Skip to content

Bring your own CLIP model

Most customers can get great performance from publicly available CLIP models. However, some use cases will benefit even more from a model fine-tuned for their domain specific task. In this circumstance, you should use your own model with fine-tuned weights and parameters. It is very convenient to incorporate your own model in Marqo as long as your model belongs to one of the following frameworks:

To use your fine-tuned model, here are the detailed steps:

1. Fine-tune your model

The first step is to fine-tune your model using the frameworks mentioned above. Here we use Open CLIP framework as an example. You should follow the guide to fine-tune your own model and store the trained model (checkpoint) as a *.pt file.

2. Upload your model to a server

You need to upload your model (the *.pt file) to a cloud storage (e.g., Amazon S3, GitHub) and use the downloading address to reference it in Marqo.

3. Use your model in Marqo

To use your custom model, you need to create an index in Marqo and define it in your index settings via model and model_properties. For an example Open CLIP model, the code is:

settings = {
    "index_defaults": {
        "treat_urls_and_pointers_as_images": True,
        "model": 'generic-clip-test-model-1',
        "model_properties": {
            "name": "ViT-B-32-quickgelu",
                "dimensions": 512,
                "url": "https://github.com/mlfoundations/open_clip/releases/download/v0.2-weights/vit_b_32-quickgelu-laion400m_avg-8a00ab3c.pt",
                "type": "open_clip",
            },
        "normalize_embeddings": True,
    },
}
response = mq.create_index("my-own-clip", settings_dict=settings)
You can check Generic CLIP for more detailed information and settings in "model_properties".

3. Preloading your model

There may be cases wherein you want to preload (or prewarm, in other terms) your model before using it to index. This can be done by adding your model (with model and model_properties) to the list of models on startup in your marqo configuration.

The syntax for this can be found in Configuring preloaded models