets, identify appropriate statistical tests, and execute them.
- P-value Reporting: It reliably reports p-values derived from its statistical test executions.
- Conclusion Drawing: Based on the statistical analysis, Fisher-R1-14B can draw and present conclusions.
- Advanced Training Methodology: Its training involved supervised fine-tuning on synthetic executable statistical tasks, followed by reinforcement learning guided by verified statistical rewards, ensuring high reliability.
- Evaluated on P-Bench: Performance is assessed using the P-Bench benchmark, a dataset specifically designed for evaluating statistical reasoning in LLMs.
When to Use This Model
This model is particularly well-suited for applications requiring automated and reliable statistical inference. It excels in scenarios where an LLM needs to:
- Perform data inspection and select appropriate statistical methods.
- Execute statistical tests and interpret their results.
- Generate p-values and provide clear, data-driven conclusions.
For detailed information on its training setup, evaluation protocols, and results, refer to the research paper: Fisher-R1: Training LLM Agents for Reliable Hypothesis Testing.