multilingual e5 large instruct embedding model from intfloat

Embedding

35 Pulls Updated 13 days ago

Readme

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": ["passage: my text 1", "passage: 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)

Model Notes

You should use the structure for queries:

Instruct: {task_description}
Query: {query}

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

Links