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.