DeepSeek-OCR
A local OCR model processes images and PDFs through vLLM or Hugging Face Transformers, with CUDA inference and an MIT license.
DeepSeek-OCR is an OCR model for developers processing document images and PDFs on their own hardware. It also supports research into visual text compression, representing text-bearing images with compact sets of vision tokens.
- Image processing: Read individual images through vLLM with streaming output.
- PDF processing: Process PDFs concurrently with the supplied vLLM inference tools.
- Resolution choices: Use fixed image resolutions or a dynamic resolution mode to vary the image detail passed to the model.
- Benchmark evaluation: Evaluate the model in batches using the included benchmark tools.
The model supports local inference through vLLM or Hugging Face Transformers. Its inference environment uses PyTorch and CUDA, and the PDF example runs on an NVIDIA A100 GPU. The Python project is open source under the MIT license.
Logotyper
