Token count zero for streaming AzureOpenAI calls in langchain

[addirathod]

Issue Description: Whenever the streaming is set to True for the LLM initialization (see code below), the token count shows up as zero (as shown in the image).

1. Code 1: I have my LLM instance initialized in the following way:

from langchain.callbacks import AsyncIteratorCallbackHandler
from langchain_openai import AzureChatOpenAI
def get_chain_details(streaming= True, deployment_name="gpt4-8k"):
callback = AsyncIteratorCallbackHandler()
LLM_BIG_STREAMING = AzureChatOpenAI(
streaming=streaming,
callbacks=[callback],
temperature=0,
azure_endpoint=AZURE_OPENAI_API_BASE,
api_version="2023-05-15",
azure_deployment=deployment_name,
api_key=AZURE_OPENAI_API_KEY,
)

2. Code 2: And my instrumentation initialized like:

import phoenix as px
session = px.launch_app()
from phoenix.trace.langchain import LangChainInstrumentor
LangChainInstrumentor().instrument()

3. Versions
   Langchain: 0.1.12
   Phoenix: 4.1.2

Will appreciate any help here Thanks

[RogerHYang]

Token counting for streaming is only recently available, so it's probably not in your API version, i.e. 2023-05-15.