pub fn build_dequantize_linear_node(
names: &DequantLinearNames,
axis: Option<usize>,
) -> NodeProtoExpand description
Build a DequantizeLinear NodeProto.
ONNX spec (opset ≥ 10): inputs = [x (INT8), x_scale (FP32), x_zero_point (INT8)] outputs = [y (FP32)] y = (x - x_zero_point) × x_scale
When axis is Some(a), the axis attribute is set on the node,
enabling per-channel dequantization (opset ≥ 13).