Skip to main content

compress_weights

Function compress_weights 

Source
pub fn compress_weights(
    weights: &[[f32; 4]],
) -> impl ExactSizeIterator<Item = CompressedWeights> + '_
Expand description

Quantize joint weights to Unorm16x4, clamping out-of-range components.

The caller is responsible for weights that already sum to 1; this helper does not renormalize. Computed on demand, so nothing is allocated.