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

crew

-0.70

started

-0.67

craft

-0.67

eus

-0.66

 clicks

-0.60

 started

-0.60

reddits

-0.59

BUG

-0.59

bean

-0.57

alis

-0.57

POSITIVE LOGITS

 Adin

0.89

ĪĴ

0.83

©¶æ¥µ

0.83

¿½

0.72

 unden

0.72

 Parables

0.72

¥µ

0.71

 Rivals

0.69

enty

0.69

©¶æ

0.67

Activations Density 0.000%

No Known Activations

This feature has no known activations.