pub fn hc_head_mean(
gpu: &dyn GpuBackend,
kernel: KernelHandle,
streams: DevicePtr,
y_out: DevicePtr,
num_tokens: u32,
hidden_size: u32,
hc_mult: u32,
stream: u64,
) -> Result<()>Expand description
Final collapse before the LM head: an unweighted mean over the hc_mult streams.
🔴 Deliberately takes NO weight pointers. GLM’s Glm5NextTextHyperHead has no parameters and
the checkpoint carries zero hc_head tensors; DeepSeek-V4’s ops::hc_head reads
hc_head.{fn,base,scale}. The absent arguments are the guard against reaching for weights
that do not exist.