mirror of
https://github.com/Cinnamon/kotaemon.git
synced 2026-07-09 11:59:53 +02:00
9ad3e4e49a
* docs: typo * fix: enhance mention @ (file, websearch) in chat * fix: shorter placeholder guide in chat * fix: enhance UI (file filter, chunks preview, lightrag reader settings) * feat: feat: integrate paddleocr (PaddleOCR-VL, PP-StructureV3) * fix: import * fix: support both urls and mention files when chatting * feat: restore the test connection panel in resources * fix: paddleocr output adapter * tests: update * fix: parse image block content * fix: docker gpu stage for paddleocr * tests: update unit tests * tests: add skip_when_paddleocr_not_installed * fix: reset chunk preview filter when re-selecting file * fix: paddleocr[all] -> paddleocr[doc-parser] * fix: allow configure paddleocr device with env * feat: parameterize CUDA version in Dockerfile for paddlepaddle installation * docs: add integration guides for PaddleOCR and Docling * fix: updating mcp server must reflect in the tool choice UI * fix: disable enable_mkldnn for cpu inference * docs: update installation doc with uv * fix: use chat_input_text for llm input instead of the UI display message * fix: render file mentions as HTML and support unquoted names * fix: ensures proper cleanup of event listeners before replacing the Tribute instance. * test: updae unit tests --------- Co-authored-by: GSD Agent <kotaemon@example.com>
2.0 KiB
2.0 KiB
PaddleOCR
Kotaemon provides two PaddleOCR readers to enable document ingestion with full layout understanding, including multilingual text, tables, figures, formulas, and seals.
PaddleOCRVLReader: Wraps the PaddleOCR-VL 1.5 visual-language model for robust layout and VQA-based parsing.PPStructureV3Reader: Uses the PPStructureV3 pipeline for structured layout analysis, including table and chart detection.
Both readers are located under kotaemon/loaders/paddleocr_loader.
Prerequisites
- Install PaddlePaddle:
- Ensure that the installed PaddlePaddle version matches your system configuration and hardware (CPU/GPU). For additional wheel options, refer to the PaddlePaddle official website.
- Check device support for PaddleOCR-VL.
# CPU
uv pip install paddlepaddle==3.3.0 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
# gpu，requires GPU driver version ≥550.54.14 (Linux) or ≥550.54.14 (Windows)
uv pip install paddlepaddle-gpu==3.3.0 -i https://www.paddlepaddle.org.cn/packages/stable/cu130/
- Install the PaddleOCR doc parser extras:
uv pip install -e "libs/kotaemon[paddleocr]"
- Configure the device: You can set the
PADDLE_DEVICEenvironment variable in your .env file to control the execution device.
PADDLE_DEVICE=gpu # cpu, gpu:0
Configure the loader
- Run Kotaemon and open the app UI.
- Navigate to Settings → File Loader (under indexing or ingestion settings).
- Select one of the PaddleOCR loaders:
PaddleOCR PPStructureV3 (table+figure extraction)PaddleOCR-VL (VLM document parsing)
- Save the settings, then upload or ingest a document. Kotaemon will automatically use the selected PaddleOCR loader during indexing and convert extracted content into Document objects.