XythicK/DeepSeek-R1-GGUF

1---2base_model:3- deepseek-ai/DeepSeek-R14---5# DeepSeek-R16<!-- markdownlint-disable first-line-h1 -->7<!-- markdownlint-disable html -->8<!-- markdownlint-disable no-duplicate-header -->9 10<div align="center">11  <img src="https://github.com/deepseek-ai/DeepSeek-V2/blob/main/figures/logo.svg?raw=true" width="60%" alt="DeepSeek-V3" />12</div>13<hr>14<div align="center" style="line-height: 1;">15  <a href="https://www.deepseek.com/" target="_blank" style="margin: 2px;">16    <img alt="Homepage" src="https://github.com/deepseek-ai/DeepSeek-V2/blob/main/figures/badge.svg?raw=true" style="display: inline-block; vertical-align: middle;"/>17  </a>18  <a href="https://chat.deepseek.com/" target="_blank" style="margin: 2px;">19    <img alt="Chat" src="https://img.shields.io/badge/🤖%20Chat-DeepSeek%20R1-536af5?color=536af5&logoColor=white" style="display: inline-block; vertical-align: middle;"/>20  </a>21  <a href="https://huggingface.co/deepseek-ai" target="_blank" style="margin: 2px;">22    <img alt="Hugging Face" src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-DeepSeek%20AI-ffc107?color=ffc107&logoColor=white" style="display: inline-block; vertical-align: middle;"/>23  </a>24</div>25 26<div align="center" style="line-height: 1;">27  <a href="https://discord.gg/Tc7c45Zzu5" target="_blank" style="margin: 2px;">28    <img alt="Discord" src="https://img.shields.io/badge/Discord-DeepSeek%20AI-7289da?logo=discord&logoColor=white&color=7289da" style="display: inline-block; vertical-align: middle;"/>29  </a>30  <a href="https://github.com/deepseek-ai/DeepSeek-V2/blob/main/figures/qr.jpeg?raw=true" target="_blank" style="margin: 2px;">31    <img alt="Wechat" src="https://img.shields.io/badge/WeChat-DeepSeek%20AI-brightgreen?logo=wechat&logoColor=white" style="display: inline-block; vertical-align: middle;"/>32  </a>33  <a href="https://twitter.com/deepseek_ai" target="_blank" style="margin: 2px;">34    <img alt="Twitter Follow" src="https://img.shields.io/badge/Twitter-deepseek_ai-white?logo=x&logoColor=white" style="display: inline-block; vertical-align: middle;"/>35  </a>36</div>37 38<div align="center" style="line-height: 1;">39  <a href="https://github.com/deepseek-ai/DeepSeek-R1/blob/main/LICENSE-CODE" style="margin: 2px;">40    <img alt="Code License" src="https://img.shields.io/badge/Code_License-MIT-f5de53?&color=f5de53" style="display: inline-block; vertical-align: middle;"/>41  </a>42  <a href="https://github.com/deepseek-ai/DeepSeek-R1/blob/main/LICENSE-MODEL" style="margin: 2px;">43    <img alt="Model License" src="https://img.shields.io/badge/Model_License-Model_Agreement-f5de53?&color=f5de53" style="display: inline-block; vertical-align: middle;"/>44  </a>45</div>46 47 48<p align="center">49  <a href="https://github.com/deepseek-ai/DeepSeek-R1/blob/main/DeepSeek_R1.pdf"><b>Paper Link</b>👁️</a>50</p>51 52 53## 1. Introduction54 55We introduce our first-generation reasoning models, DeepSeek-R1-Zero and DeepSeek-R1. 56DeepSeek-R1-Zero, a model trained via large-scale reinforcement learning (RL) without supervised fine-tuning (SFT) as a preliminary step, demonstrated remarkable performance on reasoning.57With RL, DeepSeek-R1-Zero naturally emerged with numerous powerful and interesting reasoning behaviors.58However, DeepSeek-R1-Zero encounters challenges such as endless repetition, poor readability, and language mixing. To address these issues and further enhance reasoning performance,59we introduce DeepSeek-R1, which incorporates cold-start data before RL.60DeepSeek-R1 achieves performance comparable to OpenAI-o1 across math, code, and reasoning tasks. 61To support the research community, we have open-sourced DeepSeek-R1-Zero, DeepSeek-R1, and six dense models distilled from DeepSeek-R1 based on Llama and Qwen. DeepSeek-R1-Distill-Qwen-32B outperforms OpenAI-o1-mini across various benchmarks, achieving new state-of-the-art results for dense models.62 63**NOTE: Before running DeepSeek-R1 series models locally, we kindly recommend reviewing the [Usage Recommendation](#usage-recommendations) section.