pub(crate) fn qwen4_exp_gen_once(
gpu: &mut Gpu<'_>,
cpu: Option<&CpuModel<'_>>,
cpu_state: Option<&mut State>,
tok: &ChatTokenizer,
ids: &[u32],
max_tokens: usize,
options: &Options,
resume: Option<PrefillResume>,
emit: &mut dyn FnMut(u32) -> Result<()>,
) -> Result<()>