GitHub – Deepseek-ai/DeepSeek-V3

We present DeepSeek-V3, a strong Mixture-of-Experts (MoE) language design with 671B overall specifications with 37B triggered for each token. To accomplish efficient reasoning and economical training, DeepSeek-V3 adopts Multi-head Latent Attention (MLA) and DeepSeekMoE architectures, which were thoroughly verified in DeepSeek-V2. Furthermore, DeepSeek-V3 leaders an auxiliary-loss-free strategy for load balancing and sets a multi-token prediction training objective for stronger efficiency. We pre-train DeepSeek-V3 on 14.8 trillion varied and premium tokens, followed by Supervised Fine-Tuning and Reinforcement Learning stages to completely harness its capabilities. Comprehensive examinations expose that DeepSeek-V3 exceeds other open-source designs and achieves performance comparable to leading closed-source designs. Despite its exceptional efficiency, DeepSeek-V3 requires just 2.788 M H800 GPU hours for its complete training. In addition, its training process is remarkably steady. Throughout the whole 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 technique for load balancing, which lessens the performance deterioration that develops from encouraging load balancing.
– We investigate a Multi-Token Prediction (MTP) objective and prove it helpful to model performance. It can likewise be used for speculative decoding for reasoning acceleration.

Pre-Training: Towards Ultimate Training Efficiency

– We create an FP8 blended accuracy training structure and, for the very first time, validate the expediency and efficiency of FP8 training on an extremely massive model.
– Through co-design of algorithms, frameworks, and hardware, we conquer the interaction bottleneck in cross-node MoE training, nearly attaining full computation-communication overlap.
This considerably improves our training efficiency and minimizes the training expenses, enabling us to even more scale up the design size without additional overhead.
– At an economical cost of only 2.664 M H800 GPU hours, we complete the pre-training of DeepSeek-V3 on 14.8 T tokens, producing the presently strongest open-source base model. The subsequent training stages after pre-training require just 0.1 M GPU hours.

Post-Training: Knowledge Distillation from DeepSeek-R1

– We introduce an innovative method to distill reasoning abilities from the long-Chain-of-Thought (CoT) model, particularly from one of the DeepSeek R1 series models, into standard LLMs, particularly DeepSeek-V3. Our pipeline elegantly incorporates the confirmation and reflection patterns of R1 into DeepSeek-V3 and especially enhances its reasoning efficiency. Meanwhile, we likewise maintain a control over the output design and length of DeepSeek-V3.

3. Model Downloads

The overall size of DeepSeek-V3 models on Hugging Face is 685B, which consists of 671B of the Main Model weights and 14B of the Multi-Token Prediction (MTP) Module weights. **

To make sure optimal performance and flexibility, we have actually partnered with and hardware suppliers to provide several ways to run the design in your area. For step-by-step guidance, take a look at Section 6: How_to Run_Locally.

For developers seeking to dive deeper, we suggest checking out README_WEIGHTS. md for information on the Main Model weights and the Multi-Token Prediction (MTP) Modules. Please note that MTP assistance is presently under active development within the neighborhood, and we invite your contributions and feedback.

4. Evaluation Results

Base Model

Standard Benchmarks

Best results are shown in strong. Scores with a gap not surpassing 0.3 are thought about to be at the same level. DeepSeek-V3 achieves the finest efficiency on a lot of benchmarks, particularly on mathematics and code tasks. For more assessment details, please examine our paper.

Context Window

Evaluation results on the Needle In A Haystack (NIAH) tests. DeepSeek-V3 performs well throughout all context window lengths approximately 128K.

Chat Model

Standard Benchmarks (Models bigger than 67B)

All designs are evaluated in a setup that limits the output length to 8K. Benchmarks including fewer than 1000 samples are tested numerous times using varying temperature settings to obtain robust last results. DeepSeek-V3 stands as the best-performing open-source design, and also displays competitive efficiency versus frontier closed-source models.

Open Ended Generation Evaluation

English open-ended conversation evaluations. For AlpacaEval 2.0, we utilize the length-controlled win rate as the metric.

