README.md

---
pipeline_tag: image-text-to-text
language:
- multilingual
tags:
- deepseek
- vision-language
- ocr
- custom_code
license: mit
library_name: transformers
---
<div align="center">

# Neural Metrics · Compressing pages into far fewer vision tokens.

<img src="https://img.shields.io/badge/Neural%20Metrics-document%20extraction-4F46E5?style=for-the-badge" alt="Neural Metrics" />
<img src="https://img.shields.io/badge/fork%20of-deepseek--ai%2FDeepSeek--OCR-2563EB?style=flat-square" alt="fork" />

</div>

DeepSeek-OCR explores optical context compression - representing a full page with an order of magnitude fewer tokens than conventional pipelines. For long multi-page documents, token budget is the binding constraint, so this matters commercially.

**We use it for:** long-document processing where context length is the bottleneck - cost-per-page experiments.

> ### Attribution
> This is an **unmodified fork** of [`deepseek-ai/DeepSeek-OCR`](https://huggingface.co/deepseek-ai/DeepSeek-OCR), created by the [Qwen team](https://huggingface.co/Qwen).
> All weights, files and behaviour are identical to upstream — we rehost it so our experiments stay
> reproducible and version-pinned. The original license and all credit remain with the Qwen team.
> If you want the canonical model, please use [the original](https://huggingface.co/deepseek-ai/DeepSeek-OCR).

---

<details>
<summary><b>Original model card from deepseek-ai/DeepSeek-OCR</b> (click to expand)</summary>

<div align="center">
  <img src="https://github.com/deepseek-ai/DeepSeek-V2/blob/main/figures/logo.svg?raw=true" width="60%" alt="DeepSeek AI" />
</div>
<hr>
<div align="center">
  <a href="https://www.deepseek.com/" target="_blank">
    <img alt="Homepage" src="https://github.com/deepseek-ai/DeepSeek-V2/blob/main/figures/badge.svg?raw=true" />
  </a>
  <a href="https://huggingface.co/deepseek-ai/DeepSeek-OCR" target="_blank">
    <img alt="Hugging Face" src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-DeepSeek%20AI-ffc107?color=ffc107&logoColor=white" />
  </a>

</div>

<div align="center">

  <a href="https://discord.gg/Tc7c45Zzu5" target="_blank">
    <img alt="Discord" src="https://img.shields.io/badge/Discord-DeepSeek%20AI-7289da?logo=discord&logoColor=white&color=7289da" />
  </a>
  <a href="https://twitter.com/deepseek_ai" target="_blank">
    <img alt="Twitter Follow" src="https://img.shields.io/badge/Twitter-deepseek_ai-white?logo=x&logoColor=white" />
  </a>

</div>



<p align="center">
  <a href="https://github.com/deepseek-ai/DeepSeek-OCR"><b>🌟 Github</b></a> |
  <a href="https://huggingface.co/deepseek-ai/DeepSeek-OCR"><b>📥 Model Download</b></a> |
  <a href="https://github.com/deepseek-ai/DeepSeek-OCR/blob/main/DeepSeek_OCR_paper.pdf"><b>📄 Paper Link</b></a> |
  <a href="https://arxiv.org/abs/2510.18234"><b>📄 Arxiv Paper Link</b></a> |
</p>
<h2>
<p align="center">
  <a href="https://huggingface.co/papers/2510.18234">DeepSeek-OCR: Contexts Optical Compression</a>
</p>
</h2>
<p align="center">
<img src="assets/fig1.png" style="width: 1000px" align=center>
</p>
<p align="center">
<a href="https://huggingface.co/papers/2510.18234">Explore the boundaries of visual-text compression.</a>       
</p>

