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

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

Neuron 1904 in Layer 1

Load this data into an Interactive Neuroscope

See Documentation here

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

Text #0

Max Range: 3.6986. Min Range: -3.6986

Max Act: 3.6986. Min Act: -0.1700

Data Index: 548537 (C4 (Web Text))

Max Activating Token Index: 27

Text #1

Max Range: 3.6986. Min Range: -3.6986

Max Act: 3.3539. Min Act: -0.1700

Data Index: 1213498 (C4 (Web Text))

Max Activating Token Index: 812

Text #2

Max Range: 3.6986. Min Range: -3.6986

Max Act: 3.3539. Min Act: -0.1700

Data Index: 1510321 (Python Code)

Max Activating Token Index: 1020

Text #3

Max Range: 3.6986. Min Range: -3.6986

Max Act: 3.3853. Min Act: -0.1700

Data Index: 1566994 (Python Code)

Max Activating Token Index: 743

Text #4

Max Range: 3.6986. Min Range: -3.6986

Max Act: 3.3225. Min Act: -0.1700

Data Index: 1401539 (Python Code)

Max Activating Token Index: 979

Text #5

Max Range: 3.6986. Min Range: -3.6986

Max Act: 3.2911. Min Act: -0.1700

Data Index: 888377 (C4 (Web Text))

Max Activating Token Index: 486

Text #6

Max Range: 3.6986. Min Range: -3.6986

Max Act: 3.2282. Min Act: -0.1700

Data Index: 1509935 (Python Code)

Max Activating Token Index: 603

Text #7

Max Range: 3.6986. Min Range: -3.6986

Max Act: 3.2282. Min Act: -0.1700

Data Index: 1559924 (Python Code)

Max Activating Token Index: 222

Text #8

Max Range: 3.6986. Min Range: -3.6986

Max Act: 3.2125. Min Act: -0.1700

Data Index: 178342 (C4 (Web Text))

Max Activating Token Index: 165

Text #9

Max Range: 3.6986. Min Range: -3.6986

Max Act: 3.1968. Min Act: -0.1700

Data Index: 807096 (C4 (Web Text))

Max Activating Token Index: 591

Text #10

Max Range: 3.6986. Min Range: -3.6986

Max Act: 3.1810. Min Act: -0.1700

Data Index: 1126592 (C4 (Web Text))

Max Activating Token Index: 801

Text #11

Max Range: 3.6986. Min Range: -3.6986

Max Act: 3.1810. Min Act: -0.1700

Data Index: 1410019 (Python Code)

Max Activating Token Index: 61

Text #12

Max Range: 3.6986. Min Range: -3.6986

Max Act: 3.1810. Min Act: -0.1700

Data Index: 1508463 (Python Code)

Max Activating Token Index: 975

Text #13

Max Range: 3.6986. Min Range: -3.6986

Max Act: 3.1653. Min Act: -0.1700

Data Index: 1599579 (Python Code)

Max Activating Token Index: 830

Text #14

Max Range: 3.6986. Min Range: -3.6986

Max Act: 3.1653. Min Act: -0.1700

Data Index: 1543800 (Python Code)

Max Activating Token Index: 267

Text #15

Max Range: 3.6986. Min Range: -3.6986

Max Act: 3.1496. Min Act: -0.1700

Data Index: 1528007 (Python Code)

Max Activating Token Index: 123

Text #16

Max Range: 3.6986. Min Range: -3.6986

Max Act: 3.1496. Min Act: -0.1700

Data Index: 87802 (C4 (Web Text))

Max Activating Token Index: 926

Text #17

Max Range: 3.6986. Min Range: -3.6986

Max Act: 3.1496. Min Act: -0.1700

Data Index: 1381064 (Python Code)

Max Activating Token Index: 40

Text #18

Max Range: 3.6986. Min Range: -3.6986

Max Act: 3.1180. Min Act: -0.1700

Data Index: 372540 (C4 (Web Text))

Max Activating Token Index: 759

Text #19

Max Range: 3.6986. Min Range: -3.6986

Max Act: 3.1810. Min Act: -0.1700

Data Index: 445496 (C4 (Web Text))

Max Activating Token Index: 269