- Added ability to change context prompt, switched to
condense_plus_context
chat mode.
- Added dockerfiles and database export and import functionality.
- Moved context window to only be handled on an LLM-level.
- Fixed chat memory limit overflowing the context window leading the model to stop answering.
- Added Google News support for generation of a corpus.
- Made model parameters persistent per user between sessions.
- Split the LLM from the user sesssion. All users share the same LLM, unlimited users.
- Initial release, depends on local_rag_llm==0.0.11. Separate LLMs for every user. Number of users restricted by amount of VRAM.