To install SmolDocling, follow these steps:
- environmental preparation: Ensure that Python 3.8+ is installed, a virtual environment is recommended
- Installation of dependencies: Execute command
pip install torch transformers docling_core - GPU acceleration(Optional): Install the CUDA version of PyTorch for speed, available via the
torch.cuda.is_available()Detection of support
The utilization process is divided into five stages:
- Image Loading: Use
load_image()Function to import the image to be processed - Model initialization: Automatic download of model weights via Hugging Face (requires initial internet connection)
- Document Conversion: Generate DocTags using a specific prompt template
- format conversion: Export DocTags to popular formats such as Markdown.
- Advanced Optimization: GPU users can enable flash_attention_2 accelerated processing
Note that adjustments may be required when processing large images max_new_tokens parameter (default 8192), it is recommended to print the intermediate results for debugging when using it for the first time.
This answer comes from the articleSmolDocling: a visual language model for efficient document processing in a small volumeThe
May not be reproduced without permission:AI productivity tools " How do I install and run SmolDocling on my local computer?
English 
简体中文 
日本語 
Deutsch 
Português do Brasil