Interventions

Analysis Methods

Logit Lens

See predictions at each layer:

from nnterp.interventions import logit_lens

prompts = ["The capital of France is", "The sun rises in the"]
probs = logit_lens(model, prompts)
# Shape: (batch, layers, vocab)

Patchscope

Replace activations from one context into another:

from nnterp.interventions import patchscope_lens, TargetPrompt, repeat_prompt

source_prompts = ["Paris is beautiful", "London is foggy"]
target_prompt = TargetPrompt("city: Paris\\nfood: croissant\\n?", -1)

# Or use repeat prompt
target_prompt = repeat_prompt(words=["car", "cross", "azdrfa"])

patchscope_probs = patchscope_lens(
    model, source_prompts=source_prompts, target_patch_prompts=target_prompt
)