Great work! Any plan to release more detailed documentation?

[xjtupanda]

Thanks for the great work! I wonder if there would be documentation/tutorial on how to access the quantities (e.g., importance) and draw plots in a customized way (Like Fig.9 in the paper)？

The GUI is very friendly for specific case studies. But if I want to calculate the importance of attention heads for a specific dataset (e.g., reasoning task), is there a simple class/interface to access related quantities and calculate them in batches?

Though I noticed that there's an argument called "preloaded_dataset_filename":

llm-transparency-tool/llm_transparency_tool/server/app.py

Lines 91 to 94 in ee81984

	preloaded_dataset_filename: Optional[str] = field(
	default=None,
	metadata={"help": "The name of the text file to load the lines from."},
	)

I think this is just for the ease of loading multiple cases as in sample_input.txt.