INDEX

Explanations

Twitter handles prefixed with '@'

oai_token-act-pair · gpt-3.5-turbo

names of people and their social media handles

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 8-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.8.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.8.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

 wards

-0.67

 striking

-0.65

 sights

-0.65

 notebooks

-0.64

 bruising

-0.64

 tense

-0.63

 fingerprints

-0.62

 certificates

-0.62

 waivers

-0.62

 "âĢ¦

-0.62

POSITIVE LOGITS

1.18

Writ

1.08

Jew

1.03

GoldMagikarp

1.00

CBC

0.97

Blog

0.95

Ide

0.92

News

0.91

Reports

0.91

Report

0.87

Activations Density 0.145%

Twitter handles prefixed with '@'

names of people and their social media handles

No Comments

No Known Activations