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

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

Neuron 2016 in Layer 1

Load this data into an Interactive Neuroscope

See Documentation here

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

Text #0

Max Range: 3.8491. Min Range: -3.8491

Max Act: 3.8491. Min Act: -0.1700

Data Index: 29789 (C4 (Web Text))

Max Activating Token Index: 957

Text #1

Max Range: 3.8491. Min Range: -3.8491

Max Act: 3.8178. Min Act: -0.1700

Data Index: 765792 (C4 (Web Text))

Max Activating Token Index: 22

Text #2

Max Range: 3.8491. Min Range: -3.8491

Max Act: 3.8178. Min Act: -0.1700

Data Index: 538667 (C4 (Web Text))

Max Activating Token Index: 120

Text #3

Max Range: 3.8491. Min Range: -3.8491

Max Act: 3.7553. Min Act: -0.1700

Data Index: 560200 (C4 (Web Text))

Max Activating Token Index: 928

Text #4

Max Range: 3.8491. Min Range: -3.8491

Max Act: 3.7553. Min Act: -0.1700

Data Index: 1186066 (C4 (Web Text))

Max Activating Token Index: 907

Text #5

Max Range: 3.8491. Min Range: -3.8491

Max Act: 3.6927. Min Act: -0.1700

Data Index: 343406 (C4 (Web Text))

Max Activating Token Index: 518

Text #6

Max Range: 3.8491. Min Range: -3.8491

Max Act: 3.7240. Min Act: -0.1700

Data Index: 1594073 (Python Code)

Max Activating Token Index: 790

Text #7

Max Range: 3.8491. Min Range: -3.8491

Max Act: 3.6927. Min Act: -0.1700

Data Index: 898080 (C4 (Web Text))

Max Activating Token Index: 473

Text #8

Max Range: 3.8491. Min Range: -3.8491

Max Act: 3.7240. Min Act: -0.1700

Data Index: 854943 (C4 (Web Text))

Max Activating Token Index: 776

Text #9

Max Range: 3.8491. Min Range: -3.8491

Max Act: 3.6927. Min Act: -0.1700

Data Index: 1075061 (C4 (Web Text))

Max Activating Token Index: 190

Text #10

Max Range: 3.8491. Min Range: -3.8491

Max Act: 3.6927. Min Act: -0.1700

Data Index: 201909 (C4 (Web Text))

Max Activating Token Index: 968

Text #11

Max Range: 3.8491. Min Range: -3.8491

Max Act: 3.6927. Min Act: -0.1700

Data Index: 1621112 (Python Code)

Max Activating Token Index: 55

Text #12

Max Range: 3.8491. Min Range: -3.8491

Max Act: 3.6301. Min Act: -0.1700

Data Index: 1442640 (Python Code)

Max Activating Token Index: 779

Text #13

Max Range: 3.8491. Min Range: -3.8491

Max Act: 3.5988. Min Act: -0.1700

Data Index: 1493234 (Python Code)

Max Activating Token Index: 432

Text #14

Max Range: 3.8491. Min Range: -3.8491

Max Act: 3.6614. Min Act: -0.1700

Data Index: 1646216 (Python Code)

Max Activating Token Index: 907

Text #15

Max Range: 3.8491. Min Range: -3.8491

Max Act: 3.7240. Min Act: -0.1700

Data Index: 824943 (C4 (Web Text))

Max Activating Token Index: 130

Text #16

Max Range: 3.8491. Min Range: -3.8491

Max Act: 3.6301. Min Act: -0.1700

Data Index: 1094542 (C4 (Web Text))

Max Activating Token Index: 646

Text #17

Max Range: 3.8491. Min Range: -3.8491

Max Act: 3.6614. Min Act: -0.1700

Data Index: 53557 (C4 (Web Text))

Max Activating Token Index: 211

Text #18

Max Range: 3.8491. Min Range: -3.8491

Max Act: 3.6301. Min Act: -0.1700

Data Index: 825852 (C4 (Web Text))

Max Activating Token Index: 801

Text #19

Max Range: 3.8491. Min Range: -3.8491

Max Act: 3.6614. Min Act: -0.1700

Data Index: 633985 (C4 (Web Text))

Max Activating Token Index: 743