INDEX

Explanations

names or titles followed by positions or roles

oai_token-act-pair · gpt-3.5-turbo

phrases related to editorial positions and roles

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 11-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.11.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.11.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

abouts

-0.66

olid

-0.64

wob

-0.64

OH

-0.62

 PLUS

-0.61

tery

-0.60

 collapsing

-0.59

snap

-0.58

THING

-0.57

 crashing

-0.57

POSITIVE LOGITS

 chief

1.21

Chief

1.15

chief

1.15

 Chief

1.09

 charge

0.97

 chiefs

0.87

âĹ¼

0.82

 Charge

0.81

 Chiefs

0.78

ashington

0.71

Activations Density 0.036%

names or titles followed by positions or roles

phrases related to editorial positions and roles

No Comments

No Known Activations