INDEX

Explanations

No Explanations Found

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 0-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.0.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.0.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

hog

-0.79

arij

-0.70

cca

-0.70

ciplinary

-0.67

ira

-0.64

akening

-0.64

orbit

-0.62

zx

-0.61

izen

-0.60

 Friendly

-0.60

POSITIVE LOGITS

itaire

0.75

Redditor

0.71

 deform

0.66

è¦ļéĨĴ

0.66

kefeller

0.66

ÃŃs

0.64

 surg

0.64

 skelet

0.63

 conclud

0.63

 traged

0.62

Activations Density 0.000%

No Known Activations

This feature has no known activations.