llm-embed-onnx

Run embedding models using ONNX

This LLM plugin is a wrapper around onnx_embedding_models by Benjamin Anderson.

Installation

Install this plugin in the same environment as LLM.

llm install llm-embed-onnx

Usage

This plugin adds the following embedding models, which can be listed using llm embed-models:

onnx-bge-micro
onnx-gte-tiny
onnx-minilm-l6
onnx-minilm-l12
onnx-bge-small
onnx-bge-base
onnx-bge-large

You can run any of these models using llm embed command:

llm embed -m onnx-bge-micro -c "Example content"

This will output a 384 length JSON array of floating point numbers, starting:

[-0.03910085942622519, -0.0030843335461659795, 0.032797761260860724,

The first time you use any of these models the model will be downloaded to the llm_embed_onnx directory in your LLM data directory. On macOS this defaults to:

~/Library/Application Support/io.datasette.llm/llm_embed_onnx

For more on how to use these embedding models see the LLM embeddings documentation.

Development

To set up this plugin locally, first checkout the code. Then create a new virtual environment:

cd llm-embed-onnx
python3 -m venv venv
source venv/bin/activate

Now install the dependencies and test dependencies:

llm install -e '.[test]'

To run the tests:

pytest