Add CLI tools for chat fine-tuning #2659
Conversation
This implements the feature requested in issue openai#622 to provide CLI tools for chat fine-tuning that work with the new chat message format. Added functionality: - `openai tools chat_fine_tunes.prepare_data` for validating and preparing chat fine-tuning data in the messages format - `openai api chat_fine_tunes.*` commands (create, list, get, cancel) as convenience aliases for the existing fine_tuning.jobs commands - Chat-specific validators that check for proper message format with role and content fields - Support for system, user, and assistant roles The tools work with JSONL files containing messages arrays, which is the format required for chat fine-tuning with the new API. Fixes openai#622
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
Codex Review: Here are some suggestions.
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, or 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 fix this CI failure" or "@codex address that feedback".
| from __future__ import annotations | ||
|
|
||
| from typing import TYPE_CHECKING | ||
| from argparse import ArgumentParser | ||
|
|
||
| from .chat_fine_tunes import jobs | ||
|
|
||
| if TYPE_CHECKING: | ||
| from argparse import _SubParsersAction | ||
|
|
||
|
|
||
| def register(subparser: _SubParsersAction[ArgumentParser]) -> None: | ||
| jobs.register(subparser) No newline at end of file |
There was a problem hiding this comment.
[P1] Avoid circular import in chat_fine_tunes CLI module
The new alias module is defined in cli/_api/chat_fine_tunes.py while a package with the same name (cli/_api/chat_fine_tunes/) contains jobs.py. When _api/_main imports chat_fine_tunes, Python loads this file and the statement from .chat_fine_tunes import jobs resolves to the module itself rather than the sibling package, leaving jobs undefined and raising ImportError: cannot import name 'jobs' from partially initialized module .... As a result, none of the chat_fine_tunes commands can be registered. The alias module needs a different name or should be removed in favour of the package so the import resolves correctly.
Useful? React with 👍 / 👎.
Summary
openai tools chat_fine_tunes.prepare_datafor data validation and preparationopenai api chat_fine_tunes.*commands as convenience aliasesChanges Made
src/openai/cli/_tools/chat_fine_tunes.pywith data preparation functionalitysrc/openai/cli/_api/chat_fine_tunes/module with API command aliasessrc/openai/lib/_validators.pyTest Plan
The implementation follows the existing patterns in the codebase and provides the interface proposed by @henriqueln7 in the issue comments.
Fixes #622