Skip to main content

tp_stacked_layer

Function tp_stacked_layer 

Source
pub fn tp_stacked_layer(
    spec: &ModelSpec,
    l: usize,
    dir: &str,
    ctx: usize,
    poss: &[usize],
    tp: usize,
) -> Result<TpAuthored, NotYet>
Expand description

Layer l of a batched tensor-parallel decode step alone.