Megalodon: Efficient LLM Pretraining and Inference with Unlimited Context Length (via sanxiyn) — discussion

#ai