3 ms·Megalodon: Efficient LLM Pretraining and Inference with Unlimited Context Length4 points by tosh 3y ago