Added embeddings endpoint to server

[Montagon]

This PR include the endpoint for the embeddings in the same way that OpenAI supports it.

[javipacheco]

In order to finish this endpoint we should be sure that the embeddings are working fine. For that, we are using the SpaceCraftLocal example

We should change this example using the local embedding. Something like that

  val localOpenAi = OpenAI(host = "http://localhost:8081/")

  val model = localOpenAi.DEFAULT_SERIALIZATION

  val scope =
    Conversation(LocalVectorStore(OpenAIEmbeddings(localOpenAi.DEFAULT_EMBEDDING)))

....

[Intex32]

@Montagon have you tested streaming? when I tested that earlier, it hang forever. probably something was just wrong with my setup.

[Montagon]

@Montagon have you tested streaming? when I tested that earlier, it hang forever. probably something was just wrong with my setup.

In which example did you try? I tried the SpaceCraftLocal with the new endpoint and it fails, not sure if related. But I'm checking it. Thanks!

[Intex32]

@Montagon have you tested streaming? when I tested that earlier, it hang forever. probably something was just wrong with my setup.

In which example did you try? I tried the SpaceCraftLocal with the new endpoint and it fails, not sure if related. But I'm checking it. Thanks!

I used Postman, the response hanging and never finished.

[Intex32]

@Montagon have you tested streaming? when I tested that earlier, it hang forever. probably something was just wrong with my setup.

As I found out, whether the request hangs or not depends on the length of the response from OpenAI:
Prompting to "Write an essay about climate change" does always hang, as an essay is a long text.
Short prompts do work as expected.
Now, the threshold appears to be around 80,000 chars (or bytes not sure) - mas o menos.

I haven't figured out why this is.

[Intex32]

I think we're set. Venga!

My issue with long prompts is going to be targeted in a different PR.

[javipacheco]

🚀