INDEX

Explanations

numerical patterns in the format of "digit-digit" with potential accompanying terms

oai_token-act-pair · gpt-3.5-turbo

phrases indicating the number of times someone has achieved a particular status or title

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

 Kimmel

-0.61

 liking

-0.60

IPM

-0.59

eas

-0.59

 Coul

-0.59

 Koen

-0.57

 overrun

-0.57

 Carnage

-0.55

 "$:/

-0.55

 Tomorrow

-0.54

POSITIVE LOGITS

time

1.54

term

1.18

timer

1.17

times

1.12

etime

1.00

year

0.97

Timer

0.95

generation

0.94

 timers

0.94

tim

0.90

Activations Density 0.101%

numerical patterns in the format of "digit-digit" with potential accompanying terms

phrases indicating the number of times someone has achieved a particular status or title

No Comments

No Known Activations