Skip to content

Add Image Embedding Provenance Pipeline - #58

Open
NetZissou wants to merge 3 commits into
mainfrom
update/tol-200m-image-embeddings
Open

NetZissou wants to merge 3 commits into
mainfrom
update/tol-200m-image-embeddings

Conversation

@NetZissou

Copy link
Copy Markdown
Contributor

Adds processing/tol200m_image_embeddings/:

the SLURM jobs, configs, and standardization scripts used to produce the image embedding configs of TreeOfLife-200M-Embeddings (bioclip-2_float16, bioclip-2.5-vith14_float16)

Different models share the same pipeline, differing in weights, and normalization options:

  1. Batch embedding: hpc-inference open_clip_embed over the TOL-200M HDF5 shards (multi-rank job for small sources; an image-count-balanced SLURM array for gbif). Outputs raw float32 embeddings.
  2. Standardization: join to the TOL-200M catalog metadata via the published row order (keeps all configs row-aligned), global taxonomic sort, L2 normalization (BioCLIP 2.5 only), float16 cast, and contract parquet write.
  3. Validation: counts, contract fields, full order check, sample recompute.

All site-specific paths and accounts are placeholders.

@NetZissou
NetZissou requested a review from egrace479 August 31, 2026 14:44
@NetZissou NetZissou self-assigned this Aug 31, 2026
@NetZissou NetZissou added the documentation Improvements or additions to documentation label Aug 31, 2026
Comment thread processing/README.md

## TreeOfLife-200M Image Embeddings

The [`tol200m_image_embeddings/`](tol200m_image_embeddings) directory contains the SLURM jobs, configs, and standardization scripts used to produce the image-embedding configs of [TreeOfLife-200M-Embeddings](https://huggingface.co/datasets/imageomics/TreeOfLife-200M-Embeddings) (BioCLIP 2 and BioCLIP 2.5 Huge): batch embedding via the [hpc-inference](https://github.com/Imageomics/hpc-inference) package, then metadata join, taxonomic ordering, normalization (where applicable), float16 casting, and contract parquet writing. See its `README.md` for the pipeline and placeholder conventions.

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
The [`tol200m_image_embeddings/`](tol200m_image_embeddings) directory contains the SLURM jobs, configs, and standardization scripts used to produce the image-embedding configs of [TreeOfLife-200M-Embeddings](https://huggingface.co/datasets/imageomics/TreeOfLife-200M-Embeddings) (BioCLIP 2 and BioCLIP 2.5 Huge): batch embedding via the [hpc-inference](https://github.com/Imageomics/hpc-inference) package, then metadata join, taxonomic ordering, normalization (where applicable), float16 casting, and contract parquet writing. See its `README.md` for the pipeline and placeholder conventions.
The [`tol200m_image_embeddings/`](tol200m_image_embeddings) directory contains the SLURM jobs, configs, and standardization scripts used to produce the image-embedding configs of [TreeOfLife-200M-Embeddings](https://huggingface.co/datasets/imageomics/TreeOfLife-200M-Embeddings) (BioCLIP 2 and BioCLIP 2.5 Huge): batch embedding via the [hpc-inference](https://github.com/Imageomics/hpc-inference) package, then metadata join, taxonomic ordering, normalization (where applicable), float16 casting, and contract parquet writing. See [`tol200m_image_embeddings/README.md`](tol200m_image_embeddings/README.md) for the pipeline and placeholder conventions.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

documentation Improvements or additions to documentation

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants