INDEX

Explanations

This neuron seems to be a bit confused and is activating on a variety of characters and symbols without a clear pattern

oai_token-act-pair · gpt-3.5-turbo

special characters or unique symbols

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

undai

-1.01

 mathemat

-0.87

sembly

-0.85

eatures

-0.83

rongh

-0.82

glomer

-0.82

ebus

-0.78

chio

-0.77

 showc

-0.76

uggest

-0.75

POSITIVE LOGITS

ãģ®

1.01

ãģ®æ

1.00

åº

0.99

ÙĨ

0.96

Ùĩ

0.94

çİĭ

0.94

ä¹ĭ

0.93

åŃĲ

0.93

é

0.93

æŃ¦

0.93

Activations Density 0.044%

Top

Activations

21.90

êµŃ ìĦłìĪĺëĵ¤ìĿ´ NASLìĹ

21.90

ìĦłìĪĺëĵ¤ìĿ´ NASLìĹĲ ë¶

21.77

´ìļ©ìĿĦ íĨµíķ´ íķľêµ

21.77

êµ¬ì²´ìłģìĿ¸ ëĤ´ìļ©

21.77

ģìĿ¸ ëĤ´ìļ©ìĿĦ íĨµíķ´

21.77

¸ ëĤ´ìļ©ìĿĦ íĨµíķ´ íķ

19.27

¶Īì°¸íķĺê²Į ëĲľ ìĿ´ì

19.27

¸íķĺê²Į ëĲľ ìĿ´ì<|endoftext|>Rep. St

19.27

Ĳ ë¶Īì°¸íķĺê²Į ëĲľ ìĿ

18.84

¸åĴĮè§£å¾¡çĶ¨; 1811).[9]

18.84

æĽ¸åĴĮè§£å¾¡çĶ¨; 1811).[

18.62

åľ¾èĬĤçĽ®æĹ©çĤ¹åģľåĲ§

18.62

®æĹ©çĤ¹åģľåĲ§ — YinVikki (@

18.33

ä¸ľåĮĹçİ©æ³¥å·´" ("I'm playing clay

18.33

name, "æĪĳåľ¨ä¸ľåĮĹçİ©æ³¥å·

18.33

åľ¨ä¸ľåĮĹçİ©æ³¥å·´" ("I'm

18.26

íĨµíķ´ íķľêµŃ ìĦłìĪ

18.22

ŀç¬ĳæİĴè¡Įæ¦ľ: A restaurant in Cheng

18.22

ĳæİĴè¡Įæ¦ľ: A restaurant in Chengdu hired foreign

17.36

´æĴŃç²īä¸Ŀä¹Łä¸įçĶ¨æĬķ

17.36

ĭä¸įåĪ°çĽ´æĴŃç²īä¸Ŀä¹Łä¸į

17.36

°çĽ´æĴŃç²īä¸Ŀä¹Łä¸įçĶ¨æ

17.27

łåľ¨ä¸ŃåĽ½åģľæĴŃåĩłåĳ¨ðŁĺ

17.23

showæį®è¯´æĺ¯åĽłä¸ºä¸įåı¯æĬĹ

17.23

łä¸ºä¸įåı¯æĬĹæĭĴçļĦæĶ¿æ

16.77

Ŀä¹Łä¸įçĶ¨æĬķç¥¨äºĨãĢĤ<|endoftext|> America

16.02

"X-Wing :×ŀ×ľ×Ĺ×ŀ×ª ×Ķ×Ľ×ķ×

16.02

×ľ×Ĺ×ŀ×ª ×Ķ×Ľ×ķ×Ľ×ĳ×Ļ×Ŀ" --

16.01

Office (æĺĮå¹³åĿĤåŃ¦åķıæīĢ;

15.80

Korea ban order<|endoftext|>è¿Ļåŀĥåľ¾èĬĤçĽ®æĹ

15.34

ĭĴçļĦæĶ¿æ²»åİŁåĽłåľ¨ä¸ŃåĽ

15.31

å°ıéĻ¢åŃĲï¼ļ↵
↵
Tisk tisk, seeing

14.55

?↵
↵
æĽ¾éĿĻèĸĩ_ï¼ļ↵

14.37

innÅį (ä½ĵä»ģè¦ªçİĭ).[5] He was also known

13.17

²ÑĭÐ¹ Ð·Ð°ÑīÐ¸ÑĤÐ½ÑĭÐ¹ Ð¿

12.29

ç½ĳç¾İçŀ³åĸµï¼ļ↵
↵

12.05

ĩÐµÑģÐºÐ¸ Ð¾Ð±ÑĬÐµÐ´Ð¸Ð½ÐµÐ½

12.05

ÑģÐºÐ¸ Ð¾Ð±ÑĬÐµÐ´Ð¸Ð½ÐµÐ½Ð½Ñĭ

11.69

12th month (æ°¸æ²»åħĥå¹´;<|endoftext|> happened,

11.25

↵
å®ĺç½ĳç¾İçŀ³åĸµï¼

10.46

togetherâĢ¦↵
↵
åĸµéļ¾æıįï¼ļ↵
↵

9.32

Share ïģ¤ ïĤļ ïĤĻ ïĥķ ïĥł ï

9.21

Î£; Attic: á¼ĮÏĥÏĢÎµÎ½Î´Î¿ÏĤ

8.77

å¦¹åĦ¿-ï¼ļ↵
↵
There are guys saying this group

8.15

ï¾Ł(ï¾īÐĶ`)ï¾Ł+ It's

8.15

¡+ï¾Ł(ï¾īÐĶ`)ï¾Ł+

7.59

º ÐĲ. ÐĹÐ°ÑħÐ°ÑĢÐ¾Ð²Ð° - Citro

7.43

himself! (((ï¾ŁÐĶï¾Ł)))↵
↵
To show

5.91

ľèµµå®¶äººâĢĿ). New phrases pop up regularly online in

5.73

ivating thighs!!Î£(ï¾ŁâĪĢï¾Ł*)↵
↵

3.34

<|endoftext|> _0x446d=[âĢľ\x5F\x6D

3.14

↵
↵
{âĢľ_idâĢĿ : ObjectId( âĢľ4b