INDEX

Explanations

capital letters followed by numbers, where the letter sequence "AF" is particularly strong in activation

oai_token-act-pair · gpt-3.5-turbo

references to specific organizations and acronyms

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 1-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.1.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.1.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

 Heller

-0.70

 Jenner

-0.68

 Gott

-0.68

 Kessler

-0.68

 Freak

-0.66

 Holt

-0.65

 McCull

-0.65

wikipedia

-0.64

 Walls

-0.63

ulative

-0.63

POSITIVE LOGITS

ornia

1.05

ntil

0.94

RA

0.85

avorite

0.85

htaking

0.83

SPA

0.82

AF

0.82

clair

0.81

unte

0.81

Æ

0.80

Activations Density 0.006%

capital letters followed by numbers, where the letter sequence "AF" is particularly strong in activation

references to specific organizations and acronyms

No Comments

No Known Activations