does LangGraph provide any support to stream the LLM's output？

[qiankunli]

In many cases, a LangGraph node is an LLM call. For such nodes, does LangGraph provide any support to stream the LLM's output through methods like graph.stream()? This way, we can stream the responses of a multi-step/multi-turn LLM inference task to the front end, thereby improving user experience.

[hinthornw]

Yes.

https://langchain-ai.github.io/langgraph/how-tos/#streaming

https://langchain-ai.github.io/langgraph/how-tos/streaming-tokens/

https://langchain-ai.github.io/langgraph/how-tos/streaming-tokens-without-langchain/

Is there a specific pattern in mind you find unaddressed?

[qiankunli]

@hinthornw Can you briefly explain the principle behind this?
How does LangGraph know that a node has executed an LLM call, and how does it transfer the LLM call's stream to LangGraph's stream call?