INDEX

Explanations

phrases indicating instructions or conditions that need to be followed

oai_token-act-pair · gpt-3.5-turbo

phrases that directly address the reader or user

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

åħī

-0.70

interstitial

-0.66

Georg

-0.65

iah

-0.62

WOR

-0.62

¿½

-0.60

 adolesc

-0.60

Contents

-0.60

Writ

-0.58

Appearances

-0.57

POSITIVE LOGITS

're

1.49

've

1.20

 intend

1.18

 want

1.16

 prefer

1.14

 haven

1.12

 wish

1.09

'd

1.05

are

1.02

 aren

1.00

Activations Density 0.094%

phrases indicating instructions or conditions that need to be followed

phrases that directly address the reader or user

No Comments

No Known Activations