Function mnist_valid 

pub fn mnist_valid<D: Float>() -> impl Fn(SymTensor) -> SymTensor

LeNet-5 (valid-convolution variant) — no padding, no bias, with final softmax. Use this for inference when you need class probabilities.

Input         [N,  1, 28, 28]
Conv2d(1→6,   5×5, pad=0) →  [N,  6, 24, 24]
ReLU
AvgPool2d(2×2, stride=2)  →  [N,  6, 12, 12]
Conv2d(6→16,  5×5, pad=0) →  [N, 16,  8,  8]
ReLU
AvgPool2d(2×2, stride=2)  →  [N, 16,  4,  4]
Flatten                   →  [N, 256]
Linear(256→120)
ReLU
Linear(120→84)
ReLU
Linear(84→10)
Softmax(dim=1)            →  [N, 10]