vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Add XCFramework building

Browse source
This commit is contained in:
Philipp Zagar 2024-02-28 12:19:44 -08:00
parent 87c91c0766
commit 8776d1c051
No known key found for this signature in database
GPG key ID: 263361B2C2850C93
64 changed files with 10658 additions and 2677 deletions
Download patch file Download diff file Expand all files Collapse all files

11
.github/ISSUE_TEMPLATE/bug.md vendored
View file

@ -1,11 +0,0 @@
---
name: Bug template
about: Used to report bugs in llama.cpp
labels: ["bug-unconfirmed"]
assignees: ''
---
Please include information about your system, the steps to reproduce the bug, and the version of llama.cpp that you are using. If possible, please provide a minimal code example that reproduces the bug.
If the bug concerns the server, please try to reproduce it first using the [server test scenario framework](https://github.com/ggerganov/llama.cpp/tree/master/examples/server/tests).

28
.github/ISSUE_TEMPLATE/enhancement.md vendored
View file

@ -1,28 +0,0 @@
---
name: Enhancement template
about: Used to request enhancements for llama.cpp
labels: ["enhancement"]
assignees: ''
---
# Prerequisites
Please answer the following questions for yourself before submitting an issue.
- [ ] I am running the latest code. Development is very rapid so there are no tagged versions as of now.
- [ ] I carefully followed the [README.md](https://github.com/ggerganov/llama.cpp/blob/master/README.md).
- [ ] I [searched using keywords relevant to my issue](https://docs.github.com/en/issues/tracking-your-work-with-issues/filtering-and-searching-issues-and-pull-requests) to make sure that I am creating a new issue that is not already open (or closed).
- [ ] I reviewed the [Discussions](https://github.com/ggerganov/llama.cpp/discussions), and have a new bug or useful enhancement to share.
# Feature Description
Please provide a detailed written description of what you were trying to do, and what you expected `llama.cpp` to do as an enhancement.
# Motivation
Please provide a detailed written description of reasons why this feature is necessary and how it is useful to `llama.cpp` users.
# Possible Implementation
If you have an idea as to how it can be implemented, please write a detailed description. Feel free to give links to external sources or share visuals that might be helpful to understand the details better.

968
.github/workflows/build.yml vendored
View file

@ -1,949 +1,31 @@
name: CI
#
# This source file is part of the Stanford Biodesign Digital Health Group open-source organization
#
# SPDX-FileCopyrightText: 2022 Stanford University and the project authors (see CONTRIBUTORS.md)
#
# SPDX-License-Identifier: MIT
#
name: Build XCArchive
on:
workflow_dispatch: # allows manual triggering
inputs:
create_release:
description: 'Create new release'
required: true
type: boolean
push:
branches:
- master
paths: ['.github/workflows/**', '**/CMakeLists.txt', '**/Makefile', '**/*.h', '**/*.hpp', '**/*.c', '**/*.cpp', '**/*.cu', '**/*.swift', '**/*.m']
pull_request:
types: [opened, synchronize, reopened]
paths: ['**/CMakeLists.txt', '**/Makefile', '**/*.h', '**/*.hpp', '**/*.c', '**/*.cpp', '**/*.cu', '**/*.swift', '**/*.m']
env:
BRANCH_NAME: ${{ github.head_ref || github.ref_name }}
GGML_NLOOP: 3
GGML_N_THREADS: 1
workflow_dispatch:
workflow_call:
inputs:
version:
description: 'The version number of the framework embedded in the XCArchives.'
