Function dequantize 

pub fn dequantize<'a>(
    w: impl AsRef<Array>,
    scales: impl AsRef<Array>,
    biases: impl Into<Option<&'a Array>>,
    options: DequantizeOptions,
) -> Result<Array, QuantizationError>

Reconstruct weights using the same mode, group size, and bits used to quantize them. Pass biases for affine weights and None for the floating-point modes. Floating-point modes dequantize to bfloat16; affine uses the scale/bias dtype.