[PartiallyTyped]

The most interesting thing in this whole saga is that decoder only models (aka causal transformers like GPT) are as effective as they are.