Skip to main content

gen_qwen38_decode_batch_tp_lg

Function gen_qwen38_decode_batch_tp_lg 

Source
pub fn gen_qwen38_decode_batch_tp_lg(
    cfg: &Qwen38Cfg,
    dir: &str,
    ctx: usize,
    seqs: &[(usize, usize)],
    tp: usize,
) -> TpAuthored
Expand description

A batched decode step, tensor-parallel over tp chips. See author::tp_decode_batch.