pub fn qsa_block_pool(
gpu: &dyn GpuBackend,
kernel: KernelHandle,
raw_keys: DevicePtr,
k_norm_w: DevicePtr,
block_keys: DevicePtr,
first_block: u32,
n_new: u32,
ratio: u32,
hd: u32,
rot: u32,
theta: f32,
eps: f32,
stream: u64,
) -> Result<()>Expand description
Pool n_new freshly complete blocks starting at first_block:
mean over ratio raw keys -> RMSNorm*(1+w) -> rope at block-start pos.