Great video as always, have been following you for a while now and always appreciate the knowledge you share for free! You have helped me go from a beginner in LLMs to embedding advanced agents into my applications, cannot thank you enough!
Hey James, really appreaciate your videos. I am currently working on a Project that is supposed to be a RAG Chatbot with multiple Website(s)-Data. The "problem" I am facing is that I need the assistant/bot to be able to make a "knowledge" switch, depending on the context of the conversation. Do you have any idea, how I could implement that? The additional data is stored in Chroma Vectorstore(s). I could also think of using the chunk metadata for that purpose but I wanted to know what you would do/recommend in this kind of situation. Cheers
I tried it with other models too but it didn't work as the documentation suggested this: Use with regular LLMs, not with chat models. Is it because Anthropic models are good at reasoning/writing XML?
yeah they've been trained to do function calling / tool usage via XML tags - I find that comment in the docs odd, as they recommend using with Anthropic models, but Anthropic models at all chat LLMs afaik (maybe Claude 1 is not?) - but in any case, it works well!
you can just edit the prompt that you pass in create_xml_agent to use a ReAct style prompt. That's what I have been doing, otherwise, it often use the tool when it doesnt need to
Great video as always, have been following you for a while now and always appreciate the knowledge you share for free! You have helped me go from a beginner in LLMs to embedding advanced agents into my applications, cannot thank you enough!
Thanks I appreciate the comment, I'm happy it's helped!
Hey, will you be doing tutorials about the new Self-reflective RAG? :)
Hey James, really appreaciate your videos. I am currently working on a Project that is supposed to be a RAG Chatbot with multiple Website(s)-Data. The "problem" I am facing is that I need the assistant/bot to be able to make a "knowledge" switch, depending on the context of the conversation. Do you have any idea, how I could implement that? The additional data is stored in Chroma Vectorstore(s). I could also think of using the chunk metadata for that purpose but I wanted to know what you would do/recommend in this kind of situation. Cheers
I tried it with other models too but it didn't work as the documentation suggested this: Use with regular LLMs, not with chat models. Is it because Anthropic models are good at reasoning/writing XML?
yeah they've been trained to do function calling / tool usage via XML tags - I find that comment in the docs odd, as they recommend using with Anthropic models, but Anthropic models at all chat LLMs afaik (maybe Claude 1 is not?) - but in any case, it works well!
New to GenAI. I wanted to know whether LangChain is used in production ?
Yeah I’ve seen it used, it typically won’t be the *only* thing and there’s often a lot of tweaking required, but it’s used in prod
Can we use XML Agents together with chain of thought prompting (ReAct)? If anyone has tried this out, please share your notebook -- much appreciated!
I don't know about using over many iterations, but for single tool usage this has worked as a replacement of ReAct for me
you can just edit the prompt that you pass in create_xml_agent to use a ReAct style prompt. That's what I have been doing, otherwise, it often use the tool when it doesnt need to