# Add comprehensive Build_Guide.md for llama-cpp-python with CUDA 12.8 support for Gemma 3
## This guide provides detailed step-by-step instructions for building a working llama-cpp-python (v0.3.8) wheel with NVIDIA CUDA 12.8 acceleration on Windows 10/11 (x64) that properly supports Google's Gemma 3 models with GPU offloading. Includes prerequisites, environment variable configuration, build process, verification steps, and troubleshooting tips. Created to fill the ecosystem gap for Windows users needing GPU-accelerated Gemma 3 inference via llama.cpp.
# llama-cpp-python Prebuilt Wheel (Windows x64, CUDA 12.8, Gemma 3 Support)
---
🛠️ **Built with** [llama.cpp (b5192)](https://github.com/ggml-org/llama.cpp) + [CUDA 12.8](https://developer.nvidia.com/cuda-toolkit)
---
**Prebuilt `.whl` for llama-cpp-python 0.3.8 — CUDA 12.8 acceleration with full Gemma 3 model support (Windows x64).**
This repository provides a prebuilt Python wheel (`.whl`) file for **llama-cpp-python**, specifically compiled for Windows 10/11 (x64) with NVIDIA CUDA 12.8 acceleration enabled.
Building `llama-cpp-python` with CUDA support on Windows can be a complex process involving specific Visual Studio configurations, CUDA Toolkit setup, and environment variables. This prebuilt wheel aims to simplify installation for users with compatible systems.
This build is based on **llama-cpp-python** version `0.3.8` of the Python bindings, and the underlying **llama.cpp** source code as of **April 26, 2025**. It has been verified to work with **Gemma 3 models**, correct…
