INDEX

Explanations

words related to the broadcasting or airing of content

oai_token-act-pair · gpt-3.5-turbo

instances of the word "aired" and its variations, indicating content related to broadcasts or television shows

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

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

atos

-0.75

 keys

-0.69

 geometry

-0.65

 commands

-0.64

 productivity

-0.62

 dict

-0.61

oci

-0.61

rington

-0.61

keys

-0.61

ston

-0.60

POSITIVE LOGITS

 aired

3.70

 airing

2.73

 airs

2.30

 premiered

1.81

 broadcast

1.54

 televised

1.49

 premie

1.43

 filmed

1.35

aired

1.31

 screened

1.28

Activations Density 0.020%

words related to the broadcasting or airing of content

instances of the word "aired" and its variations, indicating content related to broadcasts or television shows

No Comments

No Known Activations