**64 65<p align="center">66  <img width="80%" src="figures/benchmark.jpg">67</p>68 69## 2. Model Summary70 71---72 73**Post-Training: Large-Scale Reinforcement Learning on the Base Model**74 75-  We directly apply reinforcement learning (RL) to the base model without relying on supervised fine-tuning (SFT) as a preliminary step. This approach allows the model to explore chain-of-thought (CoT) for solving complex problems, resulting in the development of DeepSeek-R1-Zero. DeepSeek-R1-Zero demonstrates capabilities such as self-verification, reflection, and generating long CoTs, marking a significant milestone for the research community. Notably, it is the first open research to validate that reasoning capabilities of LLMs can be incentivized purely through RL, without the need for SFT. This breakthrough paves the way for future advancements in this area.76 77-   We introduce our pipeline to develop DeepSeek-R1. The pipeline incorporates two RL stages aimed at discovering improved reasoning patterns and aligning with human preferences, as well as two SFT stages that serve as the seed for the model's reasoning and non-reasoning capabilities.78    We believe the pipeline will benefit the industry by creating better models. 79 80---81 82**Distillation: Smaller Models Can Be Powerful Too**83 84-  We demonstrate that the reasoning patterns of larger models can be distilled into smaller models, resulting in better performance compared to the reasoning patterns discovered through RL on small models. The open source DeepSeek-R1, as well as its API, will benefit the research community to distill better smaller models in the future. 85- Using the reasoning data generated by DeepSeek-R1, we fine-tuned several dense models that are widely used in the research community. The evaluation results demonstrate that the distilled smaller dense models perform exceptionally well on benchmarks. We open-source distilled 1.5B, 7B, 8B, 14B, 32B, and 70B checkpoints based on Qwen2.5 and Llama3 series to the community.86 87## 3. Model Downloads88 89### DeepSeek-R1 Models90 91<div align="center">92 93| **Model** | **#Total Params** | **#Activated Params** | **Context Length** | **Download** |94| :------------: | :------------: | :------------: | :------------: | :------------: |95| DeepSeek-R1-Zero | 671B | 37B | 128K   | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Zero)   |96| DeepSeek-R1   | 671B | 37B |  128K   | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1)   |97 98</div>99 100DeepSeek-R1-Zero & DeepSeek-R1 are trained based on DeepSeek-V3-Base. 101For more details regarding the model architecture, please refer to [DeepSeek-V3](https://github.com/deepseek-ai/DeepSeek-V3) repository.102 103### DeepSeek-R1-Distill Models104 105<div align="center">106 107| **Model** | **Base Model** | **Download** |108| :------------: | :------------: | :------------: |109| DeepSeek-R1-Distill-Qwen-1.5B  | [Qwen2.5-Math-1.5B](https://huggingface.co/Qwen/Qwen2.5-Math-1.5B) | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B)   |110| DeepSeek-R1-Distill-Qwen-7B  | [Qwen2.5-Math-7B](https://huggingface.co/Qwen/Qwen2.5-Math-7B) | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B)   |111| DeepSeek-R1-Distill-Llama-8B  | [Llama-3.1-8B](https://huggingface.co/meta-llama/Llama-3.1-8B) | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-8B)   |112| DeepSeek-R1-Distill-Qwen-14B   | [Qwen2.5-14B](https://huggingface.co/Qwen/Qwen2.5-14B) | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-14B)   |113|DeepSeek-R1-Distill-Qwen-32B  | [Qwen2.5-32B](https://huggingface.co/Qwen/Qwen2.5-32B) | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B)   |114| DeepSeek-R1-Distill-Llama-70B  | [Llama-3.3-70B-Instruct](https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct) | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-70B)   |115 116</div>117 118DeepSeek-R1-Distill models are fine-tuned based on open-source models, using samples generated by DeepSeek-R1.119We slightly change their configs and tokenizers. Please use our setting to run these models.120 121## 4. Evaluation Results122 123### DeepSeek-R1-Evaluation124 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@1.125<div align="center">126 127 128| Category | Benchmark (Metric) | Claude-3.5-Sonnet-1022 | GPT-4o 0513 | DeepSeek V3 | OpenAI o1-mini | OpenAI o1-1217 | DeepSeek R1 |129|----------|-------------------|----------------------|------------|--------------|----------------|------------|--------------|130| | Architecture | - | - | MoE | - | - | MoE |131| | # Activated Params | - | - | 37B | - | - | 37B |132| | # Total Params | - | - | 671B | - | - | 671B |133| English | MMLU (Pass@1) | 88.3 | 87.2 | 88.5 | 85.2 | **91.8** | 90.8 |134| | MMLU-Redux (EM) | 88.9 | 88.0 | 89.1 | 86.7 | - | **92.9** |135| | MMLU-Pro (EM) | 78.0 | 72.6 | 75.9 | 80.3 | - | **84.0** |136| | DROP (3-shot F1) | 88.3 | 83.7 | 91.6 | 83.9 | 90.2 | **92.2** |137| | IF-Eval (Prompt Strict) | **86.5** | 84.3 | 86.1 | 84.8 | - | 83.3 |138| | GPQA-Diamond (Pass@1) | 65.0 | 49.9 | 59.1 | 60.0 | **75.7** | 71.5 |139| | SimpleQA (Correct) | 28.4 | 38.2 | 24.9 | 7.0 | **47.0** | 30.1 |140| | FRAMES (Acc.) | 72.5 | 80.5 | 73.3 | 76.9 | - | **82.5** |141| | AlpacaEval2.0 (LC-winrate) | 52.0 | 51.1 | 70.0 | 57.8 | - | **87.6** |142| | ArenaHard (GPT-4-1106) | 85.2 | 80.4 | 85.5 | 92.0 | - | **92.3** |143| Code | LiveCodeBench (Pass@1-COT) | 33.8 | 34.2 | - | 53.8 | 63.4 | **65.9** |144| | Codeforces (Percentile) | 20.3 | 23.6 | 58.7 | 93.4 | **96.6** | 96.3 |145| | Codeforces (Rating) | 717 | 759 | 1134 | 1820 | **2061** | 2029 |146| | SWE Verified (Resolved) | **50.8** | 38.8 | 42.0 | 41.6 | 48.9 | 49.2 |147| | Aider-Polyglot (Acc.) | 45.3 | 16.0 | 49.6 | 32.9 | **61.7** | 53.3 |148| Math | AIME 2024 (Pass@1) | 16.0 | 9.3 | 39.2 | 63.6 | 79.2 | **79.8** |149| | MATH-500 (Pass@1) | 78.3 | 74.6 | 90.2 | 90.0 | 96.4 | **97.3** |150| | CNMO 2024 (Pass@1) | 13.1 | 10.8 | 43.2 | 67.6 | - | **78.8** |151| Chinese | CLUEWSC (EM) | 85.4 | 87.9 | 90.9 | 89.9 | - | **92.8** |152| | C-Eval (EM) | 76.7 | 76.0 | 86.5 | 68.9 | - | **91.8** |153| | C-SimpleQA (Correct) | 55.4 | 58.7 | **68.0** | 40.3 | - | 63.7 |154 155</div>156 157 158### Distilled Model Evaluation159 160 161<div align="center">162 163| Model                                    | AIME 2024 pass@1 | AIME 2024 cons@64 | MATH-500 pass@1 | GPQA Diamond pass@1 | LiveCodeBench pass@1 | CodeForces rating |164|------------------------------------------|------------------|-------------------|-----------------|----------------------|----------------------|-------------------|165| GPT-4o-0513                          | 9.3              | 13.4              | 74.6            | 49.9                 | 32.9                 | 759               |166| Claude-3.5-Sonnet-1022             | 16.0             | 26.7                 | 78.3            | 65.0                 | 38.9                 | 717               |167| o1-mini                              | 63.6             | 80.0              | 90.0            | 60.0                 | 53.8                 | **1820**          |168| QwQ-32B-Preview                              | 44.0             | 60.0                 | 90.6            | 54.5               | 41.9                 | 1316              |169| DeepSeek-R1-Distill-Qwen-1.5B       | 28.9             | 52.7              | 83.9            | 33.8                 | 16.9                 | 954               |170| DeepSeek-R1-Distill-Qwen-7B          | 55.5             | 83.3              | 92.8            | 49.1                 | 37.6                 | 1189              |171| DeepSeek-R1-Distill-Qwen-14B         | 69.7             | 80.0              | 93.9            | 59.1                 | 53.1                 | 1481              |172| DeepSeek-R1-Distill-Qwen-32B        | **72.6**         | 83.3              | 94.3            | 62.1                 | 57.2                 | 1691              |173| DeepSeek-R1-Distill-Llama-8B         | 50.4             | 80.0              | 89.1            | 49.0                 | 39.6                 | 1205              |174| DeepSeek-R1-Distill-Llama-70B        | 70.0             | **86.7**          | **94.5**        | **65.2**             | **57.5**             | 1633              |175 176</div>177 178 179## 5. Chat Website & API Platform180You can chat with DeepSeek-R1 on DeepSeek's official website: [chat.deepseek.com](https://chat.deepseek.com), and switch on the button "DeepThink"181 182We also provide OpenAI-Compatible API at DeepSeek Platform: [platform.deepseek.com](https://platform.deepseek.com/)183 184## 6. How to Run Locally185 186### DeepSeek-R1 Models187 188Please visit [DeepSeek-V3](https://github.com/deepseek-ai/DeepSeek-V3) repo for more information about running DeepSeek-R1 locally.189 190### DeepSeek-R1-Distill Models191 192DeepSeek-R1-Distill models can be utilized in the same manner as Qwen or Llama models.193 194For instance, you can easily start a service using [vLLM](https://github.com/vllm-project/vllm):195 196```shell197vllm serve deepseek-ai/DeepSeek-R1-Distill-Qwen-32B --tensor-parallel-size 2 --max-model-len 32768 --enforce-eager198```199 200You can also easily start a service using [SGLang](https://github.com/sgl-project/sglang)201 202```bash203python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-R1-Distill-Qwen-32B --trust-remote-code --tp 2204```205 206### Usage Recommendations207 208**We recommend adhering to the following configurations when utilizing the DeepSeek-R1 series models, including benchmarking, to achieve the expected performance:**209