decode_attention

metrale_model_arch::glm5next_dsa::attend

Function decode_attention 

Source 
pub fn decode_attention(
    gpu: &dyn GpuBackend,
    kernel: Glm5NextDsaDecodeKernel,
    cfg: &Glm5NextDsaConfig,
    geom: &DsaSelectGeometry,
    paging: &DsaDecodePaging,
    inputs: &DsaDecodeInputs,
    stream: u64,
) -> Result<()>
Expand description

2026-09-25: Launch the selected-index decode. Enqueued on stream, not synchronised.

geom.q_rows must equal paging.num_seqs: one query row per sequence. A mismatch would index the selection rows with the wrong stride, so it is an error.