INDEX

Explanations

negative statements or contradictions

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

kamp

-0.76

ixel

-0.68

USH

-0.66

å¥

-0.64

velt

-0.64

stakes

-0.63

éĥ

-0.63

Ĥİ

-0.61

æ©

-0.61

 Circuit

-0.60

POSITIVE LOGITS

 necessarily

1.54

icably

1.39

epad

1.31

icable

1.30

eworthy

1.20

withstanding

1.14

orious

1.10

hin

1.09

 exactly

0.98

 bothering

0.95

Activations Density 1.620%

negative statements or contradictions

No Comments

No Known Activations