INDEX

Explanations

the word "Off" followed by a numeric value indicating a level or setting

oai_token-act-pair · gpt-3.5-turbo

instances of the word "Off" in various formats and contexts

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

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

 chain

-0.69

 bureaucracy

-0.66

 hierarchy

-0.65

 vain

-0.64

 precious

-0.63

 painstaking

-0.62

 mushroom

-0.61

 ageing

-0.61

tea

-0.60

 unification

-0.60

POSITIVE LOGITS

Off

3.89

OFF

2.28

Off

2.24

off

2.04

OFF

1.96

offs

1.75

 Offense

1.30

offer

1.27

 Offer

1.26

Def

1.20

Activations Density 0.009%

the word "Off" followed by a numeric value indicating a level or setting

instances of the word "Off" in various formats and contexts

No Comments

No Known Activations