INDEX

Explanations

No Explanations Found

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

gaard

-0.69

gem

-0.68

 Improvement

-0.66

cor

-0.66

iciency

-0.64

 newsletter

-0.64

 HSBC

-0.63

 contam

-0.61

hra

-0.61

bit

-0.60

POSITIVE LOGITS

è¦

0.67

 rall

0.65

 carrots

0.63

Redditor

0.63

 apes

0.62

—-

0.62

 Roose

0.61

efully

0.60

Magikarp

0.60

 poop

0.60

Activations Density 0.000%

No Known Activations

This feature has no known activations.