README.md

---
license: mit
tags: [onnx, int8, semantex, embedder]
base_model: nomic-ai/CodeRankEmbed
---

# CodeRankEmbed-onnx-int8

int8 ONNX build of [`nomic-ai/CodeRankEmbed`](https://huggingface.co/nomic-ai/CodeRankEmbed)
(Nomic AI), prepared for [semantex](https://github.com/MisterTK/semantex)'s
local CPU `embedder` path.

- **License:** MIT — inherited from the upstream model; original
  copyright and attribution to **Nomic AI** are retained.
- **Derived from ONNX export:** `sirasagi62/code-rank-embed-onnx`.
- **Files:** `model_int8.onnx`, `tokenizer.json`, `config.json` (the layout semantex's
  downloader expects).

## Notes
- Single-vector bi-encoder (nomic_bert). Documents embed RAW code (no prefix); queries get the query_prefix above. Pool = mean over tokens, then L2-normalize. Alt source already int8: mrsladoje/CodeRankEmbed-onnx-int8 (--source-repo + --no-quantize).

Produced by `benchmarks/onnx_models/prepare_models.py`. Do not edit by hand.