INDEX

Explanations

phrases related to going against authority or the system

oai_token-act-pair · gpt-3.5-turbo Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 8-res_fs24576-jb

Configuration

jbloom/GPT2-Small-Feature-Splitting-Experiment-Layer-8/blocks.8.hook_resid_pre_24576

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

24,576

Data Type

torch.float32

Hook Point

blocks.8.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

 Cosponsors

-0.79

£ı

-0.69

aic

-0.68

Images

-0.66

reddits

-0.63

>:

-0.63

uid

-0.62

 recip

-0.62

MpServer

-0.61

icion

-0.61

POSITIVE LOGITS

 bounds

1.06

 existence

0.83

 nowhere

0.81

 hiber

0.79

 harms

0.78

 trouble

0.74

 shape

0.73

 sight

0.72

 captivity

0.72

bed

0.72

Activations Density 0.047%

phrases related to going against authority or the system

No Comments

No Known Activations