From b44a82d547b71963016835fa1ffb439c867a25ed Mon Sep 17 00:00:00 2001 From: Victor Biederbeck Date: Mon, 12 Oct 2026 14:58:25 -0700 Subject: [PATCH] chore: bump vendor/llama.cpp to support gemma-embedding2 (EmbeddingGemma 2) Bumps the vendored llama.cpp submodule to ggml-org/llama.cpp@4fbc76dec ("model: support embeddinggemma2 (text+vision+audio)", #30054), the commit that added the `gemma-embedding2` GGUF architecture. Without it, loading an EmbeddingGemma 2 GGUF fails with: error loading model architecture: unknown model architecture: 'gemma-embedding2' No llama-cpp-python Python binding changes are needed: the new architecture is registered entirely inside llama.cpp's internal model loader (src/llama-arch.*, src/models/gemma-embedding2.cpp) and doesn't touch the public C API in include/llama.h that the ctypes bindings mirror (the only llama.h diff in this range is two internal session/state-file version-number bumps, unrelated to this feature). --- vendor/llama.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/vendor/llama.cpp b/vendor/llama.cpp index 0c1e57098..4fbc76dec 160000 --- a/vendor/llama.cpp +++ b/vendor/llama.cpp @@ -1 +1 @@ -Subproject commit 0c1e57098bba43ac29e6e3b677cdceebdd22334f +Subproject commit 4fbc76dec51d0add466f0210855c0596589b60d4