INDEX

Explanations

instances where a concept is emphasized as being on top of another concept

oai_token-act-pair · gpt-3.5-turbo

the phrase "on top" in various contexts

oai_token-act-pair · gpt-4o-mini Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 5-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.5.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.5.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

uld

-0.86

cci

-0.82

catentry

-0.78

æ©Ł

-0.66

chie

-0.64

yssey

-0.63

 pearl

-0.63

 Gaul

-0.62

mercial

-0.62

encers

-0.61

POSITIVE LOGITS

most

0.87

retty

0.80

ography

0.80

deck

0.80

 tier

0.74

mast

0.74

ology

0.69

ographical

0.69

flight

0.68

iary

0.67

Activations Density 0.011%

instances where a concept is emphasized as being on top of another concept

the phrase "on top" in various contexts

No Comments

No Known Activations