## Usage
Inference using Huggingface transformers on NVIDIA GPUs. Requirements tested on python 3.12.9 + CUDA11.8：

```
torch==2.6.0
transformers==4.46.3
tokenizers==0.20.3
einops
addict 
easydict
pip install flash-attn==2.7.3 --no-build-isolation
```

```python
from transformers import AutoModel, AutoTokenizer
import torch
import os
os.environ["CUDA_VISIBLE_DEVICES"] = '0'
model_name = 'deepseek-ai/DeepSeek-OCR'

tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
model = AutoModel.from_pretrained(model_name, _attn_implementation='flash_attention_2', trust_remote_code=True, use_safetensors=True)
model = model.eval().cuda().to(torch.bfloat16)

# prompt = "<image>\nFree OCR. "
prompt = "<image>\n<|grounding|>Convert the document to markdown. "
image_file = 'your_image.jpg'
output_path = 'your/output/dir'

# infer(self, tokenizer, prompt='', image_file='', output_path = ' ', base_size = 1024, image_size = 640, crop_mode = True, test_compress = False, save_results = False):

# Tiny: base_size = 512, image_size = 512, crop_mode = False
# Small: base_size = 640, image_size = 640, crop_mode = False
# Base: base_size = 1024, image_size = 1024, crop_mode = False
# Large: base_size = 1280, image_size = 1280, crop_mode = False

# Gundam: base_size = 1024, image_size = 640, crop_mode = True

res = model.infer(tokenizer, prompt=prompt, image_file=image_file, output_path = output_path, base_size = 1024, image_size = 640, crop_mode=True, save_results = True, test_compress = True)
```

## vLLM
Refer to [🌟GitHub](https://github.com/deepseek-ai/DeepSeek-OCR/) for guidance on model inference acceleration and PDF processing, etc.<!--  -->

[2025/10/23] 🚀🚀🚀 DeepSeek-OCR is now officially supported in upstream [vLLM](https://docs.vllm.ai/projects/recipes/en/latest/DeepSeek/DeepSeek-OCR.html#installing-vllm).
```shell
uv venv
source .venv/bin/activate
# Until v0.11.1 release, you need to install vLLM from nightly build
uv pip install -U vllm --pre --extra-index-url https://wheels.vllm.ai/nightly
```

```python
from vllm import LLM, SamplingParams
from vllm.model_executor.models.deepseek_ocr import NGramPerReqLogitsProcessor
from PIL import Image

# Create model instance
llm = LLM(
    model="deepseek-ai/DeepSeek-OCR",
    enable_prefix_caching=False,
    mm_processor_cache_gb=0,
    logits_processors=[NGramPerReqLogitsProcessor]
)

# Prepare batched input with your image file
image_1 = Image.open("path/to/your/image_1.png").convert("RGB")
image_2 = Image.open("path/to/your/image_2.png").convert("RGB")
prompt = "<image>\nFree OCR."

model_input = [
    {
        "prompt": prompt,
        "multi_modal_data": {"image": image_1}
    },
    {
        "prompt": prompt,
        "multi_modal_data": {"image": image_2}
    }
]

sampling_param = SamplingParams(
            temperature=0.0,
            max_tokens=8192,
            # ngram logit processor args
            extra_args=dict(
                ngram_size=30,
                window_size=90,
                whitelist_token_ids={128821, 128822},  # whitelist: <td>, </td>
            ),
            skip_special_tokens=False,
        )
# Generate output
model_outputs = llm.generate(model_input, sampling_param)

# Print output
for output in model_outputs:
    print(output.outputs[0].text)
```


## Visualizations
<table>
<tr>
<td><img src="assets/show1.jpg" style="width: 500px"></td>
<td><img src="assets/show2.jpg" style="width: 500px"></td>
</tr>
<tr>
<td><img src="assets/show3.jpg" style="width: 500px"></td>
<td><img src="assets/show4.jpg" style="width: 500px"></td>
</tr>
</table>


## Acknowledgement

We would like to thank [Vary](https://github.com/Ucas-HaoranWei/Vary/), [GOT-OCR2.0](https://github.com/Ucas-HaoranWei/GOT-OCR2.0/), [MinerU](https://github.com/opendatalab/MinerU), [PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR), [OneChart](https://github.com/LingyvKong/OneChart), [Slow Perception](https://github.com/Ucas-HaoranWei/Slow-Perception) for their valuable models and ideas.

We also appreciate the benchmarks: [Fox](https://github.com/ucaslcl/Fox), [OminiDocBench](https://github.com/opendatalab/OmniDocBench).


## Citation
```bibtex
@article{wei2025deepseek,
  title={DeepSeek-OCR: Contexts Optical Compression},
  author={Wei, Haoran and Sun, Yaofeng and Li, Yukun},
  journal={arXiv preprint arXiv:2510.18234},
  year={2025}
}

</details>