DeepSeek-V3 Technical Report

TL;DR

DeepSeek-V3 is a large, efficient Mixture-of-Experts language model with 671 billion parameters, achieving high performance through innovative architectures and training strategies while maintaining cost-effectiveness and training stability.

Contribution

Introduction of DeepSeek-V3 with novel load balancing and multi-token prediction strategies, setting new standards in large-scale language model training and performance.

Findings

Outperforms open-source models in benchmarks

Achieves comparable results to top closed-source models

Requires only 2.788M GPU hours for training

Abstract

We present DeepSeek-V3, a strong Mixture-of-Experts (MoE) language model with 671B total parameters with 37B activated for each token. To achieve efficient inference and cost-effective training, DeepSeek-V3 adopts Multi-head Latent Attention (MLA) and DeepSeekMoE architectures, which were thoroughly validated in DeepSeek-V2. Furthermore, DeepSeek-V3 pioneers an auxiliary-loss-free strategy for load balancing and sets a multi-token prediction training objective for stronger performance. We pre-train DeepSeek-V3 on 14.8 trillion diverse and high-quality tokens, followed by Supervised Fine-Tuning and Reinforcement Learning stages to fully harness its capabilities. Comprehensive evaluations reveal that DeepSeek-V3 outperforms other open-source models and achieves performance comparable to leading closed-source models. Despite its excellent performance, DeepSeek-V3 requires only 2.788M H800 GPU hours for its full training. In addition, its training process is remarkably stable. Throughout the entire training process, we did not experience any irrecoverable loss spikes or perform any rollbacks. The model checkpoints are available at https://github.com/deepseek-ai/DeepSeek-V3.