pub fn tensor_plan(
name: &str,
mixer: MixerKind,
_mlp: MlpKind,
config: &ModelConfig,
) -> (TpShardKind, usize, usize)Expand description
(kind, full_out, full_in) for one checkpoint key.
pub fn tensor_plan(
name: &str,
mixer: MixerKind,
_mlp: MlpKind,
config: &ModelConfig,
) -> (TpShardKind, usize, usize)(kind, full_out, full_in) for one checkpoint key.