PR #25173 spec: add DSpark speculative decoding

spec: add DSpark speculative decoding #25173

wjinxu wants to merge 11 commits into ggml-org:master from wjinxu:dspark-upstream

github-actions added model

github-actions added conversion

wjinxu force pushed from f3b83cdf to d74ff774 10 days ago

github-actions added testing

wjinxu force pushed from d74ff774 to 37f2513d 10 days ago

wjinxu marked this pull request as ready for review 10 days ago

wjinxu requested a review from

JohannesGaessler 10 days ago

wjinxu requested a review from

CISC 10 days ago

wjinxu requested a review from

ggerganov 10 days ago

wjinxu requested a review 10 days ago

CISC requested a review from

ruixiang63 10 days ago

CISC commented on 2026-06-30

wjinxu force pushed from 37f2513d to aae2941a 10 days ago

wjinxu force pushed from aae2941a to d8b38f23 10 days ago

lym000000 commented on 2026-07-01

wjinxu force pushed from d8b38f23 to 47f34420 9 days ago

wjinxu marked this pull request as draft 8 days ago

wjinxu force pushed from 47f34420 to 96c5be9d 8 days ago

wjinxu marked this pull request as ready for review 8 days ago

am17an commented on 2026-07-04

github-actions added documentation

am17an approved these changes on 2026-07-08

ggerganov assigned

ggerganov 2 days ago

ggerganov unassigned

ggerganov 2 days ago

ruixiang63 commented on 2026-07-08

wjinxu requested a review 1 day ago

wjinxu requested a review from

ngxson 1 day ago

spec: add DSpark speculative decoding

8006ae30

spec: read draft block size in the dflash impl

3e048b82

docs: add DSpark section to speculative.md

5bfd3219

spec: keep dspark block size read in the dspark impl

f2bd1906

dspark : add TODOs for incomplete parts

576908bd

spec: fold DSpark into the DFlash arch

c52ba2b2

spec: add confidence-based draft pruning for DSpark

42865369

fold the dspark impl into dflash, selected by spec type

80d1d08f

github-actions added server

wjinxu force pushed from a63c622b to 80d1d08f 1 day ago

ruixiang63 commented on 2026-07-09

address review comments

dc900904

dspark: clean up and improve naming

66f37202

ruixiang63 commented on 2026-07-10

remove trailing whitespace

27cc3bae

ggerganov assigned

ggerganov 8 hours ago

Reviewers

am17an

ruixiang63

CISC

lym000000

JohannesGaessler

ggerganov

ngxson

Assignees

ggerganov

Labels

documentation model testing server conversion

Milestone

No milestone

llama.cpp spec: add DSpark speculative decoding #25173 Open

spec: add DSpark speculative decoding #25173

llama.cpp
spec: add DSpark speculative decoding
#25173

Open