llama.cpp
spec : Warn when --spec-draft-p-min is used with probabilistic draft sampling
#31294
Merged

spec : Warn when --spec-draft-p-min is used with probabilistic draft sampling #31294

praneshgo
spec : skip p_min for sampled MTP and DFlash2 drafts
091b9fd1
github-actions github-actions added documentation
github-actions github-actions added server
praneshgo
praneshgo praneshgo marked this pull request as ready for review 18 days ago
praneshgo praneshgo requested a review 18 days ago
praneshgo praneshgo requested a review from ngxson ngxson 18 days ago
praneshgo praneshgo requested a review 18 days ago
praneshgo
praneshgo praneshgo changed the title Skip minimum speculative decoding probability for sampled MTP and DFlash2 drafts Skip minimum speculative decoding probability for MTP and DFlash2 drafts with probabilistic draft sampling 18 days ago
ruixiang63 ruixiang63 assigned ruixiang63 ruixiang63 18 days ago
gaugarg-nv
gaugarg-nv gaugarg-nv requested a review from gaugarg-nv gaugarg-nv 15 days ago
praneshgo
spec : replace the sampled-draft p_min skip with a warning
8c54744f
praneshgo praneshgo changed the title Skip minimum speculative decoding probability for MTP and DFlash2 drafts with probabilistic draft sampling Warn when --spec-draft-p-min is used with probabilistic draft sampling 13 days ago
praneshgo
sanmai sanmai changed the title Warn when --spec-draft-p-min is used with probabilistic draft sampling spec : Warn when --spec-draft-p-min is used with probabilistic draft sampling 13 days ago
ORippler
ORippler dismissed these changes on 2026-11-11
ORippler ORippler dismissed their stale review 4 days ago
approved by mistake, wanted to post a comment
praneshgo
praneshgo praneshgo requested a review from ORippler ORippler 4 days ago
ruixiang63
ruixiang63 approved these changes on 2026-11-12
spec : warn about p_min only for drafters that use it
489270e6
ORippler
ORippler approved these changes on 2026-11-13
ORippler
ORippler commented on 2026-11-13
ORippler ORippler merged 7874988b into master 2 days ago

Login to write a write a comment.

Login via GitHub

Assignees
Labels
Milestone