Skip to content

Upgrade llama.cpp from b9990 to b10015#342

Merged
bernardladenthin merged 2 commits into
mainfrom
claude/java-llama-cpp-b9999-y5szmx
Jul 15, 2026
Merged

Upgrade llama.cpp from b9990 to b10015#342
bernardladenthin merged 2 commits into
mainfrom
claude/java-llama-cpp-b9999-y5szmx

Conversation

@bernardladenthin

Copy link
Copy Markdown
Owner

Summary

  • Upgrade pinned llama.cpp version from b9990 to b10015 across all configuration files and version constants
  • Refactor 0007-server-attach-http-frontend.patch to consolidate experimental-feature warnings into a single location in llama_server(), removing duplicate warning blocks from the shared llama_server_register_common_routes() helper
  • Update documentation and version badges to reflect the new llama.cpp release

Changes

Version Bump

  • Updated GIT_TAG in llama/CMakeLists.txt from b9990 to b10015
  • Updated LLAMA_CPP_VERSION constant in llama/src/main/java/net/ladenthin/llama/value/LlamaCppVersion.java
  • Updated version references in CLAUDE.md, README.md, and build documentation

Patch Refactoring (0007-server-attach-http-frontend.patch)

The patch was refined to better separate concerns:

  • Shared route registration (llama_server_register_common_routes()): Registers HTTP routes for CORS proxy, built-in tools, and health/slots endpoints without logging experimental-feature warnings
  • Entry-point warnings (in llama_server()): Consolidated all experimental-feature warnings (CORS, router mode, MCP proxy, built-in tools) into a single unified warning block that depends on router state, which the shared helper does not carry
  • Removed duplicate warning blocks (8 lines) that were previously in both the helper and the main function

This refactoring ensures warnings are logged only once per server startup and only in the context where router state is available, while allowing the route registration logic to be safely shared across standalone, router, and attach modes.

Test plan

  • CI is green on this branch
  • Existing patches apply cleanly to b10015 (verified via fail-loud PATCH_COMMAND)
  • No new test coverage needed — this is a version bump with internal refactoring of existing patch logic

Related issues / PRs

Routine llama.cpp version bump as part of ongoing upstream tracking.

Checklist

  • I have read CONTRIBUTING.md and CODE_OF_CONDUCT.md
  • My commits follow Conventional Commits
  • No security-sensitive changes

https://claude.ai/code/session_01VEwuFTTgA1jrqMB2m24yqe

…25655

Bump the pin in all four required places (GIT_TAG + LLAMA_TAG in
llama/CMakeLists.txt, README badge, CLAUDE.md pinned-version line + doc
references, and the LlamaCppVersion.LLAMA_CPP_VERSION constant).

Patch refresh: upstream #25655 ("server: add --cors-* options") rewrote the
CORS-proxy/tools warning region of tools/server/server.cpp that
0007-server-attach-http-frontend.patch relocates (introducing a warn_names
vector, an is_router_server branch and a new cors_origins=='*' security
warning), so 0007 no longer applied. Refreshed so
llama_server_register_common_routes() extracts only the route bindings, while
the per-caller experimental-feature warnings (incl. upstream's new cors_origins
warning) stay in llama_server() -- they depend on router state the shared helper
does not carry, preserving #25655 with no upstream-warning regression. Patches
0001-0003, 0006, 0008 apply unchanged; #25649's prompt-cache-ownership refactor
and #25504's arg.cpp log-flush touch regions outside every patch.

Append the b9990..b10015 rows to the breaking-changes history.

Verified locally: fresh cmake configure (all six patches apply fail-loud, TTS
extracted from tts.cpp @ b10015) + full cmake --build (jllama + jllama_test link
cleanly, confirming the refreshed 0007 server.cpp restructure compiles) + ctest
485/485 passing.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01VEwuFTTgA1jrqMB2m24yqe
The b9986->b9990 bump (merged via PR #341) omitted its history rows, leaving a
gap between b9984->b9986 and b9990->b10015. Backfill the analysis + verification
rows from the upstream b9986...b9990 compare diff: additive-only (GGUF
tensor-shape accessor #24405 + Minimax2 eagle3 model support + test
harmonization), no patch-target or priority-8 file touched, all six patches
apply unchanged.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01VEwuFTTgA1jrqMB2m24yqe
@bernardladenthin
bernardladenthin merged commit 9774906 into main Jul 15, 2026
9 of 14 checks passed
@bernardladenthin
bernardladenthin deleted the claude/java-llama-cpp-b9999-y5szmx branch July 15, 2026 07:05
@sonarqubecloud

Copy link
Copy Markdown

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants