This AI Paper by DeepSeek-AI Introduces DeepSeek-V2: Harnessing Mixture-of-Experts for Enhanced AI Performance

DeepSeek-AI introduces DeepSeek-V2, a language model that reduces computational costs and improves performance. It leverages a Mixture-of-Experts architecture and Multi-head Latent Attention mechanism. DeepSeek-V2 exhibits a significant decrease in training costs, Key-Value cache size, and an increase in generation throughput. Show more