Fetching the paper…

Megalodon: Efficient LLM Pretraining and Inference with Unlimited Context Length · Around