fix: improve completions to send a final chunk with usage details

[drbh]

This PR fixes an issue with the /v1/completions endpoint where the final messages was not treated as a CompletionFinal. This resulted in all of the streaming responses retuning empty usage information and no finish_reason. Now if the chunk is final we handle it accordingly.

Repro

from openai import OpenAI
import os

client = OpenAI(
    base_url="http://localhost:3000/v1",
    api_key=os.getenv("HF_TOKEN", "YOUR_API_KEY"),
)

i = 1337

chat_completion = client.completions.create(
    model="mistralai/Mistral-7B-Instruct-v0.1",
    prompt="What are three words that describe the Python programming language?",
    max_tokens=20,
    stream=True,
    seed=i,
)

# iterate over the stream of messages
for message in chat_completion:
    print(message)

response

Completion(id='', choices=[CompletionChoice(finish_reason='', index=0, logprobs=None, text='\n')], created=1722373608, model='mistralai/Mistral-7B-Instruct-v0.1', object='text_completion', system_fingerprint='2.2.1-dev0-native', usage=None)
Completion(id='', choices=[CompletionChoice(finish_reason='', index=0, logprobs=None, text='Object')], created=1722373608, model='mistralai/Mistral-7B-Instruct-v0.1', object='text_completion', system_fingerprint='2.2.1-dev0-native', usage=None)
Completion(id='', choices=[CompletionChoice(finish_reason='', index=0, logprobs=None, text='-')], created=1722373609, model='mistralai/Mistral-7B-Instruct-v0.1', object='text_completion', system_fingerprint='2.2.1-dev0-native', usage=None)
Completion(id='', choices=[CompletionChoice(finish_reason='', index=0, logprobs=None, text='oriented')], created=1722373609, model='mistralai/Mistral-7B-Instruct-v0.1', object='text_completion', system_fingerprint='2.2.1-dev0-native', usage=None)
Completion(id='', choices=[CompletionChoice(finish_reason='', index=0, logprobs=None, text=',')], created=1722373609, model='mistralai/Mistral-7B-Instruct-v0.1', object='text_completion', system_fingerprint='2.2.1-dev0-native', usage=None)
Completion(id='', choices=[CompletionChoice(finish_reason='', index=0, logprobs=None, text=' Vers')], created=1722373609, model='mistralai/Mistral-7B-Instruct-v0.1', object='text_completion', system_fingerprint='2.2.1-dev0-native', usage=None)
Completion(id='', choices=[CompletionChoice(finish_reason='', index=0, logprobs=None, text='atile')], created=1722373609, model='mistralai/Mistral-7B-Instruct-v0.1', object='text_completion', system_fingerprint='2.2.1-dev0-native', usage=None)
Completion(id='', choices=[CompletionChoice(finish_reason='', index=0, logprobs=None, text=',')], created=1722373609, model='mistralai/Mistral-7B-Instruct-v0.1', object='text_completion', system_fingerprint='2.2.1-dev0-native', usage=None)
Completion(id='', choices=[CompletionChoice(finish_reason='', index=0, logprobs=None, text=' Easy')], created=1722373609, model='mistralai/Mistral-7B-Instruct-v0.1', object='text_completion', system_fingerprint='2.2.1-dev0-native', usage=None)
Completion(id='', choices=[CompletionChoice(finish_reason='', index=0, logprobs=None, text=' to')], created=1722373609, model='mistralai/Mistral-7B-Instruct-v0.1', object='text_completion', system_fingerprint='2.2.1-dev0-native', usage=None)
Completion(id='', choices=[CompletionChoice(finish_reason='', index=0, logprobs=None, text=' learn')], created=1722373609, model='mistralai/Mistral-7B-Instruct-v0.1', object='text_completion', system_fingerprint='2.2.1-dev0-native', usage=None)
Completion(id='', choices=[CompletionChoice(finish_reason='', index=0, logprobs=None, text='.')], created=1722373609, model='mistralai/Mistral-7B-Instruct-v0.1', object='text_completion', system_fingerprint='2.2.1-dev0-native', usage=None)
Completion(id='', choices=[CompletionChoice(finish_reason='eos_token', index=0, logprobs=None, text='</s>')], created=1722373609, model='mistralai/Mistral-7B-Instruct-v0.1', object='text_completion', system_fingerprint='2.2.1-dev0-native', usage=CompletionUsage(completion_tokens=13, prompt_tokens=12, total_tokens=25))

[Narsil]

LGTM