HTTP 400 error when using Meta's ImageBind model deployed to Vertex AI

[jackyliang]

I deployed Meta's ImageBind multi-modal embedding model to test multi-modality, however, every time I try to send a sample JSON, it fails with an HTTP 400 error telling me that I have an invalid HTTP message

This is the prediction model: https://console.cloud.google.com/vertex-ai/publishers/meta/model-garden/imagebind?project=multi-moda...

And this is one of its sample JSONs that are in the documentation (in the same page)

{
    "instances": [
        {
            "text": ["A dog", "A cat", "A car"],
            "vision": [
                "gs://your-bucket/dog_image.jpg",
                "gs://your-bucket/car_image.jpg",
                "gs://your-bucket/bird_image.jpg"],
            "audio": [
                "gs://your-bucket/dog_audio.wav",
                "gs://your-bucket/dog_audio.wav",
                "gs://your-bucket/dog_audio.wav"]
        }
    ]
}

Attached is the HTTP 400 error in logs 

Is something wrong with the model, because what the JSON in the examples does not return a correct response when I try to send the request

[jackyliang]

I gave up trying to make this work. I'm using the Google Multimodal Embedding Model instead. 

I don't recommend using this library because the docs are incomplete, the errors don't tell you much, uses out of date dependencies (if installed locally) and it is not production ready.