Function qwen38_tensor_parallel 

pub fn qwen38_tensor_parallel(cfg: &Qwen38Cfg, facts: &RduChipFacts) -> usize

The tensor-parallel width Qwen3.8 is served at on facts’ chips: the widest the model admits (qwen38_admits_tp) within the chips’ tensor-parallel domain (default_tensor_parallel); the rest of the domain serves more sequences as data-parallel replicas.