Updated the Main branch based on the requirements. The code is optimized for generating the MIND data for 3 checkpoints (could be done on more), gather the results into one csv, and use that for getting the activations of the data points on which at least one checkpoint hallucinates and another does not. At the moment the code only generates the results for [hallucinates, not hallucinate, hallucinates] which allows unifying the vision for the analysis.
The pipeline included in the PCA.ipynb file needs to be extended into a script form and automated. At the moment, the dimensionalities are hardcoded.
@JuanDGuerra20
Updated the
Main
branch based on the requirements. The code is optimized for generating the MIND data for 3 checkpoints (could be done on more), gather the results into one csv, and use that for getting the activations of the data points on which at least one checkpoint hallucinates and another does not. At the moment the code only generates the results for [hallucinates, not hallucinate, hallucinates] which allows unifying the vision for the analysis.The pipeline included in the
PCA.ipynb
file needs to be extended into a script form and automated. At the moment, the dimensionalities are hardcoded. @JuanDGuerra20