INDEX

Explanations

references to the word "Tart", potentially related to a title, name, or descriptor

oai_token-act-pair · gpt-3.5-turbo

references to a specific author and their distinctive writing style

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

ctive

-0.95

 Inqu

-0.80

ptions

-0.75

phies

-0.75

bilt

-0.74

enez

-0.74

ITE

-0.73

ULTS

-0.72

Bee

-0.69

ardless

-0.69

POSITIVE LOGITS

 Tart

1.10

arus

0.94

 tart

0.94

atars

0.81

arist

0.79

arbon

0.79

anian

0.75

gian

0.73

alist

0.73

yrinth

0.73

Activations Density 0.020%

references to the word "Tart", potentially related to a title, name, or descriptor

references to a specific author and their distinctive writing style

No Comments

No Known Activations