Repository navigation
grammar : fix max repetition threshold - #28469
Merged
Merged
Conversation
ggerganov
approved these changes
Sep 6, 2026
agentdavo
pushed a commit
to agentdavo/llama.cpp
that referenced
this pull request
Sep 6, 2026
(cherry picked from commit 9e0e220)
x1250
pushed a commit
to x1250/llama.cpp
that referenced
this pull request
Sep 9, 2026
zbrad
pushed a commit
to zbrad/llama.cpp
that referenced
this pull request
Sep 10, 2026
pl752
pushed a commit
to pl752/llama.cpp
that referenced
this pull request
Sep 15, 2026
zsogitbe
pushed a commit
to zsogitbe/llama.cpp
that referenced
this pull request
Sep 17, 2026
Te-eMster
pushed a commit
to Te-eMster/mx-llama.cpp
that referenced
this pull request
Sep 18, 2026
frostyautumnleaf
pushed a commit
to frostyautumnleaf/llama.cpp
that referenced
this pull request
Oct 5, 2026
roverdrubber
added a commit
to roverdrubber/llama.cpp
that referenced
this pull request
Oct 15, 2026
The branch is far behind upstream master and the parts that a card other than this one needs had drifted. This ports the non-backend half of that sync, content-matched against upstream rather than replayed by hash: the parser and JSON schema chain (ggml-org#27764, ggml-org#28736, ggml-org#28742, which adds common/parsers/, common/json-schema.* and tests/test-json-schema.cpp and retires the three example scripts it replaces), the chat, grammar, jinja and server fixes (ggml-org#28302, ggml-org#28469, ggml-org#26885, ggml-org#28279, ggml-org#28511, ggml-org#28620, ggml-org#28817, ggml-org#28749), the qwen4exp RMSNorm and multiply fusion (ggml-org#28896, hand-ported because the norm tensors here load with a different shape), the GDN q/k l2 norm as flash-linear-attention defines it (ggml-org#28068), the ggml_prec_set_acc API that replaces ggml_mul_mat_set_prec and ggml_flash_attn_ext_set_prec at every call site, and the scheduler fixes for zero-sized expert ids (ggml-org#28739) and split inputs (ggml-org#28387). The GPU backend half of the same sync lands in the following two commits; this one alone does not build the CUDA or HIP backend, because their call sites move to the new precision API there. test-chat, test-json-schema, test-json-schema-to-grammar, test-grammar-integration and test-grammar-parser pass. Assisted-by: Claude Opus 5
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
Fix #25746
Fix #27859
Accept a max length of 2000, for some reason clients really like this number. Repetition
> 2000is converted to unbounded, but when exactly 2000 it was still hitting this error.Requirements