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

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

Neuron 768 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: 5.1644. Min Range: -5.1644

Max Act: 5.1644. Min Act: -0.1700

Data Index: 725402 (C4 (Web Text))

Max Activating Token Index: 428

Text #1

Max Range: 5.1644. Min Range: -5.1644

Max Act: 5.0707. Min Act: -0.1700

Data Index: 752195 (C4 (Web Text))

Max Activating Token Index: 468

Text #2

Max Range: 5.1644. Min Range: -5.1644

Max Act: 5.0394. Min Act: -0.1700

Data Index: 904417 (C4 (Web Text))

Max Activating Token Index: 677

Text #3

Max Range: 5.1644. Min Range: -5.1644

Max Act: 5.0394. Min Act: -0.1700

Data Index: 967662 (C4 (Web Text))

Max Activating Token Index: 602

Text #4

Max Range: 5.1644. Min Range: -5.1644

Max Act: 5.0707. Min Act: -0.1700

Data Index: 197989 (C4 (Web Text))

Max Activating Token Index: 780

Text #5

Max Range: 5.1644. Min Range: -5.1644

Max Act: 4.8207. Min Act: -0.1700

Data Index: 1633344 (Python Code)

Max Activating Token Index: 697

Text #6

Max Range: 5.1644. Min Range: -5.1644

Max Act: 4.8207. Min Act: -0.1700

Data Index: 119711 (C4 (Web Text))

Max Activating Token Index: 346

Text #7

Max Range: 5.1644. Min Range: -5.1644

Max Act: 4.7894. Min Act: -0.1700

Data Index: 546487 (C4 (Web Text))

Max Activating Token Index: 624

Text #8

Max Range: 5.1644. Min Range: -5.1644

Max Act: 4.7582. Min Act: -0.1700

Data Index: 1051865 (C4 (Web Text))

Max Activating Token Index: 1012

Text #9

Max Range: 5.1644. Min Range: -5.1644

Max Act: 4.7894. Min Act: -0.1700

Data Index: 1605420 (Python Code)

Max Activating Token Index: 856

Text #10

Max Range: 5.1644. Min Range: -5.1644

Max Act: 4.8207. Min Act: -0.1700

Data Index: 724423 (C4 (Web Text))

Max Activating Token Index: 793

Text #11

Max Range: 5.1644. Min Range: -5.1644

Max Act: 4.7269. Min Act: -0.1700

Data Index: 251119 (C4 (Web Text))

Max Activating Token Index: 919

Text #12

Max Range: 5.1644. Min Range: -5.1644

Max Act: 4.7269. Min Act: -0.1700

Data Index: 964818 (C4 (Web Text))

Max Activating Token Index: 931

Text #13

Max Range: 5.1644. Min Range: -5.1644

Max Act: 4.6644. Min Act: -0.1700

Data Index: 327294 (C4 (Web Text))

Max Activating Token Index: 912

Text #14

Max Range: 5.1644. Min Range: -5.1644

Max Act: 4.6957. Min Act: -0.1700

Data Index: 1311335 (C4 (Web Text))

Max Activating Token Index: 494

Text #15

Max Range: 5.1644. Min Range: -5.1644

Max Act: 4.6332. Min Act: -0.1700

Data Index: 1414308 (Python Code)

Max Activating Token Index: 548

Text #16

Max Range: 5.1644. Min Range: -5.1644

Max Act: 4.6019. Min Act: -0.1700

Data Index: 1648342 (Python Code)

Max Activating Token Index: 829

Text #17

Max Range: 5.1644. Min Range: -5.1644

Max Act: 4.6332. Min Act: -0.1700

Data Index: 588834 (C4 (Web Text))

Max Activating Token Index: 987

Text #18

Max Range: 5.1644. Min Range: -5.1644

Max Act: 4.7269. Min Act: -0.1700

Data Index: 596762 (C4 (Web Text))

Max Activating Token Index: 450

Text #19

Max Range: 5.1644. Min Range: -5.1644

Max Act: 4.6019. Min Act: -0.1700

Data Index: 863189 (C4 (Web Text))

Max Activating Token Index: 221