Error: list index out of range with 2023-07-01-preview version

Streaming response doesn't work with Azure models. This can be fixed by handling the JSON chunks differently.

There is an issue with AzureChatOpenAI Streaming first response chunk
response["choices"][0] is causing the issue because the server returns an empty array.

This doesn't occur for 2023-03-15-preview version

During streaming, the first chunk may only contain the name of an OpenAI function and not any arguments. In this case, the current code presumes there is a streaming response and tries to append to it, but gets a KeyError.

This can be fixed by checking if the arguments key exists, and if not, creates a new entry instead of appending.

Example:

llm -m azure 'Tell me a quantum entaglement joke' --system 'You are Albert Einstein' --no-stream

Why don't quantum physicists make jokes about entanglement?
Because when they do, they instantly become the subject of someone else's humor halfway around the world!

llm -m azure 'Tell me a quantum entaglement joke' --system 'You are Albert Einstein'

Error: list index out of range

Please refer:
https://github.com/jerryjliu/llama_index/issues/7640

Reproduce the failure with the provided llm -m azure streaming command, then inspect the AzureChatOpenAI streaming response handling around response["choices"][0] and function-call chunks. Done means the 2023-07-01-preview response streams without an index or key error, including chunks that contain only a function name or arguments.