INDEX

Explanations

words related to protection or safety

oai_token-act-pair · gpt-3.5-turbo

mentions of protective measures or items

oai_token-act-pair · gpt-4o-mini Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 1-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.1.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.1.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

zig

-0.83

avez

-0.79

Hour

-0.78

licks

-0.76

mitt

-0.70

Redditor

-0.70

hour

-0.69

VP

-0.68

upon

-0.68

eric

-0.68

POSITIVE LOGITS

 protective

1.21

 Protective

1.10

 shielding

0.96

ective

0.91

 shield

0.89

 defences

0.88

 protection

0.87

 enclosure

0.86

 shields

0.85

 defenses

0.83

Activations Density 0.004%

words related to protection or safety

mentions of protective measures or items

No Comments

No Known Activations