type: string
required: true
jobs:
ubuntu-focal-make:
runs-on: ubuntu-20.04
steps:
- name: Clone
id: checkout
uses: actions/checkout@v3
- name: Dependencies
id: depends
run: |
sudo apt-get update
sudo apt-get install build-essential gcc-8
- name: Build
id: make_build
env:
LLAMA_FATAL_WARNINGS: 1
run: |
CC=gcc-8 make -j $(nproc)
- name: Test
id: make_test
run: |
CC=gcc-8 make tests -j $(nproc)
make test -j $(nproc)
ubuntu-latest-cmake:
runs-on: ubuntu-latest
steps:
- name: Clone
id: checkout
uses: actions/checkout@v3
- name: Dependencies
id: depends
run: |
sudo apt-get update
sudo apt-get install build-essential
- name: Build
id: cmake_build
run: |
mkdir build
cd build
cmake .. -DLLAMA_FATAL_WARNINGS=ON
cmake --build . --config Release -j $(nproc)
- name: Test
id: cmake_test
run: |
cd build
ctest -L main --verbose --timeout 900
ubuntu-latest-cmake-sanitizer:
runs-on: ubuntu-latest
continue-on-error: true
strategy:
matrix:
sanitizer: [ADDRESS, THREAD, UNDEFINED]
build_type: [Debug, Release]
steps:
- name: Clone
id: checkout
uses: actions/checkout@v3
- name: Dependencies
id: depends
run: |
sudo apt-get update
sudo apt-get install build-essential
- name: Build
id: cmake_build
run: |
mkdir build
cd build
cmake .. -DLLAMA_FATAL_WARNINGS=ON -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON -DCMAKE_BUILD_TYPE=${{ matrix.build_type }}
cmake --build . --config ${{ matrix.build_type }} -j $(nproc)
- name: Test
id: cmake_test
run: |
cd build
ctest -L main --verbose --timeout 900
ubuntu-latest-cmake-mpi:
runs-on: ubuntu-latest
continue-on-error: true
strategy:
matrix:
mpi_library: [mpich, libopenmpi-dev]
steps:
- name: Clone
id: checkout
uses: actions/checkout@v3
- name: Dependencies
id: depends
run: |
sudo apt-get update
sudo apt-get install build-essential ${{ matrix.mpi_library }}
- name: Build
id: cmake_build
run: |
mkdir build
cd build
cmake -DLLAMA_MPI=ON ..
cmake --build . --config Release -j $(nproc)
- name: Test
id: cmake_test
run: |
cd build
ctest -L main --verbose
ubuntu-22-cmake-sycl:
runs-on: ubuntu-22.04
continue-on-error: true
steps:
- uses: actions/checkout@v2
- name: add oneAPI to apt
shell: bash
run: |
cd /tmp
wget https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB
sudo apt-key add GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB
rm GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB
sudo add-apt-repository "deb https://apt.repos.intel.com/oneapi all main"
- name: install oneAPI dpcpp compiler
shell: bash
run: |
sudo apt update
sudo apt install intel-oneapi-compiler-dpcpp-cpp
- name: install oneAPI MKL library
shell: bash
run: |
sudo apt install intel-oneapi-mkl-devel
- name: Clone
id: checkout
uses: actions/checkout@v3
- name: Build
id: cmake_build
run: |
source /opt/intel/oneapi/setvars.sh
mkdir build
cd build
cmake -DLLAMA_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx ..
cmake --build . --config Release -j $(nproc)
ubuntu-22-cmake-sycl-fp16:
runs-on: ubuntu-22.04
continue-on-error: true
steps:
- uses: actions/checkout@v2
- name: add oneAPI to apt
shell: bash
run: |
cd /tmp
wget https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB
sudo apt-key add GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB
rm GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB
sudo add-apt-repository "deb https://apt.repos.intel.com/oneapi all main"
- name: install oneAPI dpcpp compiler
shell: bash
run: |
sudo apt update
sudo apt install intel-oneapi-compiler-dpcpp-cpp
- name: install oneAPI MKL library
shell: bash
run: |
sudo apt install intel-oneapi-mkl-devel
- name: Clone
id: checkout
uses: actions/checkout@v3
- name: Build
id: cmake_build
run: |
source /opt/intel/oneapi/setvars.sh
mkdir build
cd build
cmake -DLLAMA_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DLLAMA_SYCL_F16=ON ..
cmake --build . --config Release -j $(nproc)
# TODO: build with LLAMA_NO_METAL because test-backend-ops fail on "Apple Paravirtual device" and I don't know
# how to debug it.
# ref: https://github.com/ggerganov/llama.cpp/actions/runs/7131777249/job/19420981052#step:5:1124
macOS-latest-make:
runs-on: macos-latest
steps:
- name: Clone
id: checkout
uses: actions/checkout@v3
- name: Dependencies
id: depends
continue-on-error: true
run: |
brew update
- name: Build
id: make_build
env:
LLAMA_FATAL_WARNINGS: 1
run: |
LLAMA_NO_METAL=1 make -j $(sysctl -n hw.logicalcpu)
- name: Test
id: make_test
run: |
LLAMA_NO_METAL=1 make tests -j $(sysctl -n hw.logicalcpu)
LLAMA_NO_METAL=1 make test -j $(sysctl -n hw.logicalcpu)
# TODO: build with LLAMA_METAL=OFF because test-backend-ops fail on "Apple Paravirtual device" and I don't know
# how to debug it.
# ref: https://github.com/ggerganov/llama.cpp/actions/runs/7132125951/job/19422043567?pr=4359#step:5:6584
# would be great if we fix these
macOS-latest-cmake:
runs-on: macos-latest
steps:
- name: Clone
id: checkout
uses: actions/checkout@v3
- name: Dependencies
id: depends
continue-on-error: true
run: |
brew update
- name: Build
id: cmake_build
run: |
sysctl -a
mkdir build
cd build
cmake -DLLAMA_FATAL_WARNINGS=ON -DLLAMA_METAL=OFF ..
