lm_polygraph.stat_calculators.attention_forward_pass_visual module

class lm_polygraph.stat_calculators.attention_forward_pass_visual.AttentionForwardPassCalculatorVisual[source]

Bases: StatCalculator

For VisualLM (vision-language) WhiteboxModel: computes attention weights during forward pass using greedy tokens + image inputs.

static meta_info() → Tuple[List[str], List[str]][source]

Placeholder method to return the list of statistics and dependencies for the calculator.