Function ternary_gemm_gpu 

pub fn ternary_gemm_gpu(
    device: &Device,
    queue: &Queue,
    activations: &[f32],
    packed: &[u8],
    scale: f32,
    n_in: usize,
    n_out: usize,
    n_batch: usize,
) -> Vec<f32>

Execute the base-3 ternary GEMM on the GPU (ternary_gemm.wgsl).