INDEX

Explanations

the word "latest" followed by a number

oai_token-act-pair · gpt-3.5-turbo

references to recent updates or news

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

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

erved

-0.79

avery

-0.78

alties

-0.73

velt

-0.73

ivas

-0.69

utenant

-0.69

 Rahman

-0.69

hovah

-0.68

bara

-0.66

dinand

-0.65

POSITIVE LOGITS

 incarnation

1.06

 iteration

1.05

 edition

1.04

 installment

1.04

 update

0.91

 editions

0.85

 episode

0.82

 latest

0.81

 developments

0.81

 batch

0.80

Activations Density 0.018%

the word "latest" followed by a number

references to recent updates or news

No Comments

No Known Activations