Model: GELU Model: 3 Layers, 2048 Neurons per Layer

Dataset: 80% C4 (Web Text) and 20% Python Code

Neuron 78 in Layer 2

Load this data into an Interactive Neuroscope

See Documentation here

Transformer Lens Loading: HookedTransformer.from_pretrained('gelu-3l')

Text #0

Max Range: 8.1087. Min Range: -8.1087

Max Act: 8.1087. Min Act: -0.1693

Data Index: 1149001 (C4 (Web Text))

Max Activating Token Index: 916

Text #1

Max Range: 8.1087. Min Range: -8.1087

Max Act: 7.9837. Min Act: -0.1700

Data Index: 943926 (C4 (Web Text))

Max Activating Token Index: 669

Text #2

Max Range: 8.1087. Min Range: -8.1087

Max Act: 7.9837. Min Act: -0.1695

Data Index: 737589 (C4 (Web Text))

Max Activating Token Index: 416

Text #3

Max Range: 8.1087. Min Range: -8.1087

Max Act: 7.5774. Min Act: -0.1700

Data Index: 1553834 (Python Code)

Max Activating Token Index: 731

Text #4

Max Range: 8.1087. Min Range: -8.1087

Max Act: 7.5774. Min Act: -0.1700

Data Index: 434270 (C4 (Web Text))

Max Activating Token Index: 155

Text #5

Max Range: 8.1087. Min Range: -8.1087

Max Act: 7.4212. Min Act: -0.1700

Data Index: 318183 (C4 (Web Text))

Max Activating Token Index: 877

Text #6

Max Range: 8.1087. Min Range: -8.1087

Max Act: 7.2962. Min Act: -0.1700

Data Index: 536074 (C4 (Web Text))

Max Activating Token Index: 418

Text #7

Max Range: 8.1087. Min Range: -8.1087

Max Act: 7.2962. Min Act: -0.1700

Data Index: 1193296 (C4 (Web Text))

Max Activating Token Index: 838

Text #8

Max Range: 8.1087. Min Range: -8.1087

Max Act: 7.2337. Min Act: -0.1700

Data Index: 503005 (C4 (Web Text))

Max Activating Token Index: 1000

Text #9

Max Range: 8.1087. Min Range: -8.1087

Max Act: 7.2649. Min Act: -0.1700

Data Index: 1575858 (Python Code)

Max Activating Token Index: 855

Text #10

Max Range: 8.1087. Min Range: -8.1087

Max Act: 7.2024. Min Act: -0.1700

Data Index: 569077 (C4 (Web Text))

Max Activating Token Index: 882

Text #11

Max Range: 8.1087. Min Range: -8.1087

Max Act: 7.1399. Min Act: -0.1700

Data Index: 424716 (C4 (Web Text))

Max Activating Token Index: 1008

Text #12

Max Range: 8.1087. Min Range: -8.1087

Max Act: 7.1399. Min Act: -0.1700

Data Index: 1566943 (Python Code)

Max Activating Token Index: 163

Text #13

Max Range: 8.1087. Min Range: -8.1087

Max Act: 7.1399. Min Act: -0.1700

Data Index: 1440817 (Python Code)

Max Activating Token Index: 921

Text #14

Max Range: 8.1087. Min Range: -8.1087

Max Act: 7.1399. Min Act: -0.1700

Data Index: 1646878 (Python Code)

Max Activating Token Index: 990

Text #15

Max Range: 8.1087. Min Range: -8.1087

Max Act: 7.1087. Min Act: -0.1700

Data Index: 1519409 (Python Code)

Max Activating Token Index: 478

Text #16

Max Range: 8.1087. Min Range: -8.1087

Max Act: 7.0774. Min Act: -0.1699

Data Index: 605238 (C4 (Web Text))

Max Activating Token Index: 763

Text #17

Max Range: 8.1087. Min Range: -8.1087

Max Act: 7.0149. Min Act: -0.1700

Data Index: 1621987 (Python Code)

Max Activating Token Index: 58

Text #18

Max Range: 8.1087. Min Range: -8.1087

Max Act: 6.9837. Min Act: -0.1700

Data Index: 145242 (C4 (Web Text))

Max Activating Token Index: 475

Text #19

Max Range: 8.1087. Min Range: -8.1087

Max Act: 6.9524. Min Act: -0.1700

Data Index: 136550 (C4 (Web Text))

Max Activating Token Index: 544