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

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

bara

-0.71

 loader

-0.71

 threaded

-0.70

 scrutiny

-0.69

Oracle

-0.67

 reciproc

-0.66

 acknow

-0.65

piv

-0.65

voc

-0.64

Wiki

-0.63

POSITIVE LOGITS

âĹ¼

0.88

engineering

0.80

 Ballard

0.77

arc

0.75

 Surv

0.72

 Allison

0.71

occ

0.68

atell

0.68

utical

0.67

Ala

0.67

Activations Density 0.000%

No Known Activations

This feature has no known activations.