INDEX

Explanations

phrases marked with the abbreviation 'WH'

oai_token-act-pair · gpt-3.5-turbo

occurrences of repeated or emphasized "WH" phrases and the term "Monitor."

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

opausal

-0.64

 orbiting

-0.64

 default

-0.64

 shred

-0.64

 racer

-0.63

bru

-0.63

 fucking

-0.62

 Rolls

-0.62

 regression

-0.61

 shuff

-0.61

POSITIVE LOGITS

WH

2.33

 Monitor

1.65

azar

1.27

itta

1.11

ritz

0.99

Screen

0.96

CHR

0.94

ACP

0.94

YA

0.84

 Chambers

0.81

Activations Density 0.023%

phrases marked with the abbreviation 'WH'

occurrences of repeated or emphasized "WH" phrases and the term "Monitor."

No Comments

No Known Activations