INDEX

Explanations

proper nouns and numbers related to news events such as people's names, nationalities, and statistics

oai_token-act-pair · gpt-3.5-turbo

references to groups of people and their statuses or actions

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

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

ggles

-0.77

ãĤ¦ãĤ¹

-0.74

 superiority

-0.70

iosyncr

-0.69

ãĤ¤ãĥĪ

-0.68

ebook

-0.67

izons

-0.66

à©

-0.64

 increments

-0.62

ķ

-0.61

POSITIVE LOGITS

 indicted

1.57

 arrested

1.56

 detained

1.54

 implicated

1.50

 convicted

1.41

 involved

1.39

 killed

1.39

 jailed

1.37

 slain

1.35

 abducted

1.31

Activations Density 0.324%

proper nouns and numbers related to news events such as people's names, nationalities, and statistics

references to groups of people and their statuses or actions

No Comments

No Known Activations