Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

(EAI-353): Quiz question evaluation #440

Merged
merged 10 commits into from
Jul 9, 2024
Merged

(EAI-353): Quiz question evaluation #440

merged 10 commits into from
Jul 9, 2024

Conversation

mongodben
Copy link
Collaborator

@mongodben mongodben commented Jun 20, 2024

Jira: https://jira.mongodb.org/browse/EAI-353

Changes

In mongodb-chatbot-evaluation package:

  • Quiz question data generator (i.e. get response from model)
  • Quiz question evaluator (i.e. is model response correct)
  • Have pipeline functions return metadata with possibility of additional info.
  • Add createdAt field to GeneratedData

New model-eval package for running evals on LLMs including:

  • Create package w/ package config (based on chatbot-eval-mongodb-public)
  • Config to run model quiz question evaluations.
  • Support calling multiple models thru Radiant.
    • Currently just Azure OpenAI models. Hopefully will have more soon...

In chatbot-eval-mongodb-public package:

  • Support new pipeline function format.

Notes

@mongodben mongodben changed the title (EAI-353): Quiz question evaluatin (EAI-353): Quiz question evaluation Jun 20, 2024
@mongodben mongodben marked this pull request as ready for review July 2, 2024 16:56
Copy link
Collaborator

@nlarew nlarew left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM w/ a few non-blocking comments

packages/model-eval/.env.example Show resolved Hide resolved
packages/model-eval/src/radiantModels.ts Show resolved Hide resolved
This can be useful for evaluating how an LLM performance on the subject matter of the multiple choice questions.

The prompt is based on this blog post from Hugging Face: https://huggingface.co/blog/open-llm-leaderboard-mmlu
It follows the [HELM prompt format](https://huggingface.co/blog/open-llm-leaderboard-mmlu#mmlu-comes-in-all-shapes-and-sizes-looking-at-the-prompts).
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Would it be worth splitting this out into a reusable makeHelmPrompt() function w/ accompanying tests? Not necessary for this PR but might be useful down the line + as documentation of the format.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

HELM is just for quiz-style questions, the current makeQuizQuestionPrompt() does what you're describing. i can refactor comments a bit to make this clearer + also export this function and test it

@mongodben mongodben merged commit 364b01f into main Jul 9, 2024
1 check passed
@mongodben mongodben deleted the EAI-353 branch July 9, 2024 19:38
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants