INDEX

Explanations

words related to legal matters and court cases, as well as possibly names

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_fs12288-jb

Configuration

jbloom/GPT2-Small-Feature-Splitting-Experiment-Layer-8/blocks.8.hook_resid_pre_12288

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

12,288

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

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

amar

-0.76

Brexit

-0.72

Bloomberg

-0.67

tackle

-0.67

atz

-0.67

>>>>>>>>

-0.67

âĢİ

-0.64

Turkey

-0.63

 AMERICA

-0.62

mart

-0.62

POSITIVE LOGITS

 slightest

1.46

 wearer

1.35

 same

1.28

 entire

1.25

 entirety

1.24

 respective

1.21

 remainder

1.20

 offending

1.19

 recipient

1.17

 desired

1.17

Activations Density 0.828%

words related to legal matters and court cases, as well as possibly names

No Comments

No Known Activations