Hugging Face Server - Chroma

import { Warning } from '/snippets/callout.mdx';

Chroma provides a convenient wrapper for HuggingFace Text Embedding Server, a standalone server that provides text embeddings via a REST API. You can read more about it here.

Setting Up The Server

To run the embedding server locally you can run the following command from the root of the Chroma repository. The docker compose command will run Chroma and the embedding server together.

terminal

docker compose -f examples/server_side_embeddings/huggingface/docker-compose.yml up -d

terminal

docker run -p 8001:80 -d -rm --name huggingface-embedding-server ghcr.io/huggingface/text-embeddings-inference:cpu-0.3.0 --model-id BAAI/bge-small-en-v1.5 --revision -main

<Warning> The above docker command will run the server with the `BAAI/bge-small-en-v1.5` model. You can find more information about running the server in docker [**here**](https://github.com/huggingface/text-embeddings-inference#docker). </Warning>

Usage

python

from chromadb.utils.embedding_functions import HuggingFaceEmbeddingServer
huggingface_ef = HuggingFaceEmbeddingServer(url="http://localhost:8001/embed")

typescript

// npm install @chroma-core/huggingface-server

import { HuggingFaceEmbeddingServerFunction } from "@chroma-core/huggingface-server";

const embedder = new HuggingFaceEmbeddingServerFunction({
    url: "http://localhost:8001/embed",
});

// use directly
const embeddings = embedder.generate(["document1", "document2"]);

// pass documents to query for .add and .query
let collection = await client.createCollection({
    name: "name",
    embeddingFunction: embedder,
});
collection = await client.getCollection({
    name: "name",
    embeddingFunction: embedder,
});

</CodeGroup>

The embedding model is configured on the server side. Check the docker-compose file in examples/server_side_embeddings/huggingface/docker-compose.yml for an example of how to configure the server.

Authentication

The embedding server can be configured to only allow usage with API keys. You can use authentication in the chroma clients:

python

from chromadb.utils.embedding_functions import HuggingFaceEmbeddingServer
huggingface_ef = HuggingFaceEmbeddingServer(url="http://localhost:8001/embed", api_key="your secret key")

typescript

import { HuggingFaceEmbeddingServerFunction } from "chromadb";
const embedder = new HuggingFaceEmbeddingServerFunction({
    url: "http://localhost:8001/embed",
    apiKey: "your secret key",
});

</CodeGroup>