pub fn verify_topology_draft(
draft: &TopologyDraftBatch,
target_predictions: &[u32],
) -> SpeculativeVerifyResultExpand description
Verify a speculative TopologyDraftBatch against target model greedy predictions
or logit argmax tokens.
For each draft token $i \in [0, \text{draft_len})$, target model predicts $t^_i$. If $t_i == t^i$, draft token is accepted. Upon first divergence at index $j$, acceptance stops; $j$ tokens are accepted, and $t^_j$ is emitted as the correction token. If all draft tokens are accepted, all $\text{draft_len}$ tokens are accepted, and bonus token $t^{\text{draft_len}}$ is emitted.