Bump llama.cpp b11069 → b11080, drop patch 0010 - #451
Merged
Merged
Conversation
Eleven upstream commits, 1244 KiB, no project-source change. The size is
almost entirely #29197, which rewrites 46 files under ggml/src/ggml-hexagon/
for a backend this project does not build; the rest of ggml/src is additive
ARM repack kernels (#23492), a Metal fusion-list simplification (#29206) and
a SYCL softmax tweak (#28918). ggml/include is byte-identical.
The consequential commit is #28518 ("json: Fixed json enum handling"):
common_json_value gains an std::is_enum-gated constructor delegating to the
underlying type, and common_json_is_value now accepts enums. That fixes at
its root the trap patches/0010 cast around at upstream's get_res_model_info()
emit site, so the patch became a redundant carry and is dropped rather than
refreshed (the 0009/0011/0013 precedent).
Worth recording: 0010 still applied cleanly at b11080 -- upstream never
touched the emit site -- so the fail-loud applier said nothing. This is
exactly the case the by-hand drop-check exists for, and the reason it is
worded around the defect rather than around the patch text.
The runnable guard is kept and re-pointed (the 0011 precedent): the
CommonJsonEnumTrap pair in test_json_helpers.cpp is now the CommonJsonEnum
trio and pins upstream's contract -- an uncast enum serialises as its numeric
value, an explicit static_cast<int> is equivalent, and a real bool is still a
boolean. jllama.cpp keeps its own two "vocab_type" casts, correct either way.
verify-patches-applied.sh loses its third check (0010 was the only patch with
no runnable guard) and keeps its two generic assertions; the matching TODO.md
coverage-gap entry is resolved and removed.
Also in range, neither requiring a change: six existing sampling flags gained
environment defaults (#27380), and a router no longer forwards
LLAMA_ARG_API_KEY_FILE to spawned children (#28938). tools/server/'s schema,
task and context translation units are byte-identical, and the request-field
set (68), bounded-field set (23) and response-key set (142) were verified
unchanged mechanically, so the server wire contract cannot have moved.
The other eight patches apply unchanged and all four remaining drop-checks
still report "still required" at the pristine tag.
Verified from a fresh configure through the real FetchContent path: stamp
head 1d72b05d3 with eight SHA-256 lines, verify-patches-applied.sh green,
wire-name extraction unchanged at 138 CLI / 57 request / 15 trainer names,
Release build clean, ctest 559/559 (558 -> 559, the re-pointed guard gained a
case), nm -D 40 Java_* exports, NativeLibraryLoadSmokeTest 4/4 with 0 skipped
after a mvn clean, full mvn test 1772 run / 0 failures / 0 errors (272
model-gated skips, no GGUF in this sandbox), clang-format 23.1.1 clean,
spotless clean, SpotBugs 0 findings.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012eNXhF4RyXvmZEVZo16G4S
bernardladenthin
had a problem deploying
to
maven-central
September 21, 2026 20:28 — with 
GitHub Actions
Failure
GitHub Actions
Failure
bernardladenthin
had a problem deploying
to
maven-central
September 21, 2026 20:28 — with
GitHub Actions
Failure
GitHub Actions
Failure
bernardladenthin
had a problem deploying
to
startgate
September 21, 2026 20:28 — with
GitHub Actions
Error
GitHub Actions
Error
This branch had an error being deployed
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.
Summary
std::is_enum-gated constructor tocommon_json_value, fixing the enum-to-boolean serialization trap at its root.0010(castvocab_typetointforcommon_json). The patch became redundant once upstream fixed the underlying defect. It still applied cleanly, which is why the by-hand drop-check exists — the fail-loud applier cannot detect "upstream already fixed this".CommonJsonEnumTraptest guard toCommonJsonEnum(three tests instead of two) to pin upstream's contract: uncast enum stays numeric, explicit cast is equivalent, real booleans remain booleans. A future bump that loses the enum constructor will redC++ Testseverywhere instead of shipping a silent regression.The server wire contract is unchanged: schema, task and context translation units are byte-identical; request-field set (68), bounded-field set (23), and response-key set (142) verified mechanically. All eight remaining patches apply unchanged; all four drop-checks still say "still required".
Test plan
CommonJsonEnumtrio intest_json_helpers.cpp)Related issues / PRs
Upstream: ggml-org/llama.cpp#28518 ("json: Fixed json enum handling")
Checklist
CONTRIBUTING.mdandCODE_OF_CONDUCT.mdhttps://claude.ai/code/session_012eNXhF4RyXvmZEVZo16G4S