INDEX

Explanations

numeric quantities expressed as whole numbers

oai_token-act-pair · gpt-3.5-turbo

occurrences of the word "number" indicating statistical data or counts

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

endon

-0.70

 Passage

-0.66

 Despair

-0.66

INTON

-0.66

robe

-0.64

rador

-0.62

Materials

-0.62

axter

-0.62

dit

-0.62

 trib

-0.61

POSITIVE LOGITS

 number

0.86

 numbers

0.84

number

0.83

NUM

0.74

plate

0.73

naires

0.73

 Numbers

0.72

metry

0.72

population

0.70

Activations Density 0.037%

numeric quantities expressed as whole numbers

occurrences of the word "number" indicating statistical data or counts

No Comments

No Known Activations