INDEX

Explanations

words related to negative or offensive behavior or comments

oai_token-act-pair · gpt-3.5-turbo

derogatory terms and references to socially taboo activities

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

 Plate

-0.67

spring

-0.67

Sources

-0.65

Source

-0.65

 Source

-0.62

ensus

-0.61

 Oral

-0.60

 Celtic

-0.59

 Sacrament

-0.59

 Novel

-0.59

POSITIVE LOGITS

jer

1.22

 jerk

1.13

etsk

0.97

ithing

0.89

bucks

0.88

boa

0.85

balls

0.84

>>\

0.84

artifacts

0.83

EStream

0.82

Activations Density 0.010%

words related to negative or offensive behavior or comments

derogatory terms and references to socially taboo activities

No Comments

No Known Activations