A scanner that will scan your AI models for problems. Currently it focuses on bias testing. It is currently alpha.
First install it:
pip install daisybell
Run it in this manner (currently supports models from HuggingFace's repository):
daisybell roberta-base
The scan can output files for further analysis:
daisybell roberta-base --output results/roberta-base
We will infer the task(s) of model by default but to provide specific tasks to test explicitly use the --task switch:
daisybell cross-encoder/nli-distilroberta-base --task zero-shot-classification
Daisybell can even scan chatbot models like StableLM for AI safety and alignment issues:
daisybell StabilityAI/stablelm-tuned-alpha-3b -d cuda:0
That's it for now. More will come.
- More bias tests. More metrics for bias testing based on the research in the field.
- Integration with other types of testing (eg. adversarial robustness)
- More kinds of models besides HuggingFace models. We are especially interested in MLFlow integration.
- Documentation.
Please contribute if you can. Help is always helpful.
Apache
A project of IQT Labs.