Add rft cookbook for healthbench #1849
Merged
Conversation
This file contains hidden or 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
tompakeman-oai
requested changes
May 23, 2025
There was a problem hiding this comment.
Headlines
- Overall I think it's great but it needs more contextualising with the use case
- What problem fine tuning solves, why few shot is helpful, etc.
- Basically add some user focused language in markdown cells that can be read on the cookbook website by someone who doesn't want to run the code.
- I also think you want to submit the cookbook with the outputs generated so users can see the charts in the final output.
Other changes
- Minor python optimisations for opening files
- Fixes for job polling + using responses API format
- Bug in the matplotlib chart
examples/fine-tuned_qa/reinforcement_finetuning_healthbench.ipynb
Outdated
Show resolved
Hide resolved
examples/fine-tuned_qa/reinforcement_finetuning_healthbench.ipynb
Outdated
Show resolved
Hide resolved
examples/fine-tuned_qa/reinforcement_finetuning_healthbench.ipynb
Outdated
Show resolved
Hide resolved
examples/fine-tuned_qa/reinforcement_finetuning_healthbench.ipynb
Outdated
Show resolved
Hide resolved
examples/fine-tuned_qa/reinforcement_finetuning_healthbench.ipynb
Outdated
Show resolved
Hide resolved
examples/fine-tuned_qa/reinforcement_finetuning_healthbench.ipynb
Outdated
Show resolved
Hide resolved
|
Looks way better. Final thing is that I think it would be awesome to rerun the same evaluation post fine tuning so we can do a direct before and after on the scoring. Apart from that, LGTM |
tompakeman-oai
approved these changes
May 23, 2025
theophile-oai
pushed a commit
that referenced
this pull request
May 23, 2025
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.
Summary
New cookbook to showcase the RFT capabilities in the OpenAI API.
Motivation
Motivation for this was I was fixing/updating some of the old fine-tuning cookbooks. The biggest challenge with upgrading them is model performance has improved so much that we need new challenging datasets to show substantial finetuning gains (which is where HealthBench is ideal).
RFT is also a new feature with few examples and this cookbook covers it end-to-end including working with evals.