pub fn sample_token(logits: &[f32], temperature: f32, top_p: f32) -> usizeExpand description
Sample next token from logits using temperature + top-p (nucleus) sampling.
§Arguments
logits- Raw logits [vocab_size]temperature- Temperature for scaling (1.0 = normal, <1 = sharper, >1 = more random)top_p- Nucleus sampling threshold (0.9 = consider tokens covering 90% of probability mass)
§Returns
Token ID (index into logits array)