INDEX

Explanations

references to secretive or confidential information

oai_token-act-pair · gpt-3.5-turbo

occurrences of the term "secret."

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

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

avers

-0.77

older

-0.77

annis

-0.74

©¶æ

-0.68

 brim

-0.68

lihood

-0.68

gaard

-0.67

ammers

-0.67

ð

-0.67

È

-0.67

POSITIVE LOGITS

 secret

1.18

secret

1.10

Secret

1.03

rets

1.01

 secrets

0.99

arial

0.92

ariat

0.88

 informant

0.83

 hidden

0.80

ulously

0.79

Activations Density 0.009%

references to secretive or confidential information

occurrences of the term "secret."

No Comments

No Known Activations