Model: GELU Model: 4 Layers, 2048 Neurons per Layer

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

Neuron 1998 in Layer 3

Load this data into an Interactive Neuroscope

See Documentation here

Transformer Lens Loading: HookedTransformer.from_pretrained('gelu-4l')

Text #0

Max Range: 7.1952. Min Range: -7.1952

Max Act: 7.1952. Min Act: -0.1696

Data Index: 1598589 (Python Code)

Max Activating Token Index: 1003

Text #1

Max Range: 7.1952. Min Range: -7.1952

Max Act: 6.9765. Min Act: -0.1700

Data Index: 967150 (C4 (Web Text))

Max Activating Token Index: 92

Text #2

Max Range: 7.1952. Min Range: -7.1952

Max Act: 6.8358. Min Act: -0.1700

Data Index: 391419 (C4 (Web Text))

Max Activating Token Index: 287

Text #3

Max Range: 7.1952. Min Range: -7.1952

Max Act: 6.8202. Min Act: -0.1700

Data Index: 1556647 (Python Code)

Max Activating Token Index: 842

Text #4

Max Range: 7.1952. Min Range: -7.1952

Max Act: 6.7265. Min Act: -0.1700

Data Index: 1477124 (Python Code)

Max Activating Token Index: 160

Text #5

Max Range: 7.1952. Min Range: -7.1952

Max Act: 6.6640. Min Act: -0.1700

Data Index: 1556467 (Python Code)

Max Activating Token Index: 420

Text #6

Max Range: 7.1952. Min Range: -7.1952

Max Act: 6.7733. Min Act: -0.1700

Data Index: 1494653 (Python Code)

Max Activating Token Index: 369

Text #7

Max Range: 7.1952. Min Range: -7.1952

Max Act: 6.7577. Min Act: -0.1700

Data Index: 1561865 (Python Code)

Max Activating Token Index: 401

Text #8

Max Range: 7.1952. Min Range: -7.1952

Max Act: 6.6952. Min Act: -0.1700

Data Index: 1247918 (C4 (Web Text))

Max Activating Token Index: 506

Text #9

Max Range: 7.1952. Min Range: -7.1952

Max Act: 6.6640. Min Act: -0.1700

Data Index: 391303 (C4 (Web Text))

Max Activating Token Index: 395

Text #10

Max Range: 7.1952. Min Range: -7.1952

Max Act: 6.6640. Min Act: -0.1700

Data Index: 1224426 (C4 (Web Text))

Max Activating Token Index: 479

Text #11

Max Range: 7.1952. Min Range: -7.1952

Max Act: 6.5390. Min Act: -0.1700

Data Index: 1409096 (Python Code)

Max Activating Token Index: 323

Text #12

Max Range: 7.1952. Min Range: -7.1952

Max Act: 6.6171. Min Act: -0.1700

Data Index: 1605217 (Python Code)

Max Activating Token Index: 964

Text #13

Max Range: 7.1952. Min Range: -7.1952

Max Act: 6.5546. Min Act: -0.1700

Data Index: 633633 (C4 (Web Text))

Max Activating Token Index: 825

Text #14

Max Range: 7.1952. Min Range: -7.1952

Max Act: 6.5546. Min Act: -0.1700

Data Index: 1181780 (C4 (Web Text))

Max Activating Token Index: 176

Text #15

Max Range: 7.1952. Min Range: -7.1952

Max Act: 6.5233. Min Act: -0.1700

Data Index: 319523 (C4 (Web Text))

Max Activating Token Index: 406

Text #16

Max Range: 7.1952. Min Range: -7.1952

Max Act: 6.5858. Min Act: -0.1700

Data Index: 672460 (C4 (Web Text))

Max Activating Token Index: 689

Text #17

Max Range: 7.1952. Min Range: -7.1952

Max Act: 6.4608. Min Act: -0.1700

Data Index: 1447958 (Python Code)

Max Activating Token Index: 492

Text #18

Max Range: 7.1952. Min Range: -7.1952

Max Act: 6.4921. Min Act: -0.1700

Data Index: 1396077 (Python Code)

Max Activating Token Index: 920

Text #19

Max Range: 7.1952. Min Range: -7.1952

Max Act: 6.4452. Min Act: -0.1700

Data Index: 1570964 (Python Code)

Max Activating Token Index: 744