M.I.M.I.R - Multi-agent Intelligent Memory & Insight Repository

# llama.cpp Windows CUDA static library builder # Builds pre-compiled libraries for Windows x64 with CUDA support # # NOTE: This is primarily for CI/CD. Windows Docker containers with CUDA # require specific NVIDIA Windows container runtime support. # # For local builds, use: .\scripts\build-llama-cuda.ps1 # # Build (requires Windows container mode): # docker build -f docker/Dockerfile.llama-cuda-windows -t timothyswt/llama-cuda-libs-windows:7285 . # # For GitHub Actions: # See .github/workflows/build-llama-windows.yml # Using Windows Server Core with CUDA # Note: Requires host with NVIDIA GPU and Windows container support FROM mcr.microsoft.com/windows/servercore:ltsc2022 AS builder # Install Chocolatey for package management SHELL ["powershell", "-Command", "$ErrorActionPreference = 'Stop'; $ProgressPreference = 'SilentlyContinue';"] RUN Set-ExecutionPolicy Bypass -Scope Process -Force; \ [System.Net.ServicePointManager]::SecurityProtocol = [System.Net.ServicePointManager]::SecurityProtocol -bor 3072; \ iex ((New-Object System.Net.WebClient).DownloadString('https://chocolatey.org/install.ps1')) # Install build tools RUN choco install -y git cmake ninja visualstudio2022buildtools --package-parameters "--add Microsoft.VisualStudio.Workload.VCTools --includeRecommended" # Note: CUDA Toolkit must be pre-installed on the host or use NVIDIA's CUDA Windows container image # For CI, consider using: mcr.microsoft.com/dotnet/framework/sdk with manual CUDA install ARG LLAMA_VERSION=b4785 ARG CUDA_VERSION=12.6 WORKDIR C:\\llama # Clone llama.cpp RUN git clone --depth 1 --branch $env:LLAMA_VERSION https://github.com/ggerganov/llama.cpp.git . # Patch log.cpp for MSVC compatibility RUN $content = Get-Content 'common\\log.cpp' -Raw; \ if ($content -notmatch '#include\s*<chrono>') { \ $content = $content -replace '(#include\s*<cstdio>)', "`$1`n#include <chrono>"; \ Set-Content 'common\\log.cpp' $content -NoNewline; \ } # Setup VS environment and build SHELL ["cmd", "/S", "/C"] RUN call "C:\Program Files\Microsoft Visual Studio\2022\BuildTools\VC\Auxiliary\Build\vcvars64.bat" && cmake -B build -G Ninja -DCMAKE_BUILD_TYPE=Release -DLLAMA_STATIC=ON -DBUILD_SHARED_LIBS=OFF -DLLAMA_BUILD_TESTS=OFF -DLLAMA_BUILD_EXAMPLES=OFF -DLLAMA_BUILD_SERVER=OFF -DGGML_CUDA=ON -DGGML_CUDA_FA_ALL_QUANTS=ON && cmake --build build --config Release -j %NUMBER_OF_PROCESSORS% # Combine libraries SHELL ["powershell", "-Command", "$ErrorActionPreference = 'Stop';"] RUN New-Item -ItemType Directory -Force -Path C:\output\lib, C:\output\include; \ Get-ChildItem -Path build -Recurse -Filter "*.lib" | Where-Object { $_.Name -match "llama|ggml" } | Copy-Item -Destination C:\output\lib; \ Copy-Item include\llama.h C:\output\include\; \ Copy-Item ggml\include\*.h C:\output\include\; \ Write-Host "Libraries:"; Get-ChildItem C:\output\lib # Output stage (minimal) FROM mcr.microsoft.com/windows/nanoserver:ltsc2022 COPY --from=builder C:\\output C:\\output