INDEX

Explanations

words related to coded languages or encryption

oai_token-act-pair · gpt-3.5-turbo

instances of the letter "h" and other similar characters

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 5-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.5.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.5.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

 Cald

-0.65

 Vers

-0.59

 Barn

-0.57

 takeoff

-0.57

 Verse

-0.57

 compliment

-0.56

 Ascension

-0.55

 Suns

-0.55

ocular

-0.55

 Reboot

-0.54

POSITIVE LOGITS

arios

0.87

ateg

0.81

ificant

0.80

nikov

0.80

raper

0.74

aught

0.73

oresc

0.73

everal

0.73

uler

0.72

cific

0.72

Activations Density 0.214%

words related to coded languages or encryption

instances of the letter "h" and other similar characters

No Comments

No Known Activations