Megalodon: Efficient LLM Pretraining and Inference with Unlimited Context Length

arXiv:2404.08801v1 Announce Type: new
Abstract: The quadratic complexity and weak length extrapolation of Transformers limits their ability to scale to long sequences, and while sub-quadratic solutions like linear attention and state space models exist, they empirically underperform Transformers in pretraining efficiency and downstream task accuracy. We introduce Megalodon, a neural architecture for efficient sequence modeling with unlimited context length. Megalodon inherits the architecture of Mega (exponential moving average with gated attention), and further introduces multiple technical components to improve its …

arxiv context cs.cl cs.lg inference llm pretraining type