Skip to main content

Roboflow

You can use Roboflow Inference with Chroma to calculate multi-modal text and image embeddings with CLIP. through the RoboflowEmbeddingFunction class. Inference can be used through the Roboflow cloud, or run on your hardware.

Roboflow Cloud Inference​

To run Inference through the Roboflow cloud, you will need an API key. Learn how to retrieve a Roboflow API key.

You can pass it directly on creation of the RoboflowEmbeddingFunction:

from chromadb.utils.embedding_functions import RoboflowEmbeddingFunction

roboflow_ef = RoboflowEmbeddingFunction(api_key=API_KEY)

Alternatively, you can set your API key as an environment variable:

export ROBOFLOW_API_KEY=YOUR_API_KEY

Then, you can create the RoboflowEmbeddingFunction without passing an API key directly:

from chromadb.utils.embedding_functions import RoboflowEmbeddingFunction

roboflow_ef = RoboflowEmbeddingFunction()

Local Inference​

You can run Inference on your own hardware.

To install Inference, you will need Docker installed. Follow the official Docker installation instructions for guidance on how to install Docker on the device on which you are working.

Then, you can install Inference with pip:

pip install inference inference-cli

With Inference installed, you can start an Inference server. This server will run in the background. The server will accept HTTP requests from the RoboflowEmbeddingFunction to calculate CLIP text and image embeddings for use in your application:

To start an Inference server, run:

inference server start

Your Inference server will run at http://localhost:9001.

Then, you can create the RoboflowEmbeddingFunction:

from chromadb.utils.embedding_functions import RoboflowEmbeddingFunction

roboflow_ef = RoboflowEmbeddingFunction(api_key=API_KEY, server_url="http://localhost:9001")

This function will calculate embeddings using your local Inference server instead of the Roboflow cloud.

For a full tutorial on using Roboflow Inference with Chroma, refer to the Roboflow Chroma integration tutorial.