INDEX

Explanations

the word "Contin" or variations of it, like "Continental" or "Continuation."

oai_token-act-pair · gpt-3.5-turbo

instances of the word "continue" and its variations, indicating a focus on ongoing processes or topics

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

shy

-0.74

 weigh

-0.72

 derby

-0.71

ilty

-0.66

orer

-0.66

 haircut

-0.66

uffle

-0.65

afia

-0.64

 bash

-0.64

hit

-0.63

POSITIVE LOGITS

Contin

3.78

contin

3.29

 Contin

2.34

 continuation

1.84

 contin

1.84

 CONTIN

1.78

 continuity

1.68

 Continuing

1.64

 Continued

1.49

 Continuous

1.48

Activations Density 0.023%

the word "Contin" or variations of it, like "Continental" or "Continuation."

instances of the word "continue" and its variations, indicating a focus on ongoing processes or topics

No Comments

No Known Activations