Megalodon: Efficient LLM Pretraining and Inference with Unlimited Context Length Paper • 2404.08801 • Published Apr 12 • 63