Explore training adaptation layers

[hinerm]

I'd love to have a clear process for adapting llms to particular tasks (e.g. core fiji use, but also for community tools) and sharing those models on huggingface.