INDEX

Explanations

the word 'type' followed by a number, particularly 'type 9' and 'type 10'

oai_token-act-pair · gpt-3.5-turbo

phrases that reference different categories or classifications

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 3-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.3.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.3.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

romeda

-0.72

å§«

-0.67

olulu

-0.66

vernment

-0.65

ITED

-0.65

pload

-0.65

ernel

-0.64

borough

-0.64

 Morning

-0.63

 Liberties

-0.62

POSITIVE LOGITS

faces

1.21

face

1.16

etter

1.09

casting

0.98

etting

0.94

ahead

0.85

cast

0.77

alias

0.76

classes

0.75

geist

0.73

Activations Density 0.020%

the word 'type' followed by a number, particularly 'type 9' and 'type 10'

phrases that reference different categories or classifications

No Comments

No Known Activations