INDEX

Explanations

No Explanations Found

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 3-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.3.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.3.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

conservancy

-0.77

cock

-0.75

minus

-0.71

ibi

-0.69

ensor

-0.67

abis

-0.67

learn

-0.66

forums

-0.66

 reddit

-0.65

washer

-0.65

POSITIVE LOGITS

 indistinguishable

0.70

Gut

0.67

Lob

0.66

 Loaded

0.65

 Burg

0.64

 Dempsey

0.64

lihood

0.64

Ago

0.64

 Moroc

0.59

 Bunker

0.58

Activations Density 0.000%

No Known Activations

This feature has no known activations.