INDEX

Explanations

words related to abuse or oppressive behavior

oai_token-act-pair · gpt-3.5-turbo

references to abusive behavior and relationships

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

ye

-0.76

Fest

-0.68

ometer

-0.67

 Wonders

-0.66

Lin

-0.63

ded

-0.62

lev

-0.62

 marvel

-0.61

ometers

-0.61

Mus

-0.61

POSITIVE LOGITS

 abusive

3.55

 abuser

1.90

 abusers

1.80

 oppressive

1.67

 manipulative

1.52

 abused

1.50

 abusing

1.46

 hateful

1.38

 objectionable

1.36

 dysfunctional

1.36

Activations Density 0.025%

words related to abuse or oppressive behavior

references to abusive behavior and relationships

No Comments

No Known Activations