There was a recent-ish paper[0] from Sakana AI about baking facts from a document corpus into a LoRA adapter. Claimed near perfect recall on very large needle-i

There was a recent-ish paper[0] from Sakana AI about baking facts from a document corpus into a LoRA adapter. Claimed near perfect recall on very large needle-in-haystack testing. Haven’t tried it myself though. [0]: https://sakana.ai/doc-to-lora/ https://sakana.ai/doc-to-lora/