Skip to content

Koboldcpp Windows

Koboldcpp Windows #2

name: Koboldcpp Windows
on:
workflow_dispatch:
inputs:
commit_hash:
description: 'Optional commit hash to build from'
required: false
default: ''
env:
BRANCH_NAME: ${{ github.head_ref || github.ref_name }}
jobs:
windows:
runs-on: windows-2022
steps:
- name: Clone
id: checkout
uses: actions/checkout@v3
with:
ref: ${{ inputs.commit_hash != '' && inputs.commit_hash || github.head_ref || github.ref_name }}
- name: Show Commit Used
run: |
echo "Building from ref: ${{ inputs.commit_hash != '' && inputs.commit_hash || github.head_ref || github.ref_name }}"
- name: Get Python
uses: actions/setup-python@v2
with:
python-version: 3.8.10
- name: Install python dependencies
run: |
python -m pip install --upgrade pip
pip install customtkinter==5.2.0 pyinstaller==5.11.0 psutil==5.9.5
- name: Display full Visual Studio info Before
run: |
& "C:\Program Files (x86)\Microsoft Visual Studio\Installer\vswhere.exe" -all -products * -format json
shell: pwsh
- name: Download and install win64devkit
run: |
curl -L https://github.com/skeeto/w64devkit/releases/download/v1.22.0/w64devkit-1.22.0.zip --output w64devkit.zip
Expand-Archive w64devkit.zip -DestinationPath .
- name: Add w64devkit to PATH
run: |
echo "$(Get-Location)\w64devkit\bin" | Out-File -Append -FilePath $env:GITHUB_PATH -Encoding utf8
- name: Print System Environment Variables
id: printvars
run: |
echo "Number of processors: ${env:NUMBER_OF_PROCESSORS}"
echo "Processor Architecture: ${env:PROCESSOR_ARCHITECTURE}"
echo "Computer Name: ${env:COMPUTERNAME}"
wmic cpu get name
wmic os get TotalVisibleMemorySize, FreePhysicalMemory
- name: Build Non-CUDA
id: make_build
run: |
make LLAMA_CLBLAST=1 LLAMA_VULKAN=1 LLAMA_PORTABLE=1 -j ${env:NUMBER_OF_PROCESSORS}
- uses: Jimver/[email protected]
id: cuda-toolkit
with:
cuda: '12.4.0'
use-github-cache: false
- name: Build CUDA
id: cmake_build
run: |
mkdir build
cd build
cmake .. -DLLAMA_CUBLAS=ON -DCMAKE_SYSTEM_VERSION="10.0.19041.0"
cmake --build . --config Release -j 2
cd ..
# note: The libraries that come from the github cuda directory seem to be larger, so they are not recommended
# - name: Download CuBLAS Libraries
# run: |
# curl -L https://github.com/LostRuins/koboldcpp/releases/download/cuda11_cublas_libraries/cublas64_11.dll --output cublas64_11.dll
# curl -L https://github.com/LostRuins/koboldcpp/releases/download/cuda11_cublas_libraries/cublasLt64_11.dll --output cublasLt64_11.dll
# ls
- name: Copy CuBLAS Libraries
run: |
copy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.4\bin\cublasLt64_12.dll" .
copy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.4\bin\cublas64_12.dll" .
ls
- name: Package PyInstallers
id: make_pyinstaller
run: |
./make_pyinstaller.bat
./make_pyinstaller_cuda.bat
- name: Save artifact
uses: actions/upload-artifact@v4
with:
name: kcpp_windows_pyinstallers
path: dist/
- name: Generate VK Instructions
id: gen_vk_instructions
run: |
echo "If you cannot compile vulkan shaders yourself with glslc, you can manually patch in precompiled vulkan shader source files. Copy ggml-vulkan-shaders.cpp and ggml-vulkan-shaders.hpp to the ggml/src subdirectory in KoboldCpp source files before building." > vulkan-readme.txt
- name: Save Standalone Vulkan Shaders
uses: actions/upload-artifact@v4
with:
name: vulkan_precompiled_shaders
path: |
ggml/src/ggml-vulkan-shaders.cpp
ggml/src/ggml-vulkan-shaders.hpp
vulkan-readme.txt