imodelsX
Interpretable prompting and models for NLP
...Finetune a single linear layer on top of LLM embeddings. Use these just a like a sci-kit-learn model. During training, they fit better features via LLMs, but at test-time, they are extremely fast and completely transparent.