INDEX

Explanations

references to people being physically injured or harmed

oai_token-act-pair · gpt-3.5-turbo

mentions of injuries or casualties

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

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

ebook

-0.73

é¾

-0.72

opic

-0.71

Sche

-0.71

ordan

-0.71

perm

-0.71

rx

-0.69

RH

-0.69

ellar

-0.68

 patented

-0.66

POSITIVE LOGITS

 wounded

1.17

 wounding

0.92

 wounds

0.91

lehem

0.85

nikov

0.83

 casualties

0.81

 hostages

0.81

 fatig

0.75

 souls

0.75

 injured

0.74

Activations Density 0.005%

references to people being physically injured or harmed

mentions of injuries or casualties

No Comments

No Known Activations