INDEX

Explanations

mentions of physical attacks or violent incidents

oai_token-act-pair · gpt-3.5-turbo

instances of the word "attacked" and its related forms

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

 Valve

-0.81

 savings

-0.65

 Mazda

-0.64

 Lifetime

-0.64

 vinyl

-0.63

 paperwork

-0.63

 monthly

-0.61

CFL

-0.60

DMV

-0.60

 Guinness

-0.59

POSITIVE LOGITS

 attacked

3.39

 assaulted

2.19

 invaded

1.91

 attack

1.89

 attacking

1.77

 attacks

1.77

Attack

1.65

attack

1.64

attacks

1.59

 harassed

1.59

Activations Density 0.012%

mentions of physical attacks or violent incidents

instances of the word "attacked" and its related forms

No Comments

No Known Activations