fn prefill_row_batches(
gpu: &mut Gpu<'_>,
cpu: Option<&CpuModel<'_>>,
cpu_state: Option<&mut State>,
ids: &[u32],
n_draft: usize,
argmax_lines: Option<&mut Vec<String>>,
) -> Result<PrefillResume>Expand description
Use the decode row kernels for short prompts and explicit row-path checks.