DeepSeek-OCR README

The project's README from github.com/deepseek-ai/DeepSeek-OCR. As one markdown file: README.md.

DeepSeek AI
Homepage Hugging Face
Discord Twitter Follow

📥 Model Download | 📄 Paper Link | 📄 Arxiv Paper Link |

DeepSeek-OCR: Contexts Optical Compression

Explore the boundaries of visual-text compression.

Release

Contents

Install

Our environment is cuda11.8+torch2.6.0.

  1. Clone this repository and navigate to the DeepSeek-OCR folder
git clone https://github.com/deepseek-ai/DeepSeek-OCR.git
  1. Conda
conda create -n deepseek-ocr python=3.12.9 -y
conda activate deepseek-ocr
  1. Packages
pip install torch==2.6.0 torchvision==0.21.0 torchaudio==2.6.0 --index-url https://download.pytorch.org/whl/cu118
pip install vllm-0.8.5+cu118-cp38-abi3-manylinux1_x86_64.whl
pip install -r requirements.txt
pip install flash-attn==2.7.3 --no-build-isolation

Note: if you want vLLM and transformers codes to run in the same environment, you don't need to worry about this installation error like: vllm 0.8.5+cu118 requires transformers>=4.51.1

vLLM-Inference

cd DeepSeek-OCR-master/DeepSeek-OCR-vllm
  1. image: streaming output
python run_dpsk_ocr_image.py
  1. pdf: concurrency ~2500tokens/s(an A100-40G)
python run_dpsk_ocr_pdf.py
  1. batch eval for benchmarks
python run_dpsk_ocr_eval_batch.py

[2025/10/23] The version of upstream vLLM:

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

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)

Transformers-Inference

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'

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)

or you can

cd DeepSeek-OCR-master/DeepSeek-OCR-hf
python run_dpsk_ocr.py

Support-Modes

The current open-source model supports the following modes:

Prompts examples

# document: <image>\n<|grounding|>Convert the document to markdown.
# other image: <image>\n<|grounding|>OCR this image.
# without layouts: <image>\nFree OCR.
# figures in document: <image>\nParse the figure.
# general: <image>\nDescribe this image in detail.
# rec: <image>\nLocate <|ref|>xxxx<|/ref|> in the image.
# '先天下之忧而忧'

Visualizations

Acknowledgement

We would like to thank Vary, GOT-OCR2.0, MinerU, PaddleOCR, OneChart, Slow Perception for their valuable models and ideas.

We also appreciate the benchmarks: Fox, OminiDocBench.

Citation

@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}
}

DeepSeek-OCR (deepseek-ai/DeepSeek-OCR)

Searching Database…

24K ·GitHub
DeepSeek-OCR
What is this?DeepSeek-OCR is a vision-language model built to explore 'contexts optical compression' -- squeezing text-heavy document images into a small number of vision tokens that an LLM can still read back accurately, as a research probe into how much text a vision encoder can represent visually.
Is this a ready-made OCR app or something I run from code?It is an open-source research model, not a packaged app -- you run it yourself through either vLLM inference or Hugging Face Transformers inference scripts included in the repo.
How do I get it running?Clone the repository, set up the Conda environment, install the required packages including a specific vllm-0.8.5 wheel, then run either the vLLM inference script (for streaming images or PDF batches) or the Transformers inference script.
Do I need a CUDA GPU to run it?Yes -- the reference environment is cuda11.8+torch2.6.0, and the benchmark throughput numbers are measured on an A100-40G GPU.
Any install pitfalls?Running vLLM and Transformers inference side by side in one environment can trigger a dependency clash, since vllm 0.8.5+cu118 requires transformers>=4.51.1; the README flags this exact error ahead of time.
What resolution modes does it support?It offers four fixed native resolutions (Tiny 512x512 with 64 vision tokens, Small 640x640/100 tokens, Base 1024x1024/256 tokens, Large 1280x1280/400 tokens) plus a dynamic 'Gundam' mode that combines several 640x640 tiles with one 1024x1024 tile.
How fast is it on real documents?For PDF batch evaluation the model reaches around 2500 tokens/s on a single A100-40G GPU.
Why is it called 'optical compression'?The idea is to represent large amounts of text visually using far fewer vision tokens than the equivalent text tokens, testing how much a model can recover from a heavily compressed visual encoding of a document.
Is there a newer version?Yes -- DeepSeek-OCR2 was announced on 2026/01/27 as a successor, while this repository covers the original DeepSeek-OCR release from October 2025, which is now officially supported in upstream vLLM.
What are similar projects?ZDTaichu5.0-9B, Unlimited-OCR, InternVL, OraRL, UniVR, TeleOCR, halo, Intern-Decision
Which topics is it in?vision-language-model, natural-language-processing, transformers, vllm, ocr, document-processing, multimodal, huggingface, computer-vision, cuda, pytorch, benchmark
Creator
deepseek-ai
Size
7.8 MB
Stars
24K
Updated
CommitJan 27
Age
1y
Commits · 90 Days