pub fn hc_pre_site(
gpu: &dyn GpuBackend,
kernel: KernelHandle,
streams: DevicePtr,
site: &HcSiteWeights,
hc: &HcWeights,
y_out: DevicePtr,
post_out: DevicePtr,
comb_out: DevicePtr,
scratch: DevicePtr,
num_tokens: u32,
hidden_size: u32,
norm_eps: f32,
stream: u64,
) -> Result<()>Expand description
Collapse the streams to one and emit whatever the matching hc_post
needs — post+comb for Sinkhorn, the injection vector in post_out for
low-rank.