Conversation
Freeze the 774-episode variants, score them, and wire Modal training so the dashboard claim is reproducible. Co-authored-by: Cursor <cursoragent@cursor.com>
…col. The fork README should explain the metric, the frozen fold-clothes variants, and what the Modal training job is allowed to change. Co-authored-by: Cursor <cursoragent@cursor.com>
Spell out the CLIP representation, coverage/repetition math, farthest-first selection, and the equal-update HPT invariants. Co-authored-by: Cursor <cursoragent@cursor.com>
![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: c995fb3f4a
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".
| if last.is_file(): | ||
| return str(last) |
There was a problem hiding this comment.
When save_last: true is enabled alongside the ModelCheckpoint that monitors Valid/action_loss with save_top_k: 1, this branch makes every full run evaluate last.ckpt whenever it exists rather than the checkpoint with the lowest validation loss. In any run where the final checkpoint is not the best validation checkpoint, the reported test_action_loss is for the wrong model and the curation comparison can be misleading; persist/read the callback's best path or choose the top-k checkpoint instead of preferring last.ckpt.
Useful? React with 👍 / 👎.
One fold-clothes zarr has undecodable JPEGs in every frame; the loader used to exhaust that clip and kill the run. Co-authored-by: Cursor <cursoragent@cursor.com>
1 participant
No description provided.