Integration with LLM-based agents (for example through tlr)
#457
-
|
Hello! I want to do some comparison between different algorithms for my agent. However, I also want to give LLMs a try. Is there a goto way to integrate it with This library provides means of fine-tuning huggingface models: https://huggingface.co/docs/trl/en/index. |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 2 replies
-
|
Hey thanks for raising this point. |
Beta Was this translation helpful? Give feedback.
-
|
There are basically two things I am curious about. For one, I wanted to use an LLM to plan the agent's actions. The LLM gets a description of the environment and its action set and is allowed to derive a plan. For example for (Second) That's why I thought about fine-tuning/aligning the LLM to include a way to learn. One could use I am unsure about how to tie it all together though. If it even makes sense to use |
Beta Was this translation helpful? Give feedback.
-
|
Ok I see what you want to do. If I were you, I think I wouldn't try to integrate it into rl-zoo, but rather start from scratch and cherry pick the utilities you want from rl-zoo. |
Beta Was this translation helpful? Give feedback.
Ok I see what you want to do. If I were you, I think I wouldn't try to integrate it into rl-zoo, but rather start from scratch and cherry pick the utilities you want from rl-zoo.
If, on the contrary, you decide to integrate it into rl-zoo3, let 8000 us know here and link your branch, or any other material that could benefit the community.