5. Chat Website & API Platform

You can chat with DeepSeek-V3 on DeepSeek’s main website: chat.deepseek.com

We also supply OpenAI-Compatible API at DeepSeek Platform: platform.deepseek.com

6. How to Run Locally

DeepSeek-V3 can be deployed locally utilizing the following hardware and open-source neighborhood software application:

DeepSeek-Infer Demo: We offer a simple and lightweight demonstration for FP8 and BF16 inference.
SGLang: Fully support the DeepSeek-V3 model in both BF16 and FP8 reasoning modes, with Multi-Token Prediction coming soon.
LMDeploy: Enables efficient FP8 and BF16 inference for local and cloud implementation.
TensorRT-LLM: Currently supports BF16 reasoning and INT4/8 quantization, with FP8 support coming quickly.
vLLM: Support DeepSeek-V3 design with FP8 and BF16 modes for tensor parallelism and pipeline parallelism.
AMD GPU: Enables running the DeepSeek-V3 model on AMD GPUs by means of SGLang in both BF16 and FP8 modes.
Huawei Ascend NPU: Supports running DeepSeek-V3 on Huawei Ascend gadgets.
Since FP8 training is natively adopted in our structure, we only provide FP8 weights. If you need BF16 weights for experimentation, you can use the offered conversion script to perform the transformation.

Here is an example of converting FP8 weights to BF16:

Hugging Face’s Transformers has actually not been straight supported yet. **

6.1 Inference with DeepSeek-Infer Demo (example just)

System Requirements

Note

Linux with Python 3.10 only. Mac and Windows are not supported.

Dependencies:

Model Weights & Demo Code Preparation

First, clone our DeepSeek-V3 GitHub repository:

Navigate to the inference folder and set up dependences noted in requirements.txt. Easiest way is to use a bundle manager like conda or uv to create a new virtual environment and install the reliances.

Download the design weights from Hugging Face, and put them into/ path/to/DeepSeek-V 3 folder.

Model Weights Conversion

Convert Hugging Face model weights to a particular format:

Run

Then you can talk with DeepSeek-V3:

Or batch inference on a provided file:

6.2 Inference with SGLang (suggested)

SGLang presently supports MLA optimizations, DP Attention, FP8 (W8A8), FP8 KV Cache, and Torch Compile, delivering modern latency and throughput efficiency among open-source frameworks.

Notably, SGLang v0.4.1 fully supports running DeepSeek-V3 on both NVIDIA and AMD GPUs, making it an extremely versatile and robust service.

SGLang likewise supports multi-node tensor parallelism, enabling you to run this model on numerous network-connected machines.

Multi-Token Prediction (MTP) remains in development, and progress can be tracked in the optimization plan.

Here are the launch directions from the SGLang group: https://github.com/sgl-project/sglang/tree/main/benchmark/deepseek_v3

6.3 Inference with LMDeploy (advised)

LMDeploy, a flexible and high-performance reasoning and serving framework customized for large language models, now supports DeepSeek-V3. It offers both offline pipeline processing and online implementation abilities, flawlessly incorporating with PyTorch-based workflows.

For detailed step-by-step guidelines on running DeepSeek-V3 with LMDeploy, please refer to here: InternLM/lmdeploy # 2960

6.4 Inference with TRT-LLM (suggested)

TensorRT-LLM now supports the DeepSeek-V3 design, using precision choices such as BF16 and INT4/INT8 weight-only. Support for FP8 is currently in progress and will be launched quickly. You can access the customized branch of TRTLLM particularly for DeepSeek-V3 assistance through the following link to experience the new functions directly: https://github.com/NVIDIA/TensorRT-LLM/tree/deepseek/examples/deepseek_v3.

6.5 Inference with vLLM (suggested)

vLLM v0.6.6 supports DeepSeek-V3 inference for FP8 and BF16 modes on both NVIDIA and AMD GPUs. Aside from standard techniques, vLLM offers pipeline parallelism enabling you to run this model on several devices linked by networks. For in-depth guidance, please refer to the vLLM directions. Please feel free to follow the improvement strategy also.

6.6 Recommended Inference Functionality with AMD GPUs

In cooperation with the AMD group, we have actually attained Day-One assistance for AMD GPUs using SGLang, with full compatibility for both FP8 and BF16 accuracy. For in-depth assistance, please refer to the SGLang directions.

6.7 Recommended Inference Functionality with Huawei Ascend NPUs

The MindIE structure from the Huawei Ascend neighborhood has actually successfully adjusted the BF16 version of DeepSeek-V3. For detailed guidance on Ascend NPUs, please follow the directions here.

7. License

This code repository is licensed under the MIT License. The usage of DeepSeek-V3 Base/Chat models is subject to the Model License. DeepSeek-V3 series (consisting of Base and Chat) supports commercial use.