Hi, I am trying out hamilton and burr for a LLM fl...
# hamilton-help
j
Hi, I am trying out hamilton and burr for a LLM flow and I want to represent llm feedback mechanism which is cyclic in nature as part of hamilton data pipeline. Though since its a DAG I don't think this would be straight forward or maybe I should handle this differently
🆒 1
s
Yes this is where we see Burr helping orchestrate something like this. E.g. Burr for the higher order state management and deciding how to flow, e.g. loops & conditional branching. Hamilton for the lower level work that is DAG like, e.g. processing document text.
j
I see, so in burr I can do sub applications too by running application.run() within a function is what I remember
s
I realize this example should also be in the Hamilton repo as well — but this shows one way to use Hamilton + Burr together — https://github.com/DAGWorks-Inc/burr/tree/main/examples/conversational-rag/simple_example
j
yes, that helps but this use case is within a larger burr application so I will have to combine it with https://burr.dagworks.io/concepts/recursion/
👍 1
s
cool — let us know how that works out. We’re seeing an interesting design tension between nesting vs “flattening it”.
so if there are things that would make one or the other simpler — let us know. Otherwise I haven’t pushed the example yet, but we can now also log more details of the internal Hamilton run to the Burr UI (see this note),
j
well it's my first trial so I will follow the rules of the libraries and then form an opinion 😄 sorry not that helpful of a feedback but give me some time to play with it
👍 1
s
Sounds good!
e
Adding in — we’re planning on doing more things to make recursion easier/more natural. Keep us in the loop on what you learn so we can use the feedback to adjust the API!
❤️ 2