cmake --build . --config Release -j $(sysctl -n hw.logicalcpu)
- name: Test
id: cmake_test
run: |
cd build
ctest -L main --verbose --timeout 900
macOS-latest-cmake-ios:
runs-on: macos-latest
steps:
- name: Clone
id: checkout
uses: actions/checkout@v1
- name: Dependencies
id: depends
continue-on-error: true
run: |
brew update
- name: Build
id: cmake_build
run: |
sysctl -a
mkdir build
cd build
cmake -G Xcode .. \
-DLLAMA_BUILD_EXAMPLES=OFF \
-DLLAMA_BUILD_TESTS=OFF \
-DLLAMA_BUILD_SERVER=OFF \
-DCMAKE_SYSTEM_NAME=iOS \
-DCMAKE_OSX_DEPLOYMENT_TARGET=14.0
cmake --build . --config Release -j $(sysctl -n hw.logicalcpu)
macOS-latest-cmake-tvos:
runs-on: macos-latest
steps:
- name: Clone
id: checkout
uses: actions/checkout@v1
- name: Dependencies
id: depends
continue-on-error: true
run: |
brew update
- name: Build
id: cmake_build
run: |
sysctl -a
mkdir build
cd build
cmake -G Xcode .. \
-DLLAMA_BUILD_EXAMPLES=OFF \
-DLLAMA_BUILD_TESTS=OFF \
-DLLAMA_BUILD_SERVER=OFF \
-DCMAKE_SYSTEM_NAME=tvOS \
-DCMAKE_OSX_DEPLOYMENT_TARGET=14.0
cmake --build . --config Release -j $(sysctl -n hw.logicalcpu)
macOS-latest-swift:
runs-on: macos-latest
strategy:
matrix:
destination: ['generic/platform=macOS', 'generic/platform=iOS', 'generic/platform=tvOS']
steps:
- name: Clone
id: checkout
uses: actions/checkout@v1
- name: Dependencies
id: depends
continue-on-error: true
run: |
brew update
- name: xcodebuild for swift package
id: xcodebuild
run: |
xcodebuild -scheme llama -destination "${{ matrix.destination }}"
- name: Build Swift Example
id: make_build_swift_example
run: |
make swift
windows-latest-cmake:
runs-on: windows-latest
env:
OPENBLAS_VERSION: 0.3.23
OPENCL_VERSION: 2023.04.17
CLBLAST_VERSION: 1.6.0
SDE_VERSION: 9.33.0-2024-01-07
VULKAN_VERSION: 1.3.261.1
strategy:
matrix:
include:
- build: 'noavx'
defines: '-DLLAMA_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DLLAMA_AVX=OFF -DLLAMA_AVX2=OFF -DLLAMA_FMA=OFF -DBUILD_SHARED_LIBS=ON'
- build: 'avx2'
defines: '-DLLAMA_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DBUILD_SHARED_LIBS=ON'
- build: 'avx'
defines: '-DLLAMA_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DLLAMA_AVX2=OFF -DBUILD_SHARED_LIBS=ON'
- build: 'avx512'
defines: '-DLLAMA_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DLLAMA_AVX512=ON -DBUILD_SHARED_LIBS=ON'
- build: 'clblast'
defines: '-DLLAMA_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DLLAMA_CLBLAST=ON -DBUILD_SHARED_LIBS=ON -DCMAKE_PREFIX_PATH="$env:RUNNER_TEMP/clblast"'
- build: 'openblas'
defines: '-DLLAMA_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DLLAMA_BLAS=ON -DBUILD_SHARED_LIBS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS -DBLAS_INCLUDE_DIRS="$env:RUNNER_TEMP/openblas/include" -DBLAS_LIBRARIES="$env:RUNNER_TEMP/openblas/lib/openblas.lib"'
- build: 'kompute'
defines: '-DLLAMA_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DLLAMA_KOMPUTE=ON -DKOMPUTE_OPT_DISABLE_VULKAN_VERSION_CHECK=ON -DBUILD_SHARED_LIBS=ON'
- build: 'vulkan'
defines: '-DLLAMA_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DLLAMA_VULKAN=ON -DBUILD_SHARED_LIBS=ON'
