INDEX

Explanations

the word "persons" or variations of it

oai_token-act-pair · gpt-3.5-turbo

references to individuals or people

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

 Detention

-0.69

 Tube

-0.65

 tube

-0.65

 juice

-0.64

 Juice

-0.64

 Duke

-0.63

 bonus

-0.62

 Cancel

-0.61

gym

-0.61

 Gadget

-0.61

POSITIVE LOGITS

pers

4.69

Pers

1.97

person

1.50

perse

1.46

per

1.45

ping

1.36

people

1.34

men

1.17

pered

1.17

 Pers

1.16

Activations Density 0.011%

the word "persons" or variations of it

references to individuals or people

No Comments

No Known Activations