INDEX

Explanations

phrases related to training or education

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

bum

-0.68

pard

-0.63

 Uruguay

-0.63

adra

-0.61

 headlines

-0.60

 extrad

-0.60

govtrack

-0.59

caps

-0.58

 headline

-0.58

rants

-0.58

POSITIVE LOGITS

 instructors

1.03

 regimen

1.01

 manuals

0.99

learn

0.98

Learning

0.96

 instructor

0.95

 Instruct

0.93

 instructional

0.92

 manual

0.91

 taught

0.91

Activations Density 5.551%

phrases related to training or education

No Comments

No Known Activations