INDEX

Explanations

references to various criminal activities or acts of wrongdoing

oai_token-act-pair · gpt-3.5-turbo

mentions of various types of crimes

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 0-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.0.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.0.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

 McMaster

-0.74

 placeholder

-0.66

anian

-0.64

whe

-0.63

pring

-0.62

 impression

-0.61

ulu

-0.60

asted

-0.60

eting

-0.59

HF

-0.59

POSITIVE LOGITS

 crimes

3.59

 Crimes

2.78

 crime

2.47

 offenses

2.22

 offences

2.20

crime

2.16

 criminality

2.01

 criminals

1.98

 Crime

1.90

 murders

1.89

Activations Density 0.021%

references to various criminal activities or acts of wrongdoing

mentions of various types of crimes

No Comments

No Known Activations