v3.18.2-gemma.1

Trimmed node-llama-cpp build for the gemma-llama server.

Each tarball is platform-specific (binário nativo embutido, postinstall removido).

Build flags ativados nesta release:

• GGML_LTO=ON (todas as plataformas)
• macOS arm64: GGML_METAL_EMBED_LIBRARY=1, GGML_ACCELERATE=ON, MACOSX_DEPLOYMENT_TARGET=12.0
• CUDA (quando o runner expõe toolkit): GGML_CUDA_FA_ALL_QUANTS=ON, CMAKE_CUDA_ARCHITECTURES=native

Install

// package.json
{
  "dependencies": {
    "node-llama-cpp": "https://github.com/rafaelcaires/node-llama-cpp/releases/download/v3.18.2-gemma.1/node-llama-cpp-3.18.2-gemma.1-<plataforma>.tgz"
  }
}

Plataformas disponíveis: win-x64, win-x64-vulkan, linux-x64, mac-arm64-metal.