INDEX

Explanations

locations or spatial relationships

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

 benefited

-0.67

FTWARE

-0.67

Pac

-0.66

aceutical

-0.66

HTTP

-0.66

Owner

-0.61

fill

-0.60

View

-0.59

 thri

-0.57

 WRITE

-0.57

POSITIVE LOGITS

 least

1.46

onement

1.16

yp

1.01

 halftime

0.98

abase

0.97

las

0.97

mosp

0.96

roph

0.95

rium

0.94

rial

0.92

Activations Density 1.587%

locations or spatial relationships

No Comments

No Known Activations