INDEX

Explanations

timestamps for when something was written

oai_token-act-pair · gpt-3.5-turbo

instances of the word "wrote."

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

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

Ĭ±

-0.83

xon

-0.77

Enlarge

-0.71

Ton

-0.70

phant

-0.68

RELATED

-0.68

Nor

-0.68

amac

-0.67

 Amph

-0.67

EGA

-0.67

POSITIVE LOGITS

 penned

0.91

 written

0.89

smanship

0.88

 wrote

0.87

 eloqu

0.85

 scrib

0.85

 aloud

0.84

vironment

0.83

 write

0.83

 writes

0.82

Activations Density 0.022%

timestamps for when something was written

instances of the word "wrote."

No Comments

No Known Activations