feature_analysis.py File Reference

Go to the source code of this file.

Namespaces
namespace	aquin
namespace	aquin.compute
namespace	aquin.compute.feature_analysis

Functions
	aquin.compute.feature_analysis._get_sae_path_for_layer (str model_id, int|None layer=None)
	aquin.compute.feature_analysis._get_norm_path_for_layer (str model_id, int|None layer=None)
SparseAutoencoder	aquin.compute.feature_analysis._load_sae_native (str model_id, int|None layer=None)
SparseAutoencoder	aquin.compute.feature_analysis.load_sae (str model_id="llama-3.2-1b", int|None layer=None)
dict|None	aquin.compute.feature_analysis.load_norm (str model_id="llama-3.2-1b", int|None layer=None)
torch.Tensor	aquin.compute.feature_analysis.normalize (torch.Tensor x, str model_id="llama-3.2-1b", int|None layer=None)
torch.Tensor	aquin.compute.feature_analysis._sae_feature_ablate_hook (torch.Tensor value, int pos, int feature_idx, SparseAutoencoder sae, str model_id, int sae_layer)
str	aquin.compute.feature_analysis.label_feature_causally (int fi, str prompt, HookedTransformer model, SparseAutoencoder sae, client, int top_k=5, str model_id="llama-3.2-1b", int|None layer=None)
str	aquin.compute.feature_analysis.get_causal_label (int fi, str prompt, HookedTransformer model, SparseAutoencoder sae, client, str model_id="llama-3.2-1b", int|None layer=None)
str	aquin.compute.feature_analysis.format_feature_ref (int feature_idx, str|None label=None)
str	aquin.compute.feature_analysis.prompt_for_labeling (dict|None ctx=None, dict|None args=None)
None	aquin.compute.feature_analysis._attach_label_to_feature_dict (dict f, *, str prompt, HookedTransformer model, SparseAutoencoder sae, client, str model_id, int|None layer, set[int]|None seen=None)
dict	aquin.compute.feature_analysis.label_inspection_features (dict feat_result, *, str prompt, HookedTransformer model, client, str model_id="llama-3.2-1b", int|None layer=None)
dict	aquin.compute.feature_analysis.enrich_feature_tool_result (dict result, *, str prompt, HookedTransformer model, client, str model_id="llama-3.2-1b", int|None layer=None, bool label_neighbors=False)
str	aquin.compute.feature_analysis.resolve_feature_label (int feature_idx, *, dict ctx, dict|None args=None, int|None layer=None)
dict	aquin.compute.feature_analysis._run_sae_pass (str prompt, str response, HookedTransformer model, int top_k=TOP_K_FEATURES, str model_id="llama-3.2-1b", int|None layer=None)
dict	aquin.compute.feature_analysis.run_feature_analysis_unlabeled (str prompt, str response, HookedTransformer model, str model_id="llama-3.2-1b", int|None layer=None)
dict	aquin.compute.feature_analysis.run_feature_analysis (str prompt, str response, HookedTransformer model, client, int top_k=TOP_K_FEATURES, str model_id="llama-3.2-1b", int|None layer=None)
dict	aquin.compute.feature_analysis.get_feature_logits (int feature_idx, HookedTransformer model, str model_id="llama-3.2-1b", int|None layer=None, int top_k=10)
dict	aquin.compute.feature_analysis.get_feature_neighbors (int feature_idx, str model_id="llama-3.2-1b", int|None layer=None, int top_k=8)

Variables
int	aquin.compute.feature_analysis.TOP_K_FEATURES = 10
	aquin.compute.feature_analysis.DEVICE = resolve_compute_device()
dict	aquin.compute.feature_analysis._sae_cache = {}
dict	aquin.compute.feature_analysis._norm_cache = {}
dict	aquin.compute.feature_analysis._session_label_cache = {}
	aquin.compute.feature_analysis._sae = None
	aquin.compute.feature_analysis._norm = None
	aquin.compute.feature_analysis._kernel_feature_acts = None
	aquin.compute.feature_analysis._kernel_resid = None
list	aquin.compute.feature_analysis._kernel_top_features = []