Improve test cases for eagle infer #7173
Conversation
There was a problem hiding this comment.
Summary of Changes
Hello @merrymercy, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!
This pull request focuses on improving the test suite for the Eagle inference functionality. It refactors one test file by removing unused code and enhances another test (test_logprob_match) to be more robust by testing logprob consistency under varying temperature settings and ensuring temperature-scaled logprobs are used. Additionally, it improves the logging format in a separate test for better clarity.
Highlights
- Test Logging Improvement: Improved the print output in
test_eagle_infer_a.pyto format theacc_lengthmetric to 4 decimal places and include thespeedmetric for better readability and analysis. - Test Refactoring: Cleaned up
test_eagle_infer_b.pyby removing unused imports (torch) and variables (torch_dtype,prefill_tolerance,decode_tolerance). - Logprob Test Enhancement: Enhanced the
test_logprob_matchfunction intest_eagle_infer_b.pyto test logprob consistency across different temperatures (1.0 and 0.7) and explicitly request temperature-scaled logprobs from the API, making the test more comprehensive. - Assertion Tolerance Adjustment: Slightly increased the assertion tolerance for logprob difference in
test_logprob_matchfrom 0.25 to 0.255 to accommodate potential minor variations when testing with different temperatures.
Using Gemini Code Assist
The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.
Invoking Gemini
You can request assistance from Gemini at any point in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.
| Feature | Command | Description |
|---|---|---|
| Code Review | /gemini review |
Performs a code review for the current pull request in its current state. |
| Pull Request Summary | /gemini summary |
Provides a summary of the current pull request in its current state. |
| Comment | @gemini-code-assist | Responds in comments when explicitly tagged, both in issue comments and review comments. |
| Help | /gemini help |
Displays a list of available commands. |
Customization
To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.
Limitations & Feedback
Gemini Code Assist is currently in preview and may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments to provide feedback.
You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.
Footnotes
-
Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configureGemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩
There was a problem hiding this comment.
Code Review
This pull request enhances the test cases for 'eagle infer'. Key changes include improved print formatting in test_eagle_infer_a.py, removal of an unused import and variables in test_eagle_infer_b.py, and significant improvements to the test_logprob_match method. This test now iterates over different temperature settings and incorporates a new temp_scaled_logprobs parameter in the API call.
My review includes one comment seeking clarification on an adjustment to an assertion threshold within the updated test_logprob_match, to ensure the test's robustness and accurately reflects expected behavior under the new conditions.
| self.assertLess(max_diff, 0.25) | ||
| diff = np.abs(output_logprobs - output_logprobs_score) | ||
| max_diff = np.max(diff) | ||
| self.assertLess(max_diff, 0.255) |
There was a problem hiding this comment.
The assertion threshold for max_diff has been increased from 0.25 to 0.255 in this test. Could you provide some context for this adjustment? For instance, is this change due to testing with different temperature values, the introduction of the temp_scaled_logprobs: True parameter, or a general refinement based on observed behavior? Understanding the reason behind this slight relaxation of the threshold is helpful for ensuring the test's integrity and accurately reflects the expected logprob matching precision.
No description provided.