localize_collapse.py File Reference

Go to the source code of this file.

Namespaces
namespace	aquin
namespace	aquin.compute
namespace	aquin.compute.localize_collapse

Functions
torch.Tensor	aquin.compute.localize_collapse._unit (torch.Tensor v)
dict[str, float]	aquin.compute.localize_collapse._centroid_signal (torch.Tensor honest, torch.Tensor deceptive)
str	aquin.compute.localize_collapse._status (float signal)
dict[str, Any]	aquin.compute.localize_collapse.pick_collapse_layer (list[dict[str, Any]] layers)
list[dict[str, Any]]	aquin.compute.localize_collapse.score_layer_activations (dict[int, torch.Tensor] honest_by_layer, dict[int, torch.Tensor] deceptive_by_layer)
list[dict[str, Any]]	aquin.compute.localize_collapse.score_stressor_collapse (list[dict[str, Any]] baseline_layers, list[dict[str, Any]] stressor_layers)
dict[str, float]	aquin.compute.localize_collapse._project_on_direction (torch.Tensor honest, torch.Tensor deceptive, torch.Tensor direction)
list[dict[str, Any]]|None	aquin.compute.localize_collapse.rank_sae_features_at_layer (Any model, str model_id, int layer, list[str] honest_texts, list[str] deceptive_texts, *, int top_k=8)
dict[str, Any]	aquin.compute.localize_collapse.run_localize_collapse (Any model, str model_id, *, str|Path|None prompts=None, str|Path|None stressor_prompts=None, int|None feature_idx=None, str|Path|None vector_path=None, int|None layer=None, int top_k_features=8, Any|None collect_layer_activations=None)

Variables
float	aquin.compute.localize_collapse.SIGNAL_WEAK = 0.08
float	aquin.compute.localize_collapse.SIGNAL_COLLAPSED = 0.03
int	aquin.compute.localize_collapse.MAX_PROBES_PER_CLASS = 24