Update llama.cpp to b6482 (3d4053f)

[admiralakber]

Closes #830

• Update submodule llama.cpp to b6482 (3d4053f)
• Built and tested locally via Nix on Linux
• Bump crate versions: llama-cpp-sys-2 and llama-cpp-2 to 0.1.123

Notes:

• No changes to CI workflows or .github/*

[MarcusDunn]

If you could revert the version changes, those are handled automatically through actions on release.

[MarcusDunn]

It also looks like this needs a cargo fmt

[admiralakber]

No worries, all done.

I do notice that llama-cpp-2/Cargo.toml references llama-cpp-2-sys by version 0.1.113 which is behind what's defined in llama-cpp-sys-2/Cargo.toml is that intentional?

Thanks for the feedback and help.

[MarcusDunn]

Thanks!

Everything should be as it was set in #822. (one version ahead of latest on crates.io), if that was changed inadvertently:
a) thanks for catching it! would have had be confused come release time
b) please adjust back!

Also the cargo lock seems incorrect - you may have to revert those changes manually as well.

[admiralakber]

Hey Marcus,

Cargo.lock should be fixed now referencing 0.1.122 as per #822

Also in the latest commit, you should see specifically under dependencies where it seemed to lag behind. Is this indicating a potential tweak needed for the github action?

[MarcusDunn]

I'm not sure, I'll take a closer look when I do the next release (once this is merged), I think everything is correct as is.

[Kenrik]

Hello👋 So, the version of Llama.cpp in the current release is completely borked with GPT-OSS. To the extent possible please merge this PR soonish 🙏

[MarcusDunn]

Merged and released in 0.1.122