DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model

TL;DR

DeepSeek-V2 is an advanced Mixture-of-Experts language model that offers high performance with economical training, efficient inference, and innovative architectures, supporting very long contexts and achieving top-tier results among open-source models.

Contribution

The paper introduces DeepSeek-V2, featuring novel architectures like MLA and DeepSeekMoE, enabling strong, cost-effective training and efficient inference for large-scale language modeling.

Findings

Achieves superior performance compared to previous models.

Reduces training costs by 42.5%.

Increases generation throughput by 5.76 times.

Abstract

We present DeepSeek-V2, a strong Mixture-of-Experts (MoE) language model characterized by economical training and efficient inference. It comprises 236B total parameters, of which 21B are activated for each token, and supports a context length of 128K tokens. DeepSeek-V2 adopts innovative architectures including Multi-head Latent Attention (MLA) and DeepSeekMoE. MLA guarantees efficient inference through significantly compressing the Key-Value (KV) cache into a latent vector, while DeepSeekMoE enables training strong models at an economical cost through sparse computation. Compared with DeepSeek 67B, DeepSeek-V2 achieves significantly stronger performance, and meanwhile saves 42.5% of training costs, reduces the KV cache by 93.3%, and boosts the maximum generation throughput to 5.76 times. We pretrain DeepSeek-V2 on a high-quality and multi-source corpus consisting of 8.1T tokens, and further perform Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) to fully unlock its potential. Evaluation results show that, even with only 21B activated parameters, DeepSeek-V2 and its chat versions still achieve top-tier performance among open-source models.