steps:
- name: Clone
id: checkout
uses: actions/checkout@v3
build-xcarchive:
uses: StanfordBDHG/.github/.github/workflows/archive.yml@v2
with:
fetch-depth: 0
- name: Clone Kompute submodule
id: clone_kompute
if: ${{ matrix.build == 'kompute' }}
run: |
git submodule update --init kompute
- name: Download OpenCL SDK
id: get_opencl
if: ${{ matrix.build == 'clblast' }}
run: |
curl.exe -o $env:RUNNER_TEMP/opencl.zip -L "https://github.com/KhronosGroup/OpenCL-SDK/releases/download/v${env:OPENCL_VERSION}/OpenCL-SDK-v${env:OPENCL_VERSION}-Win-x64.zip"
mkdir $env:RUNNER_TEMP/opencl
tar.exe -xvf $env:RUNNER_TEMP/opencl.zip --strip-components=1 -C $env:RUNNER_TEMP/opencl
- name: Download CLBlast
id: get_clblast
if: ${{ matrix.build == 'clblast' }}
run: |
curl.exe -o $env:RUNNER_TEMP/clblast.7z -L "https://github.com/CNugteren/CLBlast/releases/download/${env:CLBLAST_VERSION}/CLBlast-${env:CLBLAST_VERSION}-windows-x64.7z"
curl.exe -o $env:RUNNER_TEMP/CLBlast.LICENSE.txt -L "https://github.com/CNugteren/CLBlast/raw/${env:CLBLAST_VERSION}/LICENSE"
7z x "-o${env:RUNNER_TEMP}" $env:RUNNER_TEMP/clblast.7z
rename-item $env:RUNNER_TEMP/CLBlast-${env:CLBLAST_VERSION}-windows-x64 clblast
foreach ($f in (gci -Recurse -Path "$env:RUNNER_TEMP/clblast" -Filter '*.cmake')) {
$txt = Get-Content -Path $f -Raw
$txt.Replace('C:/vcpkg/packages/opencl_x64-windows/', "$($env:RUNNER_TEMP.Replace('\','/'))/opencl/") | Set-Content -Path $f -Encoding UTF8
}
- name: Download OpenBLAS
id: get_openblas
if: ${{ matrix.build == 'openblas' }}
run: |
curl.exe -o $env:RUNNER_TEMP/openblas.zip -L "https://github.com/xianyi/OpenBLAS/releases/download/v${env:OPENBLAS_VERSION}/OpenBLAS-${env:OPENBLAS_VERSION}-x64.zip"
curl.exe -o $env:RUNNER_TEMP/OpenBLAS.LICENSE.txt -L "https://github.com/xianyi/OpenBLAS/raw/v${env:OPENBLAS_VERSION}/LICENSE"
mkdir $env:RUNNER_TEMP/openblas
tar.exe -xvf $env:RUNNER_TEMP/openblas.zip -C $env:RUNNER_TEMP/openblas
$vcdir = $(vswhere -latest -products * -requires Microsoft.VisualStudio.Component.VC.Tools.x86.x64 -property installationPath)
$msvc = $(join-path $vcdir $('VC\Tools\MSVC\'+$(gc -raw $(join-path $vcdir 'VC\Auxiliary\Build\Microsoft.VCToolsVersion.default.txt')).Trim()))
$lib = $(join-path $msvc 'bin\Hostx64\x64\lib.exe')
& $lib /machine:x64 "/def:${env:RUNNER_TEMP}/openblas/lib/libopenblas.def" "/out:${env:RUNNER_TEMP}/openblas/lib/openblas.lib" /name:openblas.dll
- name: Install Vulkan SDK
id: get_vulkan
if: ${{ matrix.build == 'kompute' || matrix.build == 'vulkan' }}
run: |
curl.exe -o $env:RUNNER_TEMP/VulkanSDK-Installer.exe -L "https://sdk.lunarg.com/sdk/download/${env:VULKAN_VERSION}/windows/VulkanSDK-${env:VULKAN_VERSION}-Installer.exe"
& "$env:RUNNER_TEMP\VulkanSDK-Installer.exe" --accept-licenses --default-answer --confirm-command install
Add-Content $env:GITHUB_ENV "VULKAN_SDK=C:\VulkanSDK\${env:VULKAN_VERSION}"
Add-Content $env:GITHUB_PATH "C:\VulkanSDK\${env:VULKAN_VERSION}\bin"
- name: Build
id: cmake_build
run: |
mkdir build
cd build
cmake .. ${{ matrix.defines }}
cmake --build . --config Release -j ${env:NUMBER_OF_PROCESSORS}
