Unified generation outputs

[jonathanhecl]

Unified generation outputs
And added the tokens in the response.

[jonathanhecl]

Create a new tag for this version :)

[xyproto]

This is great, and thanks for the PR!

It is important to me that existing programs that uses ollamaclient can continue to use the same API.

So, instead of:

generatedOutput, err := oc.GetOutput(prompt)
[...]
output := generatedOutput.Response

And:

generatedOutput := oc.MustOutput(prompt)
[...]
output := generatedOutput.Response

Could perhaps oc.GetOutput and oc.MustOutput keep the same return types as before, but new methods be added?

For instance oc.GetResult and oc.MustResult?

[jonathanhecl]

Yes, it can look like this. However, I am a bit worried that there are 3 endpoints in ollama to get answers, and I don't know when to use one or the other.

/chat/completions (brings "choices", in theory it is for a string of messages)
/api/generate (this is the one originally used for text, but also returns the prompt tokens)
/api/chat (used for the tools)

Of these we only use the final two.

[xyproto]

As far as I can tell /chat/completions is only for OpenAI compatibility, so I'm not too worried about not supporting this endpoint, unless it has advantages or possibilities that the other endpoints do not have: https://ollama.com/blog/openai-compatibility

[xyproto]

I merged and then adjusted the PR to keep backward compatibility for existing programs and packages that uses the ollamaclient package.