Neuron 1863 (Layer 0, mlp)

Universal number detector (decomposition case study)

Total weights 21

Inputs (W_in) 10

Outputs (W_out) 11

Input Weights (residual dim → neuron 1863)

Each input weight carries a different signal into the neuron. Click any row to see the full profile.

Rank	Weight	Indices	Max KL	Mean KL
#16	`blocks.0.mlp.W_in`	[464, 1863]	26.7662	0.009274
#23	`blocks.0.mlp.W_in`	[71, 1863]	22.7326	0.004690
#30	`blocks.0.mlp.W_in`	[395, 1863]	17.1137	0.002023
#34	`blocks.0.mlp.W_in`	[799, 1863]	16.0585	0.002183
#40	`blocks.0.mlp.W_in`	[578, 1863]	13.6505	0.005442
#103	`blocks.0.mlp.W_in`	[131, 1863]	8.0407	0.030843
#163	`blocks.0.mlp.W_in`	[678, 1863]	5.9731	0.004917
#1032	`blocks.0.mlp.W_in`	[87, 1863]	1.9184	0.000055
#2101	`blocks.0.mlp.W_in`	[558, 1863]	1.0570	0.000906
#3751	`blocks.0.mlp.W_in`	[299, 1863]	0.0121	0.000036

Each output weight routes the neuron's signal to a different downstream channel. Click any row to see the full profile.

Rank	Weight	Indices	Max KL	Mean KL
#718	`blocks.0.mlp.W_out`	[1863, 968]	2.4633	0.003398
#1166	`blocks.0.mlp.W_out`	[1863, 773]	1.7583	0.001808
#2617	`blocks.0.mlp.W_out`	[1863, 495]	0.8604	0.001381
#3180	`blocks.0.mlp.W_out`	[1863, 360]	0.6373	0.000885
#3694	`blocks.0.mlp.W_out`	[1863, 915]	0.4525	0.001033
#3714	`blocks.0.mlp.W_out`	[1863, 126]	0.4328	0.000428
#3745	`blocks.0.mlp.W_out`	[1863, 903]	0.2944	0.000787
#3747	`blocks.0.mlp.W_out`	[1863, 756]	0.2026	0.000323
#3748	`blocks.0.mlp.W_out`	[1863, 396]	0.1069	0.000297
#3749	`blocks.0.mlp.W_out`	[1863, 267]	0.0995	0.000183
#3750	`blocks.0.mlp.W_out`	[1863, 696]	0.0833	0.000105