Mesa-LLM developpement (Colin's version) #2773
colinfrisch
started this conversation in
Ideas
Replies: 1 comment 6 replies
-
Hey, after a few discussions with @wang-boyu, @jackiekazil and @sanika-n, we are starting to have a solid common roadmap, topics and other insights to work on. |
Beta Was this translation helpful? Give feedback.
6 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Hello everyone !
We can discuss here about the Mesa-LLM project that I'll be developing with @sanika-n under the mentorship of @wang-boyu and @jackiekazil.
My proposal, some more detailed code examples for the project as well as some refining that I did on the subject are available in my public abm-mesa-experimentations repository.
The general idea to it is to build a derived LLMAgent(Agent) class, with three main attributes :
self.memory
,self.llm
andself.strategy
, each depending on external classes to make it as modular as possible. I also chose to separate paradigm (basically the behavior of the agent, meaning the way that it observes and acts) and the reasoning (the way the agent plans its actions with the LLM). All this is explained in more details in the proposal and in the refinement.These ideas are not necessarily the exact way that Mesa-LLM is going to be developed : @sanika-n's proposal tackles this project in ways that are sometimes different but also very interesting, which is why we want to discuss it to get the better of both worlds, and also take into account other ideas coming from everyone !
If you have a new topic of discussion on Mesa-LLM that you would want to explore, please feel free to create a new discussion in the Mesa-LLM repository discussions. And if you want to be involved in this project, you can join our Matrix chat room in which questions, issues, and ideas can be (informally) discussed.
Beta Was this translation helpful? Give feedback.
All reactions