srigf/themis-embeddinggemma-300m-onnx-int8

Themis EmbeddingGemma 300M ONNX INT8

ONNX INT8 runtime artifact used by Themis — Engenharia Jurídica for local semantic retrieval.

Base model

Derived from:

google/embeddinggemma-300m

EmbeddingGemma is developed by Google DeepMind.

This repository is not an official Google distribution.

Runtime artifact

model_int8.onnx

• Size: 310016659 bytes
• SHA-256: 1b477e4439d33c26fa925f17fa5901d524daf9487a9f39b38e6d84ba7301f0ea
• ONNX IR: 8
• Opset: 18
• Output embedding dimension: 768
• Quantization: INT8
• Packaging: single ONNX file, without external_data

tokenizer.json

• Size: 33385008 bytes
• SHA-256: 6852f8d561078cc0cebe70ca03c5bfdd0d60a45f9d2e0e1e4cc05b68e9ec329e

Modification notice

model_int8.onnx is a modified/converted and quantized runtime artifact derived from EmbeddingGemma 300M.

The exact historical export and quantization toolchain used to create this artifact was not reconstructed. The artifact published here is the byte-identical runtime version validated and used by Themis and is identified by the SHA-256 checksum above.

No claim is made that this ONNX file is an official Google-produced artifact.

Intended use

This runtime is used locally by Themis for:

• semantic retrieval;
• movement-summary embeddings;
• document search;
• cosine-similarity ranking.

The model does not generate legal advice or legal conclusions.

License

This repository contains a derivative of EmbeddingGemma and is subject to the Gemma Terms of Use and the Gemma Prohibited Use Policy.

See the LICENSE and NOTICE files in this repository.

Official Gemma Terms:

https://ai.google.dev/gemma/terms

Official EmbeddingGemma model:

https://huggingface.co/google/embeddinggemma-300m