Merge pull request #7 from seanpedrick-case/dev

Updated default data source

Merge pull request #6 from seanpedrick-case/dev

Corrected Gemma model prompt format, made prompts in general ask for more detail. Wording changes.

Merge pull request #5 from seanpedrick-case/dev

Hugging Face spaces implementation seems unable to install latest llama-cpp-python version, so downgrading.

Merge pull request #4 from seanpedrick-case/dev

Updated Dockerfile so it actually works. Modified config file to include/exclude models and change relevant options as needed. Fixed thumbs up/down feedback.

Merge pull request #3 from seanpedrick-case/dev

Removed other references to boto3 client

Merge pull request #2 from seanpedrick-case/dev

Removed obligatory bedrock load in from app

Merge pull request #1 from seanpedrick-case/dev

Refocused app for working with Document Redaction guide. Added Gemma 3 1b as base model. Various UI and config improvements

Initial compatibility tested for use with Gemini and AWS Bedrock APIs

Returned to bge-base embeddings for now. Improved UI a little

Updated packages, functions, fixed a couple of bugs

Added boto3 to requirements

Updated packages, models, preparing for use with AWS (in background)

Changed 'large' model to Phi 3 Mini gguf 128k. Added requirements file for cpu. Put prompts in separate file.

Attempted Gradio downgrade to 3.50.2 to resolve version conflicts

CPU Flan inference is crashing, so trying to revert to previous package versions that worked

Removed reference to ctransformers

Upgraded gradio and packages to latest. Switched Ctransformers with Llama.cpp Python

Minor dependency updates, increased max allowable layers for GPU

Added temperature slider, more stringent checks for document relevance

Added slider for changing number of passages out

Improved prompting for csv/excel files

put back pip install requirement for gradio 3.42.0 for likeData compatibility

Added csv/Excel file support

Put sources behind accordion, minor change to model params. Gitignore update

Hid gpu layers behind accordion

Build fail when gpu_layers > 0, so will set to 0 at start and modify in app.

Made GPU layer bar invisible. Changed default GPU layers when CUDA available

Upgraded large model to Mistral OpenOrca 7B Q4. More checks for empty questions.

Changed embedding model, added reference to chat model on front page

Improved advanced model prompt, added stop generation button. context prompt improvements

Stop generation button. Better model load. Trying one source that's longer.

Setting gpu_layer slider to false to prevent accidental misuse

Removed bootstrapper.py as it may confuse HF

gpu_layers should now update correctly. Added code for creating distribution.

Downgraded Gradio back to 3.42.0 in case that's an issue.

Hopefully fixed Orca load cuda error. Loads in Orca on app load. Later gradio version installed.

Temporarily disabled Orca Mini load until Cuda compatibility sorted out

Explicitly installs required gradio version upon loading app

Added working like buttons. Added model choice. Modified requirements txt

added bot icon with white background

Cleaned up code a bit, added user icons, thumbs up/down

Reduced chat box size further

Reduced chat box height