Repository navigation
server: correct accepted tokens when need draft token replay - #26320
Merged
Merged
Conversation
Member
Author
|
@ggerganov Can you please take a look? |
ruixiang63
changed the title
ggerganov
approved these changes
Jul 31, 2026
ggerganov
added
the
merge ready
huaxel
pushed a commit
to huaxel/CachyLLama
that referenced
this pull request
Aug 2, 2026
…g#26320) * spec: correct accepted tokens when need draft token replay * cont : naming --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
rcmorano
pushed a commit
to rcmorano/ROCmFPX
that referenced
this pull request
Aug 6, 2026
Port of ggml-org/llama.cpp#26320 (upstream commit 0005475). When the context does not support partial draft acceptance, the draft is truncated and the state restored, and the replayed token was then counted as an accepted draft token. That inflated both n_draft_accepted and n_accepted_per_pos by one for every replay, so the acceptance rate and mean acceptance length reported by the server (and used to tune --spec-draft-n-max and --spec-draft-p-min) read higher than reality. Adds a spec_is_replay flag on the slot: cleared in reset(), set on the truncate-and-restore path, and consumed when tallying acceptance. This only corrects reported speculative metrics. Generated output and decode throughput are unaffected. This tree shares no git ancestry with upstream (it begins at a published source snapshot), so cherry-pick could not apply the commit directly; the change was applied by hand at the four equivalent sites. Co-authored-by: Ruixiang Wang <wangruixiang07@outlook.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
satindergrewal
pushed a commit
to satindergrewal/llama.cpp
that referenced
this pull request
Aug 12, 2026
…g#26320) * spec: correct accepted tokens when need draft token replay * cont : naming --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
thecodacus
pushed a commit
to thecodacus/llama.cpp
that referenced
this pull request
Sep 7, 2026
…g#26320) * spec: correct accepted tokens when need draft token replay * cont : naming --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
zbrad
pushed a commit
to zbrad/llama.cpp
that referenced
this pull request
Sep 10, 2026
…g#26320) * spec: correct accepted tokens when need draft token replay * cont : naming --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
pl752
pushed a commit
to pl752/llama.cpp
that referenced
this pull request
Sep 15, 2026
…g#26320) * spec: correct accepted tokens when need draft token replay * cont : naming --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
frostyautumnleaf
pushed a commit
to frostyautumnleaf/llama.cpp
that referenced
this pull request
Oct 5, 2026
…g#26320) * spec: correct accepted tokens when need draft token replay * cont : naming --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Overview
When the target context can't roll back a partially-accepted speculative draft in place — i.e. it falls back to a full checkpoint restore (
COMMON_CONTEXT_SEQ_RM_TYPE_FULL, orRSwithn_rollback > n_rs_seq) — the server restores the checkpoint and replays the accepted prefix plus the target's own correction token as the next draft.On that replayed step every token is accepted, so the per-request speculative stats counted the target-generated correction token as an accepted draft token, inflating draft acceptance, per-position acceptance and mean len by 1 for each checkpoint restore.
This marks the slot when a checkpoint replay is queued (
is_draft_replay) and excludes that single correction token from the acceptance statistics on the replayed step.This change only affects reported per-request stats only.
Additional information
Requirements