Task
Text Generation
D
deepseek-ai/DeepSeek-V3
Text Generation·deepseek-ai· 1.1M· 4.1K
transformers
684.5B params
arxiv:2412.19437region:us
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 l
# pull & run locally
pip install mlforge-sdk && mlforge pull deepseek-ai/DeepSeek-V3
pip install mlforge-sdk && mlforge pull deepseek-ai/DeepSeek-V3
Model details
Provider
deepseek-ai
Framework
transformers
Parameters
684.5B
Size
641 GB
Downloads
1.1M
Likes
4.1K
Paper
arXiv:2412.19437
Updated
2025-03-27
About deepseek-ai/DeepSeek-V3
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.