INDEX

Explanations

words related to criminal investigations, financial wrongdoing, and historical context

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_fs1536-jb

Configuration

jbloom/GPT2-Small-Feature-Splitting-Experiment-Layer-8/blocks.8.hook_resid_pre_1536

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

1,536

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

 contact

-0.51

 inaccessible

-0.51

 electronically

-0.46

 coordinate

-0.46

 spontaneously

-0.44

 offline

-0.44

 inactive

-0.44

 permanently

-0.44

 actively

-0.44

bil

-0.44

POSITIVE LOGITS

 understatement

0.62

 irony

0.58

ÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤÃĥÃĤ

0.56

quickShipAvailable

0.56

ortunately

0.54

surprisingly

0.54

 guessed

0.54

 nutshell

0.53

haps

0.52

udder

0.52

Activations Density 7.288%

words related to criminal investigations, financial wrongdoing, and historical context

No Comments

No Known Activations