Download: v1.0.0v1.0.0 ↓1.0.02025-06-27
+16
Daily
+68
Weekly
104,843
stars
16,746
forks
733
watchers
220
prs
1,239
issues
1
releases

README

Paper Link👁️


 Table of Contents

1. Introduction
2. Model Summary
3. Model Downloads
4. Evaluation Results
5. Chat Website & API Platform
6. How to Run Locally
7. License
8. Citation
9. Contact


 1. Introduction

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. 

  


 2. Model Summary

---

Architecture: Innovative Load Balancing Strategy and Training Objective

- On top of the efficient architecture of DeepSeek-V2, we pioneer an auxiliary-loss-free strategy for load balancing, which minimizes the performance degradation that arises from encouraging load balancing.
-  We investigate a Multi-Token Prediction (MTP) objective and prove it beneficial to model performance. 
    It can also be used for speculative decoding for inference acceleration. 

---

Pre-Training: Towards Ultimate Training Efficiency

- We design an FP8 mixed precision training framework and, for the first time, validate the feasibility and effectiveness of FP8 training on an extremely large-scale model.  
- Through co-design of algorithms, frameworks, and hardware, we overcome the communication bottleneck in cross-node MoE training, nearly achieving full computation-communication overlap.  
  This significantly enhances our training efficiency and reduces the training costs, enabling us to further scale up the model size without additional overhead.  
- At an economical cost of only 2.664M H800 GPU hours, we complete the pre-training of DeepSeek-V3 on 14.8T tokens, producing the currently strongest open-source base model. The subsequent training stages after pre-training require only 0.1M GPU hours.

---

Post-Training: Knowledge Distillation from DeepSeek-R1

-   We introduce an innovative methodology to distill reasoning capabilities from the long-Chain-of-Thought (CoT) model, specifically from one of the DeepSeek R1 series models, into standard LLMs, particularly DeepSeek-V3. Our pipeline elegantly incorporates the verification a

Top issues

Latest release

v1.0.0 v1.0.0 2025-06-27
This release is created for archival purposes and DOI generation.

News & coverage (21)

Hacker NewsDeepSeek-v3.2-Exp▲ 309 · 50 comments · 2025-09-29Hacker NewsDeepSeek-V3▲ 125 · 40 comments · 2024-12-26Hacker NewsDeepSeek-v3 Technical Report [pdf]▲ 26 · 2 comments · 2024-12-26Hacker NewsSuggestion: Read the DeepSeek model license - note the Governing Jurisdiction▲ 6 · 6 comments · 2025-01-31Hacker NewsDeepSeek-v3.2-Exp: Long-Context Efficiency with DeepSeek Sparse Attention [pdf]▲ 4 · 0 comments · 2025-09-29Hacker NewsShow HN: Tokiwi – An online tokenizer for any Hugging Face model▲ 2 · 0 comments · 2025-03-29Hacker NewsDeepSeek Sparse Attention: Boosting Long-Context Efficiency [pdf]▲ 1 · 0 comments · 2025-09-30opentools.aiDeepSeek v3: The Open-Source LLM Challenger Making Waves2026-09-12MashableDeepSeek v3.2: What's new and how does it compare to ChatGPT?2025-12-03githubDeepSeek-V3-0324 is now generally available in GitHub Models2025-04-08
11 moreCNETDeepSeek V3 Is Now Reportedly the Best Nonreasoning AI Model2025-03-25githubDeepSeek-V3 is now generally available in GitHub Models2025-03-07InfoQDeepSeek Open-Sources DeepSeek-V3, a 671B Parameter Mixture of Experts LLM2025-01-21thehinduChinese AI firm releases DeepSeek V3, a new leader in open-source AI models2024-12-27GIGAZINEChinese AI company DeepSeek releases 'DeepSeek-V3', an AI model comparable to GPT-4o, with an astounding 671 billion parameters2024-12-27TechCrunchDeepSeek’s new AI model appears to be one of the best ‘open’ challengers yet2024-12-26VentureBeatDeepSeek-V3, ultra-large open-source AI, outperforms Llama and Qwen on launch2024-12-26huggingfaceunsloth/DeepSeek-V3-bf162012-11-05MarkTechPostDeepSeek AI Researchers Introduce Engram: A Conditional Memory Axis For Sparse LLMs2026-01-14METRDetails about METR's preliminary evaluation of DeepSeek-R12025-03-05Gonzo MLDeepSeek-V3: Technical Details2025-02-06

Snapshot: 2026-11-16T09:23:14Z · Data license Apache-2.0 · JSON