INDEX

Explanations

instances of people acknowledging or confessing to certain actions or thoughts

oai_token-act-pair · gpt-3.5-turbo

occurrences of the word "he" and variations related to admissions or acknowledgments

oai_token-act-pair · gpt-4o-mini Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 7-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.7.hook_resid_pre

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

24,576

Data Type

torch.float32

Hook Point

blocks.7.hook_resid_pre

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Hook Point Layer

Activation Function

relu

Embeds

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

icles

-0.75

Review

-0.71

Liber

-0.69

lake

-0.68

Friends

-0.68

cellaneous

-0.66

Ec

-0.65

clock

-0.64

blogspot

-0.64

 Budd

-0.63

POSITIVE LOGITS

 mistakes

0.97

 regrets

0.91

 underestimated

0.74

 wrongdoing

0.73

 misled

0.73

 underest

0.73

 unintentional

0.71

 conflicted

0.71

 mishand

0.70

 imperfect

0.70

Activations Density 0.132%

instances of people acknowledging or confessing to certain actions or thoughts

occurrences of the word "he" and variations related to admissions or acknowledgments

No Comments

No Known Activations