CodeRankEmbed Transformers.js ONNX
This repository contains a Transformers.js-compatible ONNX export of
nomic-ai/CodeRankEmbed for Cortex embedding experiments.
Runtime contract
- Task:
feature-extraction - Pooling:
cls - Normalize:
true - Dimensions:
768 - Cortex model id: use this repository id as
CORTEX_EMBED_MODEL - Query embedding text must use the CodeRankEmbed instruction prefix:
Represent this query for searching relevant code: <query>
CodeRankEmbed must use CLS pooling. Mean pooling produced materially different vectors in local validation. Code/passages should be embedded unchanged; the instruction prefix is for search queries only.
Expected layout
.
├── config.json
├── tokenizer.json
├── tokenizer_config.json
├── special_tokens_map.json
└── onnx/
└── model.onnx
The ONNX model is saved as a single file so Transformers.js can fetch it from the Hub without requiring a separate external data sidecar.