Add patronus eval tool docs
Adding docs for the following:
- PatronusEvalTool
- PatronusLocalEvaluatorTool
- PatronusPredefinedCriteriaEvalTool
Disclaimer: This review was made by a crew of AI Agents.
Code Review Comment for Patronus Evaluator Documentation
Overview
This pull request adds documentation for three important tools in the Patronus framework:
- PatronusEvalTool
- PatronusLocalEvaluatorTool
- PatronusPredefinedCriteriaEvalTool
While the documentation generally provides clear and useful information, several areas require improvement for consistency, clarity, and usability.
Recommendations
1. Standardize Frontmatter Formatting
The frontmatter should be consistent across all documentation files. For example, ensure that all headers start with the specific YAML formatting. Here’s a corrected example:
-
---
title: Patronus Eval Tool
+ ---
+ title: Patronus Eval Tool
2. Consistent Title Formatting
Titles must be uniformly formatted to enhance clarity in navigation. For example, the title in the patronuslocalevaluatortool.mdx should align with the others:
- title: Patronus Eval Tool
+ title: Patronus Predefined Criteria Eval Tool
3. Uniform Link Formatting
Ensure all links follow a consistent format, particularly for URLs:
- app.patronus.ai](http://app.patronus.ai)
+ [app.patronus.ai](https://app.patronus.ai)
4. Improve Numbered List Formatting and Clarity
Lists should maintain a clear structure without unintended spaces:
- 1. Fetch all available criteria
- 2. Select the most fitting criteria
- 3. Evaluates the inputs/outputs/context
+ 1. Fetch all available criteria
+ 2. Select the most fitting criteria
+ 3. Evaluate the inputs/outputs/context
5. Use Language-Specific Code Block Tags
Specify the language in code snippets for improved clarity and syntax highlighting. For instance:
- ```shell
+ ```bash
6. Code Comments in Examples
Adding comments in code snippets can significantly enhance understanding, especially for new users. Here’s an added comment example:
patronus_eval_tool = PatronusLocalEvaluatorTool(
evaluator="local_evaluator_name", # Name of the registered evaluator function
evaluated_model_gold_answer="test" # Expected correct answer for comparison
)
7. Clear Configuration Example
Improve clarity in configuration examples to avoid confusion:
patronus_eval_tool = PatronusPredefinedCriteriaEvalTool(
evaluators=[{
"evaluator": "judge", # Built-in Patronus evaluator
"criteria": "contains-code" # Pre-defined evaluation criteria
}]
)
8. Additional Recommendations
- Add Links to API Reference Documentation: This will facilitate users in exploring more about the functions and their applications.
- Introduce Troubleshooting Sections: Allow users to resolve common issues quickly.
- Standardize Heading Capitalization: Ensure titles have consistent capitalization styles throughout the documentation.
- Add Parameter Type Information in Code Examples: This will improve clarity for users unfamiliar with the tool's implementation.
Positive Aspects
- Well-structured overall documentation.
- Comprehensive coverage detailing each tool's capabilities and installation instructions.
- Helpful code examples lead to better understanding and implementation.
Conclusion
The documentation lays a strong foundation for using the Patronus evaluator tools but can greatly benefit from these adjustments. By incorporating these suggestions, we will enhance the quality, clarity, and maintainability of the documentation, ultimately improving the user experience.
Thank you for your efforts to improve the Patronus documentation! I look forward to seeing these enhancements implemented.
Hey @Rebecca-Qian! Can you please merge in main and shoot me a message when you do so I can get this merged in?
Thank you for updating these docs!
Hey @Rebecca-Qian! Can you please update this PR to allow edits by maintainer so I can update this branch?
Steps for the Contributor: Go to the Pull Request page in the GitHub repository. On the right-hand sidebar (near the bottom), they should see a checkbox labeled "Allow edits by maintainers" under the "Reviewer" and "Assignees" sections. Check the box to allow maintainers to push changes to their PR branch.
Closing due to inactivity.