INDEX

Explanations

words related to crimes, shootings, and violence

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_fs768-jb

Configuration

jbloom/GPT2-Small-Feature-Splitting-Experiment-Layer-8/blocks.8.hook_resid_pre_768

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

768

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

algia

-0.71

 diaper

-0.65

aux

-0.65

Ala

-0.65

 Organ

-0.65

 Arbor

-0.65

ourced

-0.64

herty

-0.64

adian

-0.64

ç«

-0.63

POSITIVE LOGITS

powder

1.35

ãĥ£

1.01

linger

0.95

guns

0.94

smith

0.93

 rifles

0.91

fights

0.91

nikov

0.91

 rifle

0.89

 salute

0.88

Activations Density 4.632%

words related to crimes, shootings, and violence

No Comments

No Known Activations