INDEX

Explanations

phrases related to investigations or suspicions of dishonesty or wrongdoing

oai_token-act-pair · gpt-3.5-turbo

characters or symbols that are likely encoding errors or corruption in text

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 5-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.5.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.5.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

•GPT2 L5 Resid: RRLLRLLRLLRRRRLLRRRRRRRRR

No Comments

Negative Logits

 scatter

-0.72

 guiActiveUnfocused

-0.71

 lodging

-0.66

 Golem

-0.62

 buggy

-0.62

 minim

-0.62

isode

-0.60

 detached

-0.60

iliary

-0.59

odan

-0.59

POSITIVE LOGITS

£

1.14

¹

1.03

ı

1.01

º

1.01

âĸº

1.00

âĢķ

0.99

į

0.95

¡

0.92

Ĵ

0.92

»

0.91

Activations Density 0.490%

phrases related to investigations or suspicions of dishonesty or wrongdoing

characters or symbols that are likely encoding errors or corruption in text

No Comments

No Known Activations