INDEX

Explanations

words related to various names and places, potentially related to different individuals or locations

oai_token-act-pair · gpt-3.5-turbo

various forms of the letter 'H' or capital 'H' phrases

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

taboola

-0.85

ntil

-0.63

 Resp

-0.60

Ll

-0.59

 Expend

-0.59

 somew

-0.58

cc

-0.57

ioxide

-0.57

 reusable

-0.56

ħĭ

-0.56

POSITIVE LOGITS

abee

0.96

inki

0.88

icultural

0.85

kamp

0.81

azard

0.80

acan

0.79

awei

0.79

igans

0.78

iday

0.78

vard

0.77

Activations Density 0.111%

words related to various names and places, potentially related to different individuals or locations

various forms of the letter 'H' or capital 'H' phrases

No Comments

No Known Activations