Sanity checks for comparing gradient attribution maps
-
Updated
Jun 12, 2024 - MATLAB
Sanity checks for comparing gradient attribution maps
Post-hoc audit of a 22.3σ val_bpb false-positive in a 19M-parameter knowledge distillation experiment. Dual-dimension attribution: token-level gradient + sample-level pass@k.
Add a description, image, and links to the gradient-attribution topic page so that developers can more easily learn about it.
To associate your repository with the gradient-attribution topic, visit your repo's landing page and select "manage topics."