aquin.compute.interp_score Namespace Reference

Functions
dict	_generate_sentences (str label, client, int n=N_SAMPLES)
float	_get_feature_activation (str sentence, int feature_idx, HookedTransformer model, SparseAutoencoder sae, str model_id="llama-3.2-1b", int|None layer=None)
float	_cohen_d_score (list[float] pos_acts, list[float] neg_acts)
float|None	_feature_purity_score (list[str] sentences, client)
float|None	_finite (float|None x, float|None default=None)
float	_kl_div (torch.Tensor p, torch.Tensor q)
float	_entropy (torch.Tensor p)
dict	run_mui_score (int feature_idx, str prompt, HookedTransformer model, SparseAutoencoder sae, int n_positions=8, str model_id="llama-3.2-1b", int|None layer=None)
dict	run_interp_score (int feature_idx, str prompt, HookedTransformer model, SparseAutoencoder sae, client, int n_samples=N_SAMPLES, str model_id="llama-3.2-1b", int|None layer=None)

Variables
	DEVICE = resolve_compute_device()
int	N_SAMPLES = 10

Function Documentation

_cohen_d_score()

float _cohen_d_score ( list[float] pos_acts, list[float] neg_acts )

protected

Definition at line 78 of file interp_score.py.

Referenced by run_interp_score().

_entropy()

float _entropy ( torch.Tensor p )

protected

Definition at line 136 of file interp_score.py.

Referenced by run_mui_score().

_feature_purity_score()

float | None _feature_purity_score ( list[str] sentences, client )

protected

Definition at line 90 of file interp_score.py.

Referenced by run_interp_score().

_finite()

float | None _finite ( float | None x, float | None default = None )

protected

Definition at line 116 of file interp_score.py.

Referenced by run_mui_score().

_generate_sentences()

dict _generate_sentences ( str label, client, int n = N_SAMPLES )

protected

Definition at line 37 of file interp_score.py.

Referenced by run_interp_score().

_get_feature_activation()

float _get_feature_activation ( str sentence, int feature_idx, HookedTransformer model, SparseAutoencoder sae, str model_id = "llama-3.2-1b", int | None layer = None )

protected

Definition at line 63 of file interp_score.py.

Referenced by run_interp_score().

_kl_div()

float _kl_div ( torch.Tensor p, torch.Tensor q )

protected

Definition at line 130 of file interp_score.py.

Referenced by run_mui_score().

run_interp_score()

dict run_interp_score	(	int	feature_idx,
		str	prompt,
		HookedTransformer	model,
		SparseAutoencoder	sae,
			client,
		int	n_samples = N_SAMPLES,
		str	model_id = "llama-3.2-1b",
		int | None	layer = None )

Definition at line 228 of file interp_score.py.

References _cohen_d_score(), _feature_purity_score(), _generate_sentences(), _get_feature_activation(), and run_mui_score().

run_mui_score()

dict run_mui_score	(	int	feature_idx,
		str	prompt,
		HookedTransformer	model,
		SparseAutoencoder	sae,
		int	n_positions = 8,
		str	model_id = "llama-3.2-1b",
		int | None	layer = None )

Definition at line 141 of file interp_score.py.

References _entropy(), _finite(), and _kl_div().

Referenced by run_interp_score().

Variable Documentation

DEVICE

aquin.compute.interp_score.DEVICE = resolve_compute_device()

Definition at line 33 of file interp_score.py.

N_SAMPLES

int aquin.compute.interp_score.N_SAMPLES = 10

Definition at line 34 of file interp_score.py.