Skip to content

Delete templates/getting-started directory#507

Open
akshay-anyscale wants to merge 1 commit into
mainfrom
akshay-anyscale-patch-37
Open

Delete templates/getting-started directory#507
akshay-anyscale wants to merge 1 commit into
mainfrom
akshay-anyscale-patch-37

Conversation

@akshay-anyscale

Copy link
Copy Markdown
Contributor

No description provided.

Aydin-ab added a commit that referenced this pull request Jul 9, 2026
…56.0 (#896)

## Summary
Bump `audio-dataset-curation-llm-judge` to Ray 2.56.0.

## Changes
- `BUILD.yaml`: `cluster_env.image_uri`
`anyscale/ray-llm:2.55.1-py311-cu128` →
`anyscale/ray-llm:2.56.0-py312-cu130` (blessed rayllm 2.56.0 combo —
forced py311→py312 + cu128→cu130 jump;
`anyscale/ray-llm:2.56.0-py311-cu128` is not published).
- `dependencies/template.depsets.yaml`: repointed this template's
`expand` entry's `build_arg_sets` `ray2551_py311_cu128` →
`ray2560_py312_cu130`, and aligned its extra pytorch index `cu128` →
`cu130`.
- `templates/audio-dataset-curation-llm-judge/python_depset.lock`:
recompiled via `./update_deps.sh --name
audio_dataset_curation_llm_judge_depset_2.56.0_3.12_cu130` (only the
`--extra-index-url` header shifts to `cu130`; pinned deps unchanged).
- `templates/audio-dataset-curation-llm-judge/requirements.txt`:
refreshed the numpy-pin comment to the new base tag.
- `templates/audio-dataset-curation-llm-judge/README.ipynb` (+
regenerated `README.md`): fix `Decoder.__call__` for the newer
transformers Whisper tokenizer. `batch_format="pandas"` makes
`batch.pop("token_ids")` a Series of numpy arrays; recent
`WhisperTokenizer._strip_prompt` now runs plain-Python truthiness checks
(`if not token_ids:`, `if has_prompt:`), both of which raise
`ValueError: The truth value of a Series/array is ambiguous`. Hand the
tokenizer plain nested Python lists: `token_ids = [ids.tolist() for ids
in batch.pop("token_ids")]`.

## Fix iterations
- Build #507 failed at cell In[10] (`ds.write_parquet(...)`) with
`ValueError: truth value of a Series is ambiguous` inside
`WhisperTokenizer._strip_prompt` — `if not token_ids`. First attempt
(`list(batch.pop("token_ids"))`) unstuck that check but Build #509
failed on `if has_prompt:` deeper in the same path, where the Series
elements were numpy arrays.
- Second fix converts every element to a plain Python list. Build #510
passed.

## Tests / validation
- **Local:** `pre-commit run --all-files` — passed.
- **CI:** Buildkite `template-test` [build
#510](https://buildkite.com/anyscale/template-test/builds/510) — passed
(~10 min).

## Publish
`BUILDKITE_API_TOKEN` here lacks `write_builds` scope — a `tmpl-publish`
build cannot be triggered from this run. After merge, a reviewer should
trigger it manually per
`.claude/skills/template/references/publish-to-backend.md`:
- Pipeline: https://buildkite.com/anyscale/tmpl-publish
- `input-tmpl-name` fields:
`tmpl-name=audio-dataset-curation-llm-judge`, `tmpl-branch=main`,
`tmpl-commit=HEAD`.

---------

Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Aydin Abiar <Aydin-ab@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant