Question and suggestion about the structure of the codebase

[hyeok9855]

I'm not sure, but I believe the goal of this project is to support

1. Various GFN algorithms, including advanced techniques, e.g., Forward-looking, Local search, or Reward augmentation
2. Various GFN applications, e.g., Biochemical Sequence designs, Active learning, LLM finetuning, diffusion sampling or diffusion finetuning.

I want to raise the question of the best structure to achieve this.

My rough idea is the following:

• Abstraction for the key algorithmic components, as we do now.
• Provide a collection of reproduced code snippets for each application (see model zoo of rl4co as a reference)
  • Is the gym for this purpose?
  • We need to reproduce the results of the original code, and it would be best if we gain some improvement in terms of complexity.

Let's discuss this point together!!