llms-reasoning
A Python library based on the technique of activation patching to represent the factual knowledge encoded in the latent space of large language models as dynamic knowledge graphs.
