Usage

After you pull the image, you can simply use the openai compatible embedding endpoint:

curl -vvvv -H 'Content-Type: application/json' "http://localhost:11434/v1/embeddings" -d '{
  "input": ["my text 1", "my text 2"],
  "model": "jeffh/intfloat-multilingual-e5-large-instruct"
}'

Where model can also include the tag as necessary (such as jeffh/intfloat-multilingual-e5-large-instruct:f16)

Sizes

The following models quantizations are available:

• f32
• f16
• q8_0

Model Notes

You should use this structure for queries:

Instruct: {task_description}
Query: {query}

Documents do not need any special structure to be embedded for indexing.

Links

• Ollama Model
• GitHub Repo
• Original Model