Clamp supervised spans to assistant turns - #34
Conversation
![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 automated review suggestions for this pull request.
Reviewed commit: 7aa07663a8
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".
|
@codex please revie the latest state of everything and make sure it's all good and doesn't break anything existing model paths |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 2799a9923a
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".
|
Addressed the latest Codex review in 202b7fe. The audit now rejects contiguous label runs that begin at non-assistant ChatML headers; full literal ChatML transcript snippets are disambiguated with marker-derived source spans; Llama ipython tool-result turns and Phi <|end|> terminators are indexed; and turn-header validation no longer allocates progressively larger text prefixes. Added focused regressions for each case. Verification: 720 passed / 28 skipped locally, 26/26 live tokenizer tests passed, Ruff and mypy passed, wheel/sdist passed, and all required GitHub CI checks passed. |
CompactAIOfficial
added
enhancement
2 participants
No description provided.