Fixes #184: JSON parsing issue when async streaming

[athyuttamre]

I was able to reproduce #184 where sometimes calling Completion.acreate(stream=True) would result in a JSON parsing issue.

This was happening because our code was trying to parse incomplete JSON objects. It was receiving chunks that looked like this (notice the incomplete line at the end):

data: {"id": "cmpl-6hDMH0KtbEmSdkAvX4xEHy7be8XMz", "object": "text_completion", "created": 1675757473, "choices": [{"text": "\\n", "index": 0, "logprobs": null, "finish_details": null}], "model": "text-davinci-003"}\n\ndata: {"id": "cmpl-6hDMH0KtbEmSdkAvX4xEHy7be8XMz", "object": "text_completion", "created": 1675757473, "choices": [{"text": "Hello", "index": 0, "logprobs": null, "finish_details": null, "model": "text-davinci-003"}\n\ndata: {"id": "cmpl-6hDMH0KtbEmSdkAvX4xEHy7be8XMz", "object": "text_completion", "created": 1675757473, "choices": [{"text": " again", "index": 0, "logprobs": null, "finish_details": null, "model": "text-davinci-003"}\n\ndata: {"id": "cmpl-6hDMH0KtbEmSdkAvX4xEHy7be8XMz", "object": "text_completion", "created": 1675757473, "choices": [{"text": "!", "index": 0, "logprobs": null, "finish_details": null, "model": "text-davinci-003"}\n\ndata: {"id": "cmpl-6hDMH0KtbEmSdkAvX4xEHy7be8XMz", "object": "text_completion", "created": 1675757473, "choices": [{"text": " Is", "index": 0, "logprobs": null, "finish_details": null, "model": "text-davinci-003"}\n\ndata: {"id": "cmpl-6hDMH0KtbEmSdkAvX4xEHy7be8XMz", "object": "text_completion", "created": 1675757473, "choices": [{"text": " there", "index": 0, "logprobs": null, "finish_details": null, "model": "text-davinci-003"}\n\ndata: {"id": "cmpl-6hDMH0KtbEmSdkAvX4xEHy7be8XMz", "object": "text_completion", "created": 1675757473, "choices": [{"text": " something", "index": 0, "logprobs": null, "finish_details": null, "model": "text-davinci-003"}\n\ndata: {"id": "cmpl-6hDMH0KtbEmSdkAvX4xEHy7be8XMz", "object": "text_completion", "c

It looks like we were using the wrong function from aiohttp to parse streaming bodies. Instead of rbody.iter_chunks() we should be using async for line in rbody instead. See docs here: https://docs.aiohttp.org/en/stable/streams.html#asynchronous-iteration-support.

This change immediately fixed the issue in my code.

[ddeville]

We went back and forth on this that tbh I've lost sense of what the right fix for this is :)

We moved to using iter_chunks in #172 and then I I went back to looping over the body itself and splitting additional lines afterwards, and then put back the iter_chunks upon review in #177.

Reading the docs your change seems correct to me but since we've had issues in the past make sure that what was raised in #171 doesn't happen again after this PR.

[athyuttamre]

I ran the previously added tests. I think I saw those issues myself too and no longer see them. Fingers crossed!

[ddeville]

Awesome!