Skip to content

Add evals for each model#196

Merged
kamath merged 3 commits intomainfrom
anirudh/fix/openai-example
Nov 21, 2024
Merged

Add evals for each model#196
kamath merged 3 commits intomainfrom
anirudh/fix/openai-example

Conversation

@kamath
Copy link
Copy Markdown
Contributor

@kamath kamath commented Nov 21, 2024

why

We want evals for OpenAI and Anthropic. This does that and adds tags for each model if we want to break down performance.

Also, OpenAIClient wasn't working as expected, for some reason LogLine was being imported from the wrong place. This must've gotten lost in a previous PR, and due to the typing hell we're in, it got lost along the way. This fixes that.

what changed

  • Added evals for both Claude 3.5 Sonnet and GPT 4o for all evals with model name tags
  • Fixes example.ts to work with OpenAI and Anthropic

test plan

Braintrust run

Also ran example.ts with modelName of gpt-4o, claude-3-5-sonnet-20241022, and empty and they all work

@kamath kamath marked this pull request as ready for review November 21, 2024 01:52
@kamath kamath changed the title fix: openai example wasn't working Add evals for each model Nov 21, 2024
@kamath kamath requested a review from jeremypress November 21, 2024 01:59
Copy link
Copy Markdown
Contributor

@pkiv pkiv left a comment

Choose a reason for hiding this comment

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

l g t m

@kamath kamath merged commit 377d7d4 into main Nov 21, 2024
Copy link
Copy Markdown
Contributor

@jeremypress jeremypress left a comment

Choose a reason for hiding this comment

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

looks good, might make sense in the future to have "stable" models that we can pin in the LLM Providers file, then any thing we run could pull stable Anthropic, stable Open AI etc, and not have manually set model names

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.

3 participants