README.md

metadata

license: other
library_name: calyx
tags:
  - calyx
  - embeddings
  - onnx
  - sentence-transformers
  - multimodal

Calyx Embedder Pack (private)

A self-contained bundle of 42 commissioned Calyx lenses (embedders) plus the registry that wires them in. Download, extract, run the installer — they plug into any Calyx install.

Download

hf download cabdru/calyx-embedders calyx-embedders.tar.gz --repo-type model --local-dir .
# (older CLI: huggingface-cli download cabdru/calyx-embedders calyx-embedders.tar.gz --local-dir .)

Install

tar -xf calyx-embedders.tar.gz
cd calyx-embedders
CALYX_HOME=/path/to/your/calyx ./install.sh      # defaults to $HOME/calyx
calyx lens list                                  # should show 42 lenses

The installer copies lenses/ into your CALYX_HOME and rewrites all absolute paths to your machine. Full details + the per-lens list are in the bundle's README.md and INVENTORY.txt.

Coverage (42 lenses)

• text — 28: clinical (pubmedbert, medcpt, biolord), scientific (scibert, scincl), finance (finbert), legal (modernbert), code (jina-code), general/multilingual (bge, gte, nomic, jina, mxbai, e5, minilm, embeddinggemma, qwen3, potion), ColBERT ×3, SPLADE/bge-m3 sparse.
• image — 6, audio — 5, molecule — 1, dna — 1, protein — 1.

Requirements to run a lens

• ONNX GPU lenses: a CUDA ONNX Runtime built for your GPU arch (ORT_DYLIB_PATH).
• *-tei lenses: a local Text-Embeddings-Inference server at the lens endpoint.
• adapter lenses (image/audio/dna/protein/molecule): a Python helper + venv.

⚠️ Licensing

This is a private repo. Each lens keeps its upstream model's license (see each lensforge.manifest.json). Several are non-commercial / restricted — BioLORD (CC-BY-NC), MedCPT (restricted), EmbeddingGemma (Gemma license), jina-colbert-v2 (CC-BY-NC). Anyone you share this with must honor each model's original terms. This pack does not relicense any model.