Add @track(flush=True)

[dsblank]

Details

This PR adds @track(flush=True) so that you don't have to do it manually after the function has been called.

Before:

from opik import track, flush_tracker

@track()
def streaming_function(input):
    messages = [{"role": "user", "content": input}]
    response = framework.completion(
        model="gpt-3.5-turbo",
        messages=messages,
    )
    return response
streaming_function("Why?")
flush_tracker()

After:

from opik import track

@track(flush=True)
def streaming_function(input):
    messages = [{"role": "user", "content": input}]
    response = framework.completion(
        model="gpt-3.5-turbo",
        messages=messages,
    )
    return response
streaming_function("Why?")

Issues

None, except for adding some convenience.

I'm not sure if one can flush even async decorators inside the decorator?

Testing

No tests added yet. Just seeing if you think this would be worth while first.

Documentation

None yet.

[andrescrz]

LGTM, but let's get input from @alexkuzmik @japdubengsub

[alexkuzmik]

@dsblank this will not work for generators. See the code above, if the generator was decorated, we return wrapped generator before you are calling flush().

[dsblank]

Is there anything we can do to flush after the generator is done?

[alexkuzmik]

we can do flushing in _after_call method.

[dsblank]

Ok, thanks for the info! I'll revise the PR and flush in the _after_call.

[jverre]

@dsblank Closing this PR for now and we can re-open when it's ready