Credits

Models

This implementation is build on GODEL by Microsoft, Inc. and Mistral 7B Instruct by Mistral AI.

Mistral 7B Instruct

Mistral 7B Instruct is an open source model by Mistral AI. See offical paper for more information.

the version used in this project is Mistral 7B Instruct v0.2, see huggingface model hub
the model is an autoregressive language model fine-tuned for instruction following

GODEL

GODEL is an open source model by Microsoft. See offical paper for more information.

the version used in this project is GODEL Large, see huggingface model hub
the model as is a generative seq2seq transformer fine-tuned for goal directed dialog

Libraries

This project uses a number of open source libraries, only the most important ones are listed below.

captum

This application uses the captum library for the interpretation of the Mistral 7B Instruct model. The captum library is available at GitHub.

Please refer to the captum website for more information about the library.
The used KernelExplainer is based on work by Lundberg et al. - see below for more information.
For original paper about captum see Inital Paper.

Shap

This application uses a custom version of the shap library, which is available at GitHub.

Please refer to the thesis-shap repository for more information about the changes made to the library, specifically the README file.
The shap library and the used partition SHAP explainer are based on work by Lundberg et al. (2017), see offical paper for more information.

Visualizations

This application uses attention visualization inspired by the bertviz library, which is available atGitHub. It doesn't actually use BERTViz.

The bertviz was introduced by Vig et al. (2019), see offical paper for more information.
This project only uses decoder attention visualization with gradio and matplotlib and not BERTViz itself.

Data Protection

This is a non-commercial research project, which does not collect any personal data. The only data collected is the data you enter into the application. This data is only used to generate the explanations and is not stored anywhere.

However, the application may be hosted with an external service (i.e. Huggingface Spaces), which may collect data.

Please refer to the data protection policies of the respective service for more information. If you use the "flag" feature, the data you enter will be stored in publicly available csv file.

License

This Product is licensed under the MIT license. See LICENSE at GitHub for more information. Please credit the original authors of this project (Lennard Zündorf) and the credits listed above if you use this project or parts of it in your own work.