DeepSeek-OCR: Quick Start Guide

---

1. Google Colab (Recommended for Quick Testing)

If you want to test DeepSeek-OCR without using local resources:

• Use the notebook at notebooks/deepsek_ocr.ipynb.
• Open it in Google Colab.
• Important: Select a runtime with GPU. This configuration will not work with CPU.

---

2. Local Runtime

2.1. Install Git LFS

git lfs install

2.2. Clone the DeepSeek-OCR Repository

git clone https://huggingface.co/deepseek-ai/DeepSeek-OCR

2.3. GPU/CPU Support

• Default:
  DeepSeek-OCR is configured to use CUDA and requires a GPU.

• To run on CPU:
  Download the modeling_deepseekocr.py file from this discussion and replace your local file with it.

2.4. Installing PyTorch with CUDA via pyproject.toml

To set up your pyproject.toml to download a CUDA-compatible version of torch, add:

[[tool.uv.index]]
name = "pytorch-cu128"
url = "https://download.pytorch.org/whl/cu128"
explicit = true

[tool.uv.sources]
torch = [
  { index = "pytorch-cu128", marker = "sys_platform == 'linux' or sys_platform == 'win32'" },
]
torchvision = [
  { index = "pytorch-cu128", marker = "sys_platform == 'linux' or sys_platform == 'win32'" },
]

For more information about specific CUDA versions, see the official uv documentation.

2.5. Set Up the Environment

uv sync

2.6. Testing DeepSeek-OCR

• Modify the configuration parameters (image path, prompt type, etc.) in the run_ocr.py file.
• Run the script to test DeepSeek-OCR.