INDEX

Explanations

references to news organizations, particularly "Examiner"

oai_token-act-pair · gpt-3.5-turbo

references to medical examiners

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 0-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.0.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.0.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

ijn

-0.75

Å¡

-0.74

Saharan

-0.66

xual

-0.66

bet

-0.65

bil

-0.64

icka

-0.64

 collaboration

-0.63

house

-0.63

kid

-0.63

POSITIVE LOGITS

 Examiner

4.14

 examiner

2.15

aminer

1.16

 coroner

0.98

ribune

0.81

 Applicant

0.79

Crash

0.79

 Examination

0.77

 Advocate

0.74

 Forensic

0.74

Activations Density 0.012%

references to news organizations, particularly "Examiner"

references to medical examiners

No Comments

No Known Activations