INDEX

Explanations

references to criminal or wrongful acts, often related to war crimes or atrocities

oai_token-act-pair · gpt-3.5-turbo

references to violations of human rights and crimes

oai_token-act-pair · gpt-4o-mini Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 10-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.10.hook_resid_pre

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

24,576

Data Type

torch.float32

Hook Point

blocks.10.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

Goo

-0.69

iew

-0.58

 Shining

-0.58

ortment

-0.57

trak

-0.56

ouf

-0.55

antz

-0.55

isode

-0.54

 reinforcement

-0.52

 Stuff

-0.52

POSITIVE LOGITS

by

0.98

BY

0.83

 concurrently

0.81

 consecut

0.79

 anew

0.77

 aloud

0.75

 unlawfully

0.74

 correctly

0.73

 domestically

0.72

 elsewhere

0.72

Activations Density 0.259%

references to criminal or wrongful acts, often related to war crimes or atrocities

references to violations of human rights and crimes

No Comments

No Known Activations