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

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

Neuron 96 in Layer 2

Load this data into an Interactive Neuroscope

See Documentation here

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

Text #0

Max Range: 0.0541. Min Range: -0.0541

Max Act: 0.0515. Min Act: -0.0003

Data Index: 75028 (C4 (Web Text))

Max Activating Token Index: 752

Text #1

Max Range: 0.0541. Min Range: -0.0541

Max Act: 0.0540. Min Act: -0.0003

Data Index: 916740 (C4 (Web Text))

Max Activating Token Index: 433

Text #2

Max Range: 0.0541. Min Range: -0.0541

Max Act: 0.0541. Min Act: -0.0003

Data Index: 86390 (C4 (Web Text))

Max Activating Token Index: 540

Text #3

Max Range: 0.0541. Min Range: -0.0541

Max Act: 0.0500. Min Act: -0.0003

Data Index: 585658 (C4 (Web Text))

Max Activating Token Index: 737

Text #4

Max Range: 0.0541. Min Range: -0.0541

Max Act: 0.0504. Min Act: -0.0004

Data Index: 298585 (C4 (Web Text))

Max Activating Token Index: 663

Text #5

Max Range: 0.0541. Min Range: -0.0541

Max Act: 0.0476. Min Act: -0.0003

Data Index: 248886 (C4 (Web Text))

Max Activating Token Index: 1019

Text #6

Max Range: 0.0541. Min Range: -0.0541

Max Act: 0.0500. Min Act: -0.0003

Data Index: 432723 (C4 (Web Text))

Max Activating Token Index: 126

Text #7

Max Range: 0.0541. Min Range: -0.0541

Max Act: 0.0499. Min Act: -0.0003

Data Index: 271333 (C4 (Web Text))

Max Activating Token Index: 260

Text #8

Max Range: 0.0541. Min Range: -0.0541

Max Act: 0.0473. Min Act: -0.0002

Data Index: 1361512 (Python Code)

Max Activating Token Index: 52

Text #9

Max Range: 0.0541. Min Range: -0.0541

Max Act: 0.0475. Min Act: -0.0003

Data Index: 441983 (C4 (Web Text))

Max Activating Token Index: 67

Text #10

Max Range: 0.0541. Min Range: -0.0541

Max Act: 0.0472. Min Act: -0.0003

Data Index: 117212 (C4 (Web Text))

Max Activating Token Index: 149

Text #11

Max Range: 0.0541. Min Range: -0.0541

Max Act: 0.0465. Min Act: -0.0003

Data Index: 1562269 (Python Code)

Max Activating Token Index: 777

Text #12

Max Range: 0.0541. Min Range: -0.0541

Max Act: 0.0479. Min Act: -0.0003

Data Index: 868092 (C4 (Web Text))

Max Activating Token Index: 98

Text #13

Max Range: 0.0541. Min Range: -0.0541

Max Act: 0.0465. Min Act: -0.0003

Data Index: 500100 (C4 (Web Text))

Max Activating Token Index: 421

Text #14

Max Range: 0.0541. Min Range: -0.0541

Max Act: 0.0484. Min Act: -0.0004

Data Index: 579049 (C4 (Web Text))

Max Activating Token Index: 982

Text #15

Max Range: 0.0541. Min Range: -0.0541

Max Act: 0.0454. Min Act: -0.0003

Data Index: 396775 (C4 (Web Text))

Max Activating Token Index: 327

Text #16

Max Range: 0.0541. Min Range: -0.0541

Max Act: 0.0459. Min Act: -0.0003

Data Index: 57311 (C4 (Web Text))

Max Activating Token Index: 525

Text #17

Max Range: 0.0541. Min Range: -0.0541

Max Act: 0.0456. Min Act: -0.0004

Data Index: 616192 (C4 (Web Text))

Max Activating Token Index: 331

Text #18

Max Range: 0.0541. Min Range: -0.0541

Max Act: 0.0451. Min Act: -0.0003

Data Index: 1481885 (Python Code)

Max Activating Token Index: 850

Text #19

Max Range: 0.0541. Min Range: -0.0541

Max Act: 0.0448. Min Act: -0.0003

Data Index: 1199919 (C4 (Web Text))

Max Activating Token Index: 154