HealthBench RFT - Updates to use synthetic data

[robtinn]

Summary

• Adding in some functions to generate synthetic data in this example as examples from HealthBench might not always be suitable for fine-tuning.
• Also tidied up some of the wording and visualization notebook cells.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

responses.parse called with unsupported argument

The new synthetic-data cell calls client.responses.parse(..., text_format=SyntheticData) but the responses API expects a response_format argument for Pydantic parsing; text_format is not a valid parameter and will raise a TypeError before any data is generated. With this change, the notebook cannot produce the synthetic train/val/test JSONL files and the fine-tuning walk-through stops at this cell.

Useful? React with 👍 / 👎.