[mtp] add rejection sampling for speculative decoding (#26312)

Co-authored-by: lyc508653 <lyc508653@alibaba-inc.com>
Co-authored-by: Qiaolin-Yu <liin1211@outlook.com>
Co-authored-by: Huiqiang Jiang <30883354+iofu728@users.noreply.github.com>
Co-authored-by: Yi Zhang <25844240+yizhang2077@users.noreply.github.com>
Co-authored-by: Yizhong Cao <114661107+cao1zhg@users.noreply.github.com>
This commit is contained in:
Yucheng Li
2026-06-20 15:10:42 -07:00
committed by GitHub
co-authored by lyc508653 Qiaolin-Yu Huiqiang Jiang Yi Zhang Yizhong Cao
parent 95fb1ef697
commit f42ec350b4
11 changed files with 566 additions and 20 deletions
@@ -1336,6 +1336,10 @@ class ForwardBatch(ForwardBatchDeepSeekMHAMixin):
spec_info.topk_index = self._pad_tensor_to_size(
spec_info.topk_index, bs
)
if getattr(spec_info, "draft_probs", None) is not None:
spec_info.draft_probs = self._pad_tensor_to_size(
spec_info.draft_probs, bs
)
if getattr(spec_info, "num_correct_drafts", None) is not None:
spec_info.num_correct_drafts = self._pad_tensor_to_size(
spec_info.num_correct_drafts, bs