- name: Add clblast.dll
id: add_clblast_dll
if: ${{ matrix.build == 'clblast' }}
run: |
cp $env:RUNNER_TEMP/clblast/lib/clblast.dll ./build/bin/Release
cp $env:RUNNER_TEMP/CLBlast.LICENSE.txt ./build/bin/Release/CLBlast-${env:CLBLAST_VERSION}.txt
- name: Add libopenblas.dll
id: add_libopenblas_dll
if: ${{ matrix.build == 'openblas' }}
run: |
cp $env:RUNNER_TEMP/openblas/bin/libopenblas.dll ./build/bin/Release/openblas.dll
cp $env:RUNNER_TEMP/OpenBLAS.LICENSE.txt ./build/bin/Release/OpenBLAS-${env:OPENBLAS_VERSION}.txt
- name: Check AVX512F support
id: check_avx512f
if: ${{ matrix.build == 'avx512' }}
continue-on-error: true
run: |
cd build
$vcdir = $(vswhere -latest -products * -requires Microsoft.VisualStudio.Component.VC.Tools.x86.x64 -property installationPath)
$msvc = $(join-path $vcdir $('VC\Tools\MSVC\'+$(gc -raw $(join-path $vcdir 'VC\Auxiliary\Build\Microsoft.VCToolsVersion.default.txt')).Trim()))
$cl = $(join-path $msvc 'bin\Hostx64\x64\cl.exe')
echo 'int main(void){unsigned int a[4];__cpuid(a,7);return !(a[1]&65536);}' >> avx512f.c
& $cl /O2 /GS- /kernel avx512f.c /link /nodefaultlib /entry:main
.\avx512f.exe && echo "AVX512F: YES" && ( echo HAS_AVX512F=1 >> $env:GITHUB_ENV ) || echo "AVX512F: NO"
- name: Test
id: cmake_test
# not all machines have native AVX-512
if: ${{ matrix.build != 'clblast' && matrix.build != 'kompute' && matrix.build != 'vulkan' && (matrix.build != 'avx512' || env.HAS_AVX512F == '1') }}
run: |
cd build
ctest -L main -C Release --verbose --timeout 900
- name: Test (Intel SDE)
id: cmake_test_sde
if: ${{ matrix.build == 'avx512' && env.HAS_AVX512F == '0' }} # use Intel SDE for AVX-512 emulation
run: |
curl.exe -o $env:RUNNER_TEMP/sde.tar.xz -L "https://downloadmirror.intel.com/813591/sde-external-${env:SDE_VERSION}-win.tar.xz"
# for some weird reason windows tar doesn't like sde tar.xz
7z x "-o${env:RUNNER_TEMP}" $env:RUNNER_TEMP/sde.tar.xz
7z x "-o${env:RUNNER_TEMP}" $env:RUNNER_TEMP/sde.tar
$sde = $(join-path $env:RUNNER_TEMP sde-external-${env:SDE_VERSION}-win/sde.exe)
cd build
& $sde -future -- ctest -L main -C Release --verbose --timeout 900
- name: Determine tag name
id: tag
shell: bash
run: |
BUILD_NUMBER="$(git rev-list --count HEAD)"
SHORT_HASH="$(git rev-parse --short=7 HEAD)"
if [[ "${{ env.BRANCH_NAME }}" == "master" ]]; then
echo "name=b${BUILD_NUMBER}" >> $GITHUB_OUTPUT
else
SAFE_NAME=$(echo "${{ env.BRANCH_NAME }}" | tr '/' '-')
echo "name=${SAFE_NAME}-b${BUILD_NUMBER}-${SHORT_HASH}" >> $GITHUB_OUTPUT
fi
- name: Pack artifacts
id: pack_artifacts
if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
run: |
Copy-Item LICENSE .\build\bin\Release\llama.cpp.txt
7z a llama-${{ steps.tag.outputs.name }}-bin-win-${{ matrix.build }}-x64.zip .\build\bin\Release\*
- name: Upload artifacts
if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
uses: actions/upload-artifact@v3
with:
path: |
llama-${{ steps.tag.outputs.name }}-bin-win-${{ matrix.build }}-x64.zip
windows-latest-cmake-cublas:
runs-on: windows-latest
strategy:
matrix:
cuda: ['12.2.0', '11.7.1']
build: ['cublas']
steps:
- name: Clone
