pub fn generate_speculative(
model: &dyn Model,
prompt_tokens: &[u32],
params: &SamplingParams,
num_drafts: usize,
) -> Result<GenerateResult>Expand description
Generate with speculative decoding (MTP).
Delegates the speculative decode loop to model.generate_speculative(),
which has access to GPU/buffers needed for the MTP proposer.