Skip to content

chore: upgrade llama.cpp from b11222 to b11236 - #464

Merged
bernardladenthin merged 1 commit into
mainfrom
claude/busy-archimedes-lo9uyy
Sep 28, 2026
Merged

bernardladenthin merged 1 commit into
mainfrom
claude/busy-archimedes-lo9uyy

Conversation

@bernardladenthin

Copy link
Copy Markdown
Owner

Summary

  • Bumps the pinned llama.cpp from b11222 to b11236 (llama/CMakeLists.txt, README.md, CLAUDE.md, LlamaCppVersion.java).
  • 0001 refreshed: upstream #29426 rewrote tests/test-recurrent-state-rollback.cpp so its main() filters --models DIR into its own filtered_argv before calling common_params_parse. By the patch's own rule, that call site keeps common_params_parse, so the hunk is dropped. This is the same case as test-save-load-state.cpp at b10679. The other eight patches apply unchanged.
  • No project source change. The batch_ext migration (#29385) changes or removes common_batch, common_speculative_process and mtmd_helper_post_decode_callback symbols that this project does not use. gen_audio, tools/server/*.h, server-schema.cpp and server-task.cpp are unchanged, and there is no upstream .github change, so the ROCm and Windows-CUDA pins stay.

Test plan

  • All nine patches apply in order to a clean b11236 tree (git apply)
  • Drop checks: 0002 is still needed (the callback is still assigned unconditionally); the files patched by 0001, 0012, 0014 and 0015 are untouched in the range
  • CI is green on this branch (build/test is left to CI)
  • Docs updated (CLAUDE.md patch note, docs/history/llama-cpp-breaking-changes.md)

Related issues / PRs

Follows #460 (b11214 → b11222).

Checklist

  • My commits follow Conventional Commits
  • No security-sensitive changes

🤖 Generated with Claude Code

https://claude.ai/code/session_01QSXAwbXqx9u5xsMArMvgT8


Generated by Claude Code

- 0001 refreshed: upstream #29426 rewrote tests/test-recurrent-state-rollback.cpp
  so its main() filters --models into its own argv before common_params_parse;
  by the patch's own rule that call site keeps common_params_parse, so the hunk
  is dropped (as test-save-load-state.cpp at b10679). All nine patches apply in
  order to a clean b11236 tree.
- No project source change: the batch_ext migration (#29385) touches
  common_batch / speculative / mtmd post-decode symbols this project does not use;
  gen_audio, the server headers, server-schema.cpp and server-task.cpp are
  unchanged; no upstream .github change (ROCm / Windows-CUDA pins stay).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QSXAwbXqx9u5xsMArMvgT8

Copy link
Copy Markdown
Owner Author

The two Verify GPG signing key checks ("… (no secrets printed)" and "… — Gradle/BouncyCastle path …") fail here, but not because of this PR. They read the signing key from the maven-central environment, and those secrets are not available on PR branches. They fail the same way on every PR (as they did on #461 and #460). This diff changes only the llama.cpp pin, patch 0001 and docs, and touches no signing or workflow code. There is no fix to port; they go green on main / release dispatches.


Generated by Claude Code

@bernardladenthin
bernardladenthin merged commit 56891c7 into main Sep 28, 2026
9 of 14 checks passed
@bernardladenthin
bernardladenthin deleted the claude/busy-archimedes-lo9uyy branch September 28, 2026 20:21
@sonarqubecloud

Copy link
Copy Markdown

This branch had an error being deployed

1 failed deployment
maven-central — 16e24a17 Deployed Sep 28, 2026 by bernardladenthin via Verify GPG signing key (no secrets printed) #1059
startgate — 16e24a17 Deployed Sep 28, 2026 by bernardladenthin via Start gate (abort window) #1059
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