Upgrade llama.cpp from b10197 to b10217 - #373
Merged
Merged
Conversation
7 commits: server support for inp embd to generate next token (server-context.cpp, additive), SYCL q2 mul_mat + RMS_NORM/MUL fusion + dev2dev memcpy, ggml-webgpu flash-attn quantized-KV tuning, Vulkan quantized-concat support, and an mtmd Lanczos image-resize method. All 7 patches apply cleanly (no overlap with the server-context.cpp hunks they touch). Full local build + ctest (485/485) + NativeLibraryLoadSmokeTest verified.
3 commits: llama_init_from_model gains a K/V cache type-matching check for DeepSeek-V4/MLA models and folds the quantized-V-cache validation into a combined block that auto-enables flash_attn (internal validation logic, no signature change), plus SYCL cpy type coverage and a ggml-zendnn mul_mat_id refactor. All 7 patches apply cleanly. Full local build + ctest (485/485) + NativeLibraryLoadSmokeTest verified.
9 commits: include/llama.h gains a purely-additive load_mtp field; server-context.cpp adds speculative-replay accepted-token accounting (server_slot::spec_is_replay, corrects draft-acceptance counters on partial-acceptance replay) -- clear of the 0002/0003 patch hunks. Plus CUDA MMQ tile-loading refactors, SYCL oneMKL XMX flash attention, Vulkan POOL_1D op, and mtmd n_embd_head. All 7 patches apply cleanly. Full local build + ctest (485/485) + NativeLibraryLoadSmokeTest verified.
1 commit: DeepSeek-V4 tool-call-in-thinking chat-format parsing rework (common/chat.cpp internals, no common/chat.h signature change). No patches or project source needed changes. This is the final chunk of the b10197 -> b10217 bump (via b10204, b10207, b10216). Full local build + ctest (485/485) + NativeLibraryLoadSmokeTest + javadoc verified.
Upstream (ngxson) closed ggml-org/llama.cpp#22393 without merging, reasoning that slot_prompt_similarity is internal single-use-case state that shouldn't be public API. Update the patch 0003 header and its CLAUDE.md table row so both stop framing the patch as "pending merge" -- it is now carried permanently and will never be droppable via a version bump. No functional change; verified the patch still applies cleanly against the current b10217 pin.
Keep each patch's header self-contained instead of naming sibling patch numbers, which drift as patches are added/dropped over time.
bernardladenthin
had a problem deploying
to
startgate
August 1, 2026 09:49 — with 
GitHub Actions
Error
GitHub Actions
Error
bernardladenthin
had a problem deploying
to
maven-central
August 1, 2026 09:49 — with
GitHub Actions
Failure
GitHub Actions
Failure
bernardladenthin
had a problem deploying
to
maven-central
August 1, 2026 09:49 — with
GitHub Actions
Failure
GitHub Actions
Failure
|
Review Summary: llama.cpp b10197 to b10217 Upgrade This merged PR demonstrates excellent quality. All version strings consistently updated across CMakeLists.txt, LlamaCppVersion.java, README.md, and CLAUDE.md. All 7 patches re-verified and applied cleanly through chunked verification steps (b10197 to b10204 to b10207 to b10216 to b10217). Priority-8 headers unchanged. Patch 0003 correctly documented as permanently carried after upstream rejection. Test coverage complete: 485/485 C++ tests passing, NativeLibraryLoadSmokeTest green. No breaking API changes to JNI surface. OuteTTS generator anchors verified. No issues found - all changes correct and complete. |
|
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
Changes
Version pin updates:
llama/CMakeLists.txt: GIT_TAG and LLAMA_TAG updated to b10217llama/src/main/java/net/ladenthin/llama/value/LlamaCppVersion.java: LLAMA_CPP_VERSION constant updatedREADME.md: Badge link updated to b10217CLAUDE.md: Documentation updated with new version and detailed upstream verification notesPatch verification:
All 7 patches re-verified and applied cleanly against b10217:
0001-win32-arg-parse-embed-guard.patch— No changes to patch-target files in this range0002-server-preserve-caller-load-progress-callback.patch— No changes to patch-target files in this range0003-pr22393-server-add-slot-prompt-similarity-getter-setter.patch— Updated header comment to document that upstream PR #22393 was closed without merging; patch is now permanent0006-server-embed-native-server-jni.patch— No changes to patch-target files in this range0007-server-attach-http-frontend.patch— No changes to patch-target files in this range0008-server-models-worker-cmd-override.patch— No changes to patch-target files in this rangeUpstream changes in b10197→b10217 range:
server_batch::add()overload andserver_slot::inp_embdmember for TTS-style models)No project-source changes required; all upstream modifications are internal or additive with no impact on public APIs or JNI signatures.
Test plan
Per-platform build + ctest confirmation deferred to the CI pipeline.
Related issues / PRs
Upstream: ggml-org/llama.cpp#22393 (rejected; now carried permanently)
Checklist
CONTRIBUTING.mdandCODE_OF_CONDUCT.mdhttps://claude.ai/code/session_01CV2A6f4ezHubpZgsqTDkj4