Indexing works! Bug fixes and UI improvemnents
What's Changed
Updates
This update introduces several significant updates and improvements to the GraphRAG-Local-UI project:
Major Changes
-
Local Model Support Enhancement: Improved compatibility with local models, particularly for embedding. Added a note that embedding providers using a /v1/embeddings endpoint work best currently. Switch your Ollama embeddings to 14434/api (even if you get an error in Gradio, it will still set the correct tag)
-
Indexing Process Overhaul:
- Significant updates to the indexing workflow to better handle various LLM and embedder configurations.
- Improved error handling and logging for easier debugging of indexing issues.
-
Configuration Updates:
- Added example configuration files and environment variable templates.
- Updated
settings.yaml
and addedsettings-example.yaml
for easier setup.
-
Input Processing:
- Removed outdated input files.
- Added a new sample document about tree frog adhesion biomimetics.
-
Output Structure Reorganization:
- Implemented a new structure for output artifacts and reports.
- Added new output folders with updated artifacts and logs.
-
Caching Mechanism:
- Implemented caching for text embeddings and summarizations to improve performance.
-
UI and Visualization Improvements:
- Enhanced graph visualization capabilities.
- Updated the app.py file with new functionalities for the interactive UI.
Minor Changes
- Updated README.md with more detailed information about the project status, installation process, and known issues.
- Refined error messages and logging throughout the codebase.
- Updated default configuration values in
graphrag/config/defaults.py
. -
- fix:'gbk' codec can't decode byte 0x8c by @Ikaros-521 in #42
Bug Fixes
- Addressed several issues in the indexing process, particularly related to empty nodes/data frames when unpacking.
- Fixed encoding/UTF errors that were occurring on Windows systems.
Development Notes
- This update represents a major work in progress. Users should expect some instability and potential bugs, especially in the indexing process.
- The changes aim to make the application more robust with local LLMs, but further testing and refinement are needed.
- Feedback and bug reports from users are highly encouraged to help improve the system.
Next Steps
- Continue stabilizing the indexing process.
- Further optimize the workflow for various LLM and embedder configurations.
- Enhance documentation and user guides.
- Address any new issues that arise from these changes.
New Contributors
- @Ikaros-521 made their first contribution in #42
Full Changelog: v1.0...v1.01