KolmogorovArnoldNetworks.jl is a Julia library providing implementations of Kolmogorov-Arnold neural networks. Implementation inspired by Efficient KAN.
Please feel free to contribute!
Current error:
KANLinear forward - input x shape: (784, 128)
b_splines - x shape: (784, 128), grid shape: (784, 12)
b_splines - bases shape: (784, 128, 8)
KANLinear forward - base_output shape: (128, 128), spline_output shape: (128, 128)
KANLinear forward - base_output shape: (128, 128), spline_output shape: (128, 128)
KANLinear forward - input x shape: (128, 128)
b_splines - x shape: (128, 128), grid shape: (128, 12)
b_splines - bases shape: (128, 128, 8)
KANLinear forward - base_output shape: (64, 128), spline_output shape: (64, 128)
KANLinear forward - base_output shape: (64, 128), spline_output shape: (64, 128)
KANLinear forward - input x shape: (64, 128)
b_splines - x shape: (64, 128), grid shape: (64, 12)
b_splines - bases shape: (64, 128, 8)
KANLinear forward - base_output shape: (10, 128), spline_output shape: (10, 128)
KANLinear forward - base_output shape: (10, 128), spline_output shape: (10, 128)
KAN forward - final output shape: (10, 128)
ERROR: DomainError with Loss is NaN on data item 1, stopping training:
You can install KolmogorovArnoldNetworks.jl using the Julia package manager. From the Julia REPL, type ]
to enter the package manager mode, then run:
pkg> add KolmogorovArnoldNetworks