What is the difference between on_chain_stream and on_llm_stream?

[SebastianStehle]

I am testing different models, mainly vertex ai, mistral and open ai using streaming.

I have seen different events and would like to understand the difference between these events. At the moment I handle them like this:

for await (const event of stream) {
    const eventType = event.event;

    if (eventType === 'on_llm_start') {
      hasBeenStarted = true;
    } else if (hasBeenStarted && eventType === 'on_llm_stream') {
      const content = event.data?.chunk?.message?.content;

      if (content) {
        // Empty content in the context of OpenAI means that the model is asking for a tool to be invoked via function call.
        result.next({ type: 'chunk', content: normalizedMessageContent(content) });
      }
    } else if (hasBeenStarted && eventType === 'on_chain_stream') {
      const content = event.data?.chunk?.content;

      if (content) {
        result.next({ type: 'chunk', content: normalizedMessageContent(content) });
      }
    }
}

• on_chain_stream is used by Vertex AI.
• on_llm_stream is used by other clients.

[SebastianStehle]

I have seen that sometimes even both events are created. It is very weird and the documentation does not help at all.