llama.cpp
server : speculative checkpointing
#19493
Merged

server : speculative checkpointing #19493

srogmann
srogmann srogmann requested a review from ngxson ngxson 159 days ago
srogmann srogmann requested a review from ggerganov ggerganov 159 days ago
github-actions github-actions added examples
github-actions github-actions added server
ggerganov
ggerganov commented on 2026-02-11
srogmann
srogmann srogmann force pushed from c5911892 to 0fa66c27 153 days ago
srogmann srogmann force pushed from 0fa66c27 to 665893a0 145 days ago
srogmann
srogmann srogmann force pushed from 58d612a2 to edc9b889 139 days ago
srogmann
stsydow
srogmann
srogmann
srogmann
ggerganov
ggerganov commented on 2026-03-09
srogmann srogmann force pushed from 278bb0a9 to e0c2f928 132 days ago
stsydow
srogmann srogmann force pushed from e0c2f928 to f39bb5db 131 days ago
srogmann
JonathanJing
srogmann
stsydow
l0nedigit
srogmann
ggerganov
srogmann srogmann force pushed from 1f629668 to e0a61a5f 127 days ago
srogmann
srogmann
ggerganov
srogmann
ggerganov
ggerganov
petter-b
srogmann
srogmann
ggerganov
srogmann server : speculative decoding using checkpoints
a4237ea0
srogmann server : fix draft check with checkpoints
3723f8e5
srogmann server : rename spec vars
e002b095
srogmann server : log levels
01763e80
srogmann server : refactored spec logic to speculative.cpp
e994c4ec
srogmann server : renamed spec checkpoints option
bd2f7f2d
srogmann server : fix spec checkpoints, logging
af3b630e
srogmann speculative : checkpoints with draft model, logging
fe4f859a
srogmann server : n_tokens_cur and create_checkpoint in draft
91932ae0
srogmann server : fix server_speculative_callback (slot.id)
b5b3ac3b
srogmann spec : fix ngram-map/begin idx_last_check
7b7ce932
srogmann spec : init ckpt (begin() wasn't called)
21921a1e
srogmann chore: update webui build output
7d2814a9
srogmann srogmann force pushed from e0a61a5f to 7d2814a9 121 days ago
srogmann srogmann requested a review 121 days ago
srogmann srogmann requested a review 121 days ago
srogmann
ggerganov
srogmann
petter-b
srogmann
srogmann
srogmann server : restore sampler in spec checkpoint and clear mem
d0a85689
ggerganov
srogmann
ggerganov
srogmann
ggerganov Merge branch 'master' into pr/19493
63c66f15
ggerganov
ggerganov commented on 2026-04-01
ggerganov cont : avoid --spec-use-checkpoints argument
8491e154
ggerganov
am17an
ggerganov
am17an
ggerganov
ggerganov cont : remove server_prompt_checkpoint_with_size
e1141d1c
ggerganov Merge branch 'master' into pr/19493
00c0c4c1
ggerganov
srogmann spec : rename (leave_draft_state)
66f0ce2c
treo
ggerganov Merge branch 'master' into pr/19493
cfb6c4c7
ggerganov cont : clean-up
ca9e370b
ggerganov cont : do not ignore partial drafts even if the are short
5b9b0f02
ggerganov
ggerganov cont : spec callback owned by session
c1adaf99
ggerganov cont : simplify
dc59f5b1
ggerganov cont : avoid empty speculative session
50a15510
ggerganov Merge branch 'master' into pr/19493
cec07350
ggerganov cont : simplify
ee248c68
ggerganov cont : simplify
e385affc
ggerganov cont : enable mtmd speculative decoding
0d4c9118
ggerganov
ggerganov cont : keep the spec sampler alive
257bb7cc
ggerganov cont : simplify
2132f818
ggerganov cont : fix nullptr deref + draft checkpoints
89cf768c
ggerganov cont : remove common_speculative_accept_response
edf2af7c
ggerganov cont : remove callback
73f6658c
ggerganov cont : simplify
2a0a0c74
ggerganov cont : minor
0c6b7a0d
ggerganov cont : simplify
66e0c9a2
ggerganov Merge branch 'master' into pr/19493
ada21d30
ggerganov cont : fix accepted number
e1f8a1f0
ggerganov
ggerganov approved these changes on 2026-04-18
ggerganov ggerganov added merge ready
drrros
drrros
srogmann
ggerganov ggerganov merged 455d8e4b into master 91 days ago
thc1006
treo
ruixiang63
ggerganov

Login to write a write a comment.

Login via GitHub

Reviewers
Assignees
No one assigned
Labels
Milestone