qsa_score

Function qsa_score 

Source
pub fn qsa_score(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    q: DevicePtr,
    block_keys: DevicePtr,
    scores: DevicePtr,
    n_blocks: u32,
    n_heads: u32,
    hd: u32,
    stream: u64,
) -> Result<()>
Expand description

scores[b] = sum_h relu(q_h . k_b) / sqrt(hd) over n_blocks blocks.