INDEX

Explanations

phrases related to crime and justice

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_fs1536-jb

Configuration

jbloom/GPT2-Small-Feature-Splitting-Experiment-Layer-8/blocks.8.hook_resid_pre_1536

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

1,536

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

staking

-0.61

fortunately

-0.60

 anecd

-0.58

 irony

-0.56

xtap

-0.55

 nods

-0.53

 aback

-0.53

 Subtle

-0.52

 consolation

-0.52

udos

-0.52

POSITIVE LOGITS

 ObamaCare

0.59

Islam

0.55

 vagina

0.55

thood

0.55

ratom

0.54

 Ebola

0.53

 Obamacare

0.53

 virginity

0.53

 sexual

0.53

versive

0.53

Activations Density 12.245%

phrases related to crime and justice

No Comments

No Known Activations