evaluate extraction progress in separate call #110
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…ss separately from main extraction prompt
…and into complete-cond-extract
kamath
reviewed
Oct 23, 2024
evals/index.eval.ts
Outdated
| @@ -281,7 +281,9 @@ const extract_last_twenty_github_commits = async () => { | |||
|
|
|||
| try { | |||
| await stagehand.page.goto("https://github.com/facebook/react"); | |||
| await stagehand.waitForSettledDom(); | |||
There was a problem hiding this comment.
could be wrong here, but i thought we're deprecating this?
There was a problem hiding this comment.
Got it, wasn't aware—removed. Just included since I saw it in some other tests—I figure we need to remove it there as well?
kamath
approved these changes
Oct 24, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
why
Changes for improvements to completion checking and progress updating.
The original issue was that in a single extraction step, we provide previously extracted content, progress so far, completed/not completed, DOM elements, etc. all at once and ask the model to give us newly extracted content from DOM elements, new progress, completed/not completed. The issue here is that the model needs to simultaneously decide what is important in the DOM and evaluate its progress based on the information it extracts.
what changed
This PR turns this into three steps: 1. extracting the content from the DOM (without previous progress, previous extraction content, etc.), 2. Refining content to ensure that we do not add redundant/duplicate data and updating the total response, and 3. deciding based on the total extracted content what the new progress is and whether the job is completed. The main idea here is that breaking this up rather than doing it all at once shows improvements and a reduction in hallucinations.
test plan
Running evals. Shows improvement with awareness of the number of outputs in
arxiveval required since progress tracking is improved. Is able to do both of the Github evals.This shows improvement on both the GitHub tests.