llama-cpp: Bump to version b6565 (#28433)

[akos-fulop-public]

Summary

Changes to recipe: llama-cpp/b6565

Motivation

Request: #28433

Details

Patch no longer seems necessary, as of ggml-org/llama.cpp#14613

---

• Read the contributing guidelines
• Checked that this PR is not a duplicate: list of PRs by recipe
• Tested locally with at least one configuration using a recent version of Conan

[franramirez688]

Hi @akos-fulop-public, thanks for contributing!

You need to add the new version b6529 to the config.yml too:

--- a/recipes/llama-cpp/config.yml
+++ b/recipes/llama-cpp/config.yml
@@ -1,4 +1,6 @@
 versions:
+  "b6529":
+    folder: "all"
   "b4570":
     folder: "all"
   "b4079":

[akos-fulop-public]

Hi @franramirez688 ,
Thanks, sorry I missed that!

[akos-fulop-public]

Previous build failed on macOS.
As far as I could tell, this property (https://developer.apple.com/documentation/metal/mtlbuffer/gpuaddress) is only added in macOS 13, and the CI runs on lower version.
I looked at the prebuild_tool_package template, but I'm uncertain if the check I added here is a good solution.

[uilianries]

The Windows build failed with (https://github.com/conan-io/conan-center-index/pull/28435/checks?check_run_id=51103353004):

C:\workspace\cci_prod_PR-28435\conan-home\p\b\llama49bf62bb2ee1b\b\src\common\common.cpp(432): error C2039: 'system_clock': is not a member of 'std::chrono'

I opened a PR to the upstream with a quick fix: ggml-org/llama.cpp#16211

[akos-fulop-public]

Thank you @uilianries ,
I updated to the version that includes the fix.

[uilianries]

That new release fixed that error in common.cpp, but the old version is failing due missing chrono header in log.cpp. It's fixed already in newer versions.

I would suggest reverting conanfile.py to keep it as in Conan Center, so the CI will build only the version b6565 and should not fail with that error.

[akos-fulop-public]

Thanks for the help @uilianries !
Am I supposed to clean up my commits / is there anything else I need to do now?

[uilianries]

LGMT Thank you @akos-fulop-public !! 😄

I see several changes since the last version, including new options.

In special, using git commands with ggml, but it looks safe by comparing with the Linux build here: https://c3i.jfrog.io/artifactory/cci-build-logs/cci/prod/PR-28435/7/package_build_logs/build_log_llama-cpp_b6565_87db59a20d2209f720e7aefbd727fd4d_0f2ae57cd830dd618c06cce8e80c9fc879ab3710.success.txt

For more reference:

• ggml-org/llama.cpp@b4570...b6565#diff-1e7de1ae2d059d21e1dd75d5812d5a34b0222cef273b7c3a2af62eb747f9d20a
• ggml-org/llama.cpp@b4570...b6565#diff-a77d393383a84e0dd42ebd261b80443390335377813ca160e32abf8c52effac2
• ggml-org/llama.cpp@b4570...b6565#diff-a77d393383a84e0dd42ebd261b80443390335377813ca160e32abf8c52effac2
• ggml-org/llama.cpp@b4570...b6565#diff-d4bb67eec77d05ad8d4056c50492de104bd17dc6e0b0f9919854aeeaa5002842