VWFA-Localization

Setup

1. First, download the Qwen2VL-72B model.
2. Place the model inside the saved_model directory.

Scripts

• extract_activations.py
  Records the activations of the model when presented with the localizer stimuli.

• contrast_activations.py
  Computes the mean activations of units for words vs. other categories and generates VWFA masks.

• apply_VWFA_mask_roar.py
  Applies the VWFA mask to the ROAR benchmark.

• apply_VWFA_mask_KSCT.py
  Applies the VWFA mask to the Kempler benchmark.