ai-agents-llm-streaming-llmchain-stream-returns-the-full-response-instead--321aeba8

THEN Create a subclass of LLMChain that overrides the stream method. In the override, call self.prep_prompts() to prepare inputs, then yield from self.llm.stream(input=prompts[0], config=config). This bypasses the base Chain's fallback which does not support streaming.

class MyChain(LLMChain):
    def stream(self, input, config=None, run_manager=None, **kwargs):
        prompts, stop = self.prep_prompts([input], run_manager=run_manager)
        yield from self.llm.stream(input=prompts[0], config=config, **kwargs)