Find answers from the community

Updated last year

Hello, I'm using Anthropic as a llm for

At a glance

The community member is using Anthropic as an LLM (Language Model) for their ReAct agent, with verbose=False set for the agent and its query engine tools. However, they are experiencing an issue where the streaming responses include tags from the default prompt template, and they only need the Action part streamed out. They are asking if they have forgotten to set any attributes.

In the comments, other community members express confusion about the issue, with one mentioning they have used Anthropic as a React agent in the past but never encountered this problem. Another community member asks if there is an alternative way to build a full-stack application with an agent that can use multiple tools and stream the answer out to other OpenAI.

There is no explicitly marked answer in the comments.

Hello, I'm using Anthropic as a llm for my ReAct agent, with verbose=False to the agent and its query engine tools. However, streaming responses include tags from the default prompt template.

I only need the Action part streamed out. Please let me know if I forgot to set any attributes. Many thanks!
Plain Text
ReactAgent.from_tools(
    tools=query_engine_tools,
    llm=main_llm, memory=memory,
    system_prompt=updated_default_system_prompt,
    verbose=False,
)
QueryEngineTool(
    query_engine=user_guide_index.as_query_engine(
        verbose=False,
        similarity_top_k=1,
        service_context=ServiceContext.from_service_context(
            default_service_context,
            llm=sub_llm,
            prompt_helper=PromptHelper(chunk_size_limit=720),
        ),
        response_mode='compact',
        use_async=True,
    ),
    metadata=ToolMetadata(
        name='u_query',
        description='...'
    )
)
Attachment
image.png
L
a
3 comments
uhhh is that just the LLM hallucinating that? thats pretty weird tbh
I've used anthropic as a react agent in the past and never saw that, but it has been a while tbh
Is there an alternative way you can suggest for a full-stack application to build an agent that can use multiple tools and stream the answer out to other OpenAI? @Logan M
Add a reply
Sign up and join the conversation on Discord