INDEX

Explanations

phrases related to contentious or divisive topics in politics or society

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

actionDate

-0.78

lehem

-0.76

iece

-0.72

nick

-0.71

nil

-0.70

ossibility

-0.70

wcsstore

-0.70

aniel

-0.70

DERR

-0.69

avi

-0.69

POSITIVE LOGITS

 humankind

1.15

 society

1.10

 mankind

1.09

 regards

1.03

 everyday

1.03

 modern

1.00

 contemporary

1.00

 ecosystems

0.99

 regard

0.98

 psyche

0.97

Activations Density 7.330%

phrases related to contentious or divisive topics in politics or society

No Comments

No Known Activations