glm_hc_post

Function glm_hc_post 

Source
pub fn glm_hc_post(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    block_out: DevicePtr,
    residual: DevicePtr,
    post: DevicePtr,
    comb: DevicePtr,
    out: DevicePtr,
    num_tokens: u32,
    hidden_size: u32,
    hc_mult: u32,
    stream: u64,
) -> Result<()>
Expand description

glm_hc_post: out[j] = post[j] * block_out + Σ_i comb[i][j] * residual[i].

residual is the pre-hc_pre stream snapshot, block_out this site’s sublayer output. out may alias streams — each output stream is a fresh combination, so writing back over the highway is the intended flow.