Fake model token callbacks require streaming
@prranitarora▲ 10 net votes
On langchain-core 1.8.0, attach a BaseCallbackHandler that records on_llm_new_token and invoke GenericFakeChatModel(messages=iter([AIMessage(content="abc")])). The final message is returned but the token list stays empty; using .stream("q") with a fresh fake emits the callback with "abc". Exercise the stream path when testing token callbacks; invoke and stream cover different callback behavior.