Deepseek R1 Bf16

🚀 DeepSeek-R1

DeepSeek-R1 is a first - generation reasoning model. It addresses the issues of its predecessor, DeepSeek - R1 - Zero, and achieves comparable performance to OpenAI - o1 in math, code, and reasoning tasks. The model and its distilled versions are open - sourced to support the research community.

🚀 Quick Start

You can chat with DeepSeek - R1 on DeepSeek's official website: chat.deepseek.com, and switch on the button "DeepThink". We also provide an OpenAI - Compatible API at DeepSeek Platform: platform.deepseek.com.

✨ Features

• Reinforcement Learning - Based Training: Directly apply large - scale reinforcement learning to the base model without supervised fine - tuning as a preliminary step, enabling the model to explore chain - of - thought for solving complex problems.
• Model Distillation: Demonstrate that the reasoning patterns of larger models can be distilled into smaller models, resulting in better performance.
• Good Performance: Achieve performance comparable to OpenAI - o1 across math, code, and reasoning tasks. Some distilled models outperform OpenAI - o1 - mini in various benchmarks.

📦 Installation

DeepSeek - R1 Models

Please visit [DeepSeek - V3](https://github.com/deepseek - ai/DeepSeek - V3) repo for more information about running DeepSeek - R1 locally.

DeepSeek - R1 - Distill Models

DeepSeek - R1 - Distill models can be utilized in the same manner as Qwen or Llama models. For instance, you can easily start a service using [vLLM](https://github.com/vllm - project/vllm):

vllm serve deepseek - ai/DeepSeek - R1 - Distill - Qwen - 32B --tensor - parallel - size 2 --max - model - len 32768 --enforce - eager

⚠️ Important Note

We recommend setting an appropriate temperature (between 0.5 and 0.7) when running these models, otherwise you may encounter issues with endless repetition or incoherent output.

💻 Usage Examples

Basic Usage

For running DeepSeek - R1 - Distill models, you can use the following command with vLLM:

vllm serve deepseek - ai/DeepSeek - R1 - Distill - Qwen - 32B --tensor - parallel - size 2 --max - model - len 32768 --enforce - eager

📚 Documentation

Introduction

We introduce our first - generation reasoning models, DeepSeek - R1 - Zero and DeepSeek - R1. DeepSeek - R1 - Zero, trained via large - scale reinforcement learning without supervised fine - tuning, demonstrated remarkable reasoning performance but faced challenges like endless repetition. To address these issues and enhance performance, we introduce DeepSeek - R1, which incorporates cold - start data before RL. DeepSeek - R1 achieves performance comparable to OpenAI - o1 across math, code, and reasoning tasks. We have open - sourced DeepSeek - R1 - Zero, DeepSeek - R1, and six dense models distilled from DeepSeek - R1.

Model Summary

Post - Training: Large - Scale Reinforcement Learning on the Base Model

• We directly apply reinforcement learning to the base model without supervised fine - tuning. This allows the model to explore chain - of - thought, resulting in DeepSeek - R1 - Zero, which shows capabilities like self - verification and reflection. It is the first open research to validate that LLMs' reasoning capabilities can be incentivized purely through RL without SFT.
• Our pipeline to develop DeepSeek - R1 includes two RL stages for discovering better reasoning patterns and aligning with human preferences, as well as two SFT stages for the model's reasoning and non - reasoning capabilities.

Distillation: Smaller Models Can Be Powerful Too

• Demonstrate that the reasoning patterns of larger models can be distilled into smaller models, outperforming the reasoning patterns discovered through RL on small models. The open - source DeepSeek - R1 and its API will help the research community distill better smaller models.
• Fine - tune several dense models using the reasoning data generated by DeepSeek - R1. The evaluation results show that the distilled smaller dense models perform well on benchmarks. We open - source distilled 1.5B, 7B, 8B, 14B, 32B, and 70B checkpoints based on Qwen2.5 and Llama3 series.

Model Downloads

DeepSeek - R1 Models

DeepSeek - R1 - Zero & DeepSeek - R1 are trained based on DeepSeek - V3 - Base. For more details regarding the model architecture, please refer to [DeepSeek - V3](https://github.com/deepseek - ai/DeepSeek - V3) repository.

DeepSeek - R1 - Distill Models

DeepSeek - R1 - Distill models are fine - tuned based on open - source models, using samples generated by DeepSeek - R1. We slightly change their configs and tokenizers. Please use our setting to run these models.

Evaluation Results

DeepSeek - R1 - Evaluation

For all our models, the maximum generation length is set to 32,768 tokens. For benchmarks requiring sampling, we use a temperature of $0.6$, a top - p value of $0.95$, and generate 64 responses per query to estimate pass@1.

Distilled Model Evaluation

🔧 Technical Details

• Reinforcement Learning: Directly apply reinforcement learning to the base model without supervised fine - tuning, enabling the model to explore chain - of - thought for solving complex problems.
• Model Distillation: Distill the reasoning patterns of larger models into smaller models, improving the performance of smaller models.

📄 License

This code repository and the model weights are licensed under the [MIT License](https://github.com/deepseek - ai/DeepSeek - R1/blob/main/LICENSE). DeepSeek - R1 series support commercial use, allow for any modifications and derivative works, including, but not limited to, distillation for training other LLMs. Please note that:

• DeepSeek - R1 - Distill - Qwen - 1.5B, DeepSeek - R1 - Distill - Qwen - 7B, DeepSeek - R1 - Distill - Qwen - 14B and DeepSeek - R1 - Distill - Qwen - 32B are derived from Qwen - 2.5 series, which are originally licensed under [Apache 2.0 License](https://huggingface.co/Qwen/Qwen2.5 - 1.5B/blob/main/LICENSE), and now finetuned with 800k samples curated with DeepSeek - R1.
• DeepSeek - R1 - Distill - Llama - 8B is derived from Llama3.1 - 8B - Base and is originally licensed under [llama3.1 license](https://huggingface.co/meta - llama/Llama - 3.1 - 8B/blob/main/LICENSE).
• DeepSeek - R1 - Distill - Llama - 70B is derived from Llama3.3 - 70B - Instruct and is originally licensed under [llama3.3 license](https://huggingface.co/meta - llama/Llama - 3.3 - 70B - Instruct/blob/main/LICENSE).

📬 Contact

If you have any questions, please raise an issue or contact us at service@deepseek.com.