Parse OpenAI response embedding to contain all embeddings

[DZunke]

When i request the embeddings with mutliple texts the response only contaion the first entry of the response. I did not really understand why it should be this way as the VectorResponse already supports having multiple vectors and so have adapted the response converter for the OpenAi Bridge embeddings.

$texts = array_fill(0, 20, 'Foo Bar Baz');
$vectors = $platform->request(new Embeddings(), $texts)->getContent();

The response of such a request, utilizing the OpenAI embeddings is an array with a single text or an array of text. So it does not change the current result of the VectorResponse for the user of the request.

Hope this fits?

[OskarStark]

Why not using a JsonMockResponse directly?

[DZunke]

Do you maybe have more information to get me on your level here? I am a bit stuck with that you could mean with a JsonMockResponse 🤔

[OskarStark]

[DZunke]

Ah! Ok. Nice one. But what would be the benefit from not mocking the ResponseInterface here? The only thing i get from the documentation is that it would improve testing timeouts and chunked responses, which i do not want to do.

But nethertheless i have tried it and got the following exception that hints me to utilize the mock http client but i do not utilize any http client for my test as i am testing the response converter itself.

Symfony\Component\HttpClient\Exception\InvalidArgumentException: MockResponse instances must be issued by MockHttpClient before processing.

[chr-hertel]

i think in this case the benefit is not really there yet - more like a question of style, e.g. not using mock frameworks but implementations.

in general i can think of some infrastructure for MockResponse from file where also processing is done already - when oop design is done, it could be a quite common scenario to test those response converters and contract is ResponseInterface yup

[chr-hertel]

Yes, i guess that was just a short coming - thanks for the patch & test!