Decide whether to use OlmoEarth embeddings (kNN / linear probe / clustering on frozen features) vs full fine-tuning for a specific Earth-observation task — and emit a runnable Jupyter notebook that extracts embeddings on the user's data. Use whenever the user asks "should I use embeddings or fine-tune", "is my dataset too small to fine-tune", "I have <100 labels — what now", "kNN vs linear probe vs MLP head on OlmoEarth features", "how do I extract OlmoEarth embeddings from this data", "I want to do similarity search / clustering / nearest-neighbor on satellite imagery", "Nano vs Tiny vs Base embeddings", or "generate a notebook to run OlmoEarth on my rslearn dataset". Tr…
