INDEX

Explanations

instances of Japanese names

oai_token-act-pair · gpt-3.5-turbo

names and titles related to authority or leadership

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 1-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.1.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.1.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

izons

-0.75

etimes

-0.72

 Plane

-0.72

ORPG

-0.67

kell

-0.67

onic

-0.67

srf

-0.66

dimension

-0.66

uates

-0.64

ched

-0.63

POSITIVE LOGITS

Reviewer

0.82

ource

0.80

æ©

0.76

":""},{"

0.76

externalActionCode

0.75

DragonMagazine

0.74

HAEL

0.73

HOME

0.73

å§«

0.72

backs

0.71

Activations Density 0.031%

instances of Japanese names

names and titles related to authority or leadership

No Comments

No Known Activations