INDEX

Explanations

well-written statements or text

oai_token-act-pair · gpt-3.5-turbo

instances of the term "written."

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 0-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.0.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.0.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

 camp

-0.68

inas

-0.65

 camps

-0.65

lag

-0.63

 tech

-0.62

 subs

-0.62

 heights

-0.62

bot

-0.62

hub

-0.61

 isolate

-0.61

POSITIVE LOGITS

written

4.06

Written

2.64

 written

2.02

write

1.99

 Written

1.89

wrote

1.89

writing

1.87

writ

1.76

authored

1.64

ritten

1.59

Activations Density 0.015%

well-written statements or text

instances of the term "written."

No Comments

No Known Activations