INDEX

Explanations

mentions of the concept of "winning"

oai_token-act-pair · gpt-3.5-turbo

instances of the word "win"

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

 Sentinel

-0.74

 charg

-0.74

 unauthorized

-0.73

 underground

-0.72

 charging

-0.69

 charged

-0.68

 ambient

-0.66

 Anthem

-0.65

 charges

-0.64

 satirical

-0.64

POSITIVE LOGITS

win

4.51

WIN

2.27

Win

2.10

winner

1.75

winning

1.52

Win

1.49

win

1.41

wine

1.36

WIN

1.31

wyn

1.31

Activations Density 0.006%

mentions of the concept of "winning"

instances of the word "win"

No Comments

No Known Activations