GitHub – Deepseek-ai/DeepSeek-V3

We provide DeepSeek-V3, a strong Mixture-of-Experts (MoE) language model with 671B total specifications with 37B triggered for each token. To achieve efficient inference and affordable training, DeepSeek-V3 embraces Multi-head Latent Attention (MLA) and DeepSeekMoE architectures, which were thoroughly confirmed 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 varied and high-quality tokens, followed by Supervised Fine-Tuning and Reinforcement Learning stages to totally harness its capabilities. Comprehensive assessments reveal that DeepSeek-V3 exceeds other open-source models and accomplishes efficiency similar to leading closed-source models. Despite its outstanding performance, DeepSeek-V3 requires only 2.788 M H800 GPU hours for its complete training. In addition, its training process is incredibly stable. 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 effective architecture of DeepSeek-V2, we leader an auxiliary-loss-free strategy for load balancing, which reduces the efficiency degradation that arises from motivating load balancing.
– We examine a Multi-Token Prediction (MTP) objective and show it advantageous to model efficiency. It can likewise be utilized for speculative decoding for reasoning velocity.

Pre-Training: Towards Ultimate Training Efficiency

– We create an FP8 blended accuracy training structure and, for the first time, confirm the feasibility and efficiency of FP8 training on an incredibly massive design.
– Through co-design of algorithms, frameworks, and hardware, we get rid of the interaction bottleneck in cross-node MoE training, nearly accomplishing complete computation-communication overlap.
This significantly improves our training effectiveness and decreases the training costs, enabling us to further scale up the model size without additional overhead.
– At a cost-effective expense of only 2.664 M H800 GPU hours, we finish the pre-training of DeepSeek-V3 on 14.8 T tokens, producing the presently strongest open-source base design. 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 boil down reasoning abilities from the long-Chain-of-Thought (CoT) model, specifically from among the DeepSeek R1 series models, into basic LLMs, particularly DeepSeek-V3. Our pipeline elegantly integrates the confirmation and reflection patterns of R1 into DeepSeek-V3 and significantly enhances its reasoning performance. Meanwhile, we likewise maintain a control over the output design and length of DeepSeek-V3.

3. Model Downloads

The total size of DeepSeek-V3 designs on Hugging Face is 685B, that includes 671B of the Main Model weights and 14B of the Multi-Token Prediction (MTP) Module weights. **

To make sure ideal performance and flexibility, we have partnered with open-source neighborhoods and hardware vendors to supply multiple methods to run the design locally. For detailed guidance, take a look at Section 6: How_to Run_Locally.

For designers looking to dive much deeper, we advise exploring README_WEIGHTS. md for details on the Main Model weights and the Multi-Token Prediction (MTP) Modules. Please note that is currently under active advancement within the community, and we invite your contributions and feedback.

4. Evaluation Results

Base Model

Standard Benchmarks

Best outcomes are shown in strong. Scores with a gap not going beyond 0.3 are considered to be at the exact same level. DeepSeek-V3 accomplishes the very best performance on most criteria, especially on mathematics and code jobs. For more examination details, please examine our paper.

Context Window

Evaluation results on the Needle In A Haystack (NIAH) tests. DeepSeek-V3 carries out well across all context window lengths up to 128K.

Chat Model

Standard Benchmarks (Models bigger than 67B)

All designs are assessed in a configuration that restricts the output length to 8K. Benchmarks containing less than 1000 samples are evaluated several times utilizing differing temperature settings to obtain robust results. DeepSeek-V3 stands as the best-performing open-source design, and likewise displays competitive performance against frontier closed-source models.

Open Ended Generation Evaluation

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

5. Chat Website & API Platform

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

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

6. How to Run Locally

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

DeepSeek-Infer Demo: We provide an easy and light-weight demo for FP8 and BF16 inference.
SGLang: Fully support the DeepSeek-V3 model in both BF16 and FP8 inference modes, with Multi-Token Prediction coming soon.
LMDeploy: Enables efficient FP8 and BF16 reasoning for local and cloud implementation.
TensorRT-LLM: Currently supports BF16 reasoning and INT4/8 quantization, with FP8 assistance coming soon.
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 through SGLang in both BF16 and FP8 modes.
Huawei Ascend NPU: Supports running DeepSeek-V3 on Huawei Ascend devices.
Since FP8 training is natively adopted in our structure, we only provide FP8 weights. If you need BF16 weights for experimentation, you can utilize the offered conversion script to perform the improvement.

Here is an example of transforming FP8 weights to BF16:

Hugging Face’s Transformers has not been directly 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 reasoning folder and install dependences noted in requirements.txt. Easiest way is to utilize a plan supervisor like conda or uv to produce a new virtual environment and set up the dependences.

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

Model Weights Conversion

Convert Hugging Face model weights to a specific format:

Run

Then you can talk with DeepSeek-V3:

Or batch inference on a provided file:

6.2 Inference with SGLang (advised)

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

Notably, SGLang v0.4.1 completely supports running DeepSeek-V3 on both NVIDIA and AMD GPUs, making it an extremely flexible 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 advancement, and development can be tracked in the optimization plan.

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

6.3 Inference with LMDeploy (advised)

LMDeploy, a versatile and high-performance reasoning and serving structure customized for large language designs, now supports DeepSeek-V3. It offers both offline pipeline processing and online release abilities, perfectly integrating with PyTorch-based workflows.

For comprehensive detailed instructions 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 options such as BF16 and INT4/INT8 weight-only. Support for FP8 is currently in progress and will be released quickly. You can access the custom branch of TRTLLM specifically for DeepSeek-V3 assistance through the following link to experience the new features straight: https://github.com/NVIDIA/TensorRT-LLM/tree/deepseek/examples/deepseek_v3.

6.5 Inference with vLLM (advised)

vLLM v0.6.6 supports DeepSeek-V3 inference for FP8 and BF16 modes on both NVIDIA and AMD GPUs. Aside from standard methods, vLLM provides pipeline parallelism permitting you to run this design on multiple makers connected by networks. For detailed assistance, please describe the vLLM guidelines. Please feel complimentary to follow the improvement strategy also.

6.6 Recommended Inference Functionality with AMD GPUs

In collaboration with the AMD team, we have actually achieved Day-One assistance for AMD GPUs utilizing SGLang, with complete compatibility for both FP8 and BF16 precision. For comprehensive guidance, please refer to the SGLang instructions.

6.7 Recommended Inference Functionality with Huawei Ascend NPUs

The MindIE structure from the Huawei Ascend community has actually successfully adapted the BF16 variation of DeepSeek-V3. For step-by-step guidance on Ascend NPUs, please follow the directions here.

7. License

This code repository is certified under the MIT License. Making use of DeepSeek-V3 Base/Chat models is subject to the Model License. DeepSeek-V3 series (including Base and Chat) supports industrial use.