INDEX

Explanations

words related to administrative or governmental entities, specifically the term "cant" or "cantonment"

oai_token-act-pair · gpt-3.5-turbo

instances of the word "cant" and its variations, indicating a focus on expressions of inability or lack of permission

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 1-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.1.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.1.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

 Generation

-0.80

 Flavoring

-0.77

GB

-0.72

æł

-0.69

Ô

-0.69

values

-0.66

taining

-0.65

å£«

-0.65

Cam

-0.63

GN

-0.63

POSITIVE LOGITS

 dont

0.89

 cant

0.88

peg

0.83

kson

0.82

ional

0.80

cha

0.73

pole

0.72

ember

0.71

 bother

0.71

onal

0.71

Activations Density 0.006%

words related to administrative or governmental entities, specifically the term "cant" or "cantonment"

instances of the word "cant" and its variations, indicating a focus on expressions of inability or lack of permission

No Comments

No Known Activations