pub fn select_candidate_blocks(
logits: &[f32],
queries: usize,
width: usize,
compress_lens: &[usize],
topk_blocks: usize,
block: usize,
) -> Vec<bool>Expand description
2026-09-25: Candidate blocks: per query, score each block-wide block of
logits[queries][width] (unreachable positions at -inf) by its max, pin
the block holding position compress_lens[q] - 1, keep the top
topk_blocks blocks that are not all -inf. Returns the keep mask
[queries][width].