INDEX

Explanations

terms related to legal matters or actions, particularly those with negative connotations like fraud or deception

oai_token-act-pair · gpt-3.5-turbo

forms of the word "defend" or its variations

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 4-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.4.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.4.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

 Archdemon

-0.78

 Dragonbound

-0.74

 Bohem

-0.71

DAY

-0.71

 Boll

-0.70

 Madness

-0.69

externalActionCode

-0.68

OLOGY

-0.66

 Parenthood

-0.66

ocene

-0.65

POSITIVE LOGITS

ensible

1.22

ibr

1.07

erent

1.07

anged

1.02

acement

1.00

ected

0.98

ensions

0.97

erence

0.97

aul

0.97

ident

0.96

Activations Density 0.006%

terms related to legal matters or actions, particularly those with negative connotations like fraud or deception

forms of the word "defend" or its variations

No Comments

No Known Activations