INDEX

Explanations

abbreviated names or initials of individuals, potentially related to interviews or news articles

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

holders

-0.96

holder

-0.81

 Illum

-0.68

houses

-0.65

doors

-0.62

cluding

-0.61

 constit

-0.58

 rooting

-0.57

rights

-0.56

 dictators

-0.56

POSITIVE LOGITS

ournals

1.30

ealous

1.29

upiter

1.23

igsaw

1.21

unction

1.19

umbo

1.10

okers

1.06

itsu

1.04

utsu

1.04

ansson

1.03

Activations Density 1.663%

abbreviated names or initials of individuals, potentially related to interviews or news articles

No Comments

No Known Activations