pub fn multi_token_attention_f32(
scores: &[f32],
weight: &[f32],
bias: Option<&[f32]>,
batch: usize,
channels_in: usize,
channels_out: usize,
seq_len: usize,
kernel_h: usize,
kernel_w: usize,
stride_h: usize,
stride_w: usize,
padding_h: usize,
padding_w: usize,
dilation_h: usize,
dilation_w: usize,
groups: usize,
sparse: bool,
) -> Vec<f32>