id: checkout
uses: actions/checkout@v3
with:
fetch-depth: 0
- uses: Jimver/cuda-toolkit@v0.2.11
id: cuda-toolkit
with:
cuda: ${{ matrix.cuda }}
method: 'network'
sub-packages: '["nvcc", "cudart", "cublas", "cublas_dev", "thrust", "visual_studio_integration"]'
- name: Build
id: cmake_build
run: |
mkdir build
cd build
cmake .. -DLLAMA_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DLLAMA_CUBLAS=ON -DBUILD_SHARED_LIBS=ON
cmake --build . --config Release -j ${env:NUMBER_OF_PROCESSORS}
- name: Determine tag name
id: tag
shell: bash
run: |
BUILD_NUMBER="$(git rev-list --count HEAD)"
SHORT_HASH="$(git rev-parse --short=7 HEAD)"
if [[ "${{ env.BRANCH_NAME }}" == "master" ]]; then
echo "name=b${BUILD_NUMBER}" >> $GITHUB_OUTPUT
else
SAFE_NAME=$(echo "${{ env.BRANCH_NAME }}" | tr '/' '-')
echo "name=${SAFE_NAME}-b${BUILD_NUMBER}-${SHORT_HASH}" >> $GITHUB_OUTPUT
fi
- name: Pack artifacts
id: pack_artifacts
if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
run: |
7z a llama-${{ steps.tag.outputs.name }}-bin-win-${{ matrix.build }}-cu${{ matrix.cuda }}-x64.zip .\build\bin\Release\*
- name: Upload artifacts
if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
uses: actions/upload-artifact@v3
with:
path: |
llama-${{ steps.tag.outputs.name }}-bin-win-${{ matrix.build }}-cu${{ matrix.cuda }}-x64.zip
- name: Copy and pack Cuda runtime
run: |
echo "Cuda install location: ${{steps.cuda-toolkit.outputs.CUDA_PATH}}"
$dst='.\build\bin\cudart\'
robocopy "${{steps.cuda-toolkit.outputs.CUDA_PATH}}\bin" $dst cudart64_*.dll cublas64_*.dll cublasLt64_*.dll
7z a cudart-llama-bin-win-cu${{ matrix.cuda }}-x64.zip $dst\*
- name: Upload Cuda runtime
if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
uses: actions/upload-artifact@v3
with:
path: |
cudart-llama-bin-win-cu${{ matrix.cuda }}-x64.zip
windows-latest-cmake-sycl:
runs-on: windows-latest
defaults:
run:
shell: bash
env:
WINDOWS_BASEKIT_URL: https://registrationcenter-download.intel.com/akdlm/IRC_NAS/62641e01-1e8d-4ace-91d6-ae03f7f8a71f/w_BaseKit_p_2024.0.0.49563_offline.exe
WINDOWS_DPCPP_MKL: intel.oneapi.win.cpp-dpcpp-common:intel.oneapi.win.mkl.devel
steps:
- name: Clone
id: checkout
uses: actions/checkout@v3
with:
fetch-depth: 0
- name: Install
run: scripts/install-oneapi.bat $WINDOWS_BASEKIT_URL $WINDOWS_DPCPP_MKL
- name: Build
id: cmake_build
run: examples/sycl/win-build-sycl.bat
ios-xcode-build:
runs-on: macos-latest
steps:
- name: Checkout code
uses: actions/checkout@v3
- name: Build Xcode project
run: xcodebuild -project examples/llama.swiftui/llama.swiftui.xcodeproj -scheme llama.swiftui -sdk iphoneos CODE_SIGNING_REQUIRED=NO CODE_SIGN_IDENTITY= -destination 'generic/platform=iOS' build
android-build:
runs-on: ubuntu-latest
steps:
- name: Clone
uses: actions/checkout@v3
- name: Set up JDK
uses: actions/setup-java@v3
with:
java-version: 17
distribution: zulu
- name: Setup Android SDK
uses: android-actions/setup-android@v3
with:
log-accepted-android-sdk-licenses: false
- name: Build
run: |
cd examples/llama.android
./gradlew build --no-daemon
# freeBSD-latest:
# runs-on: macos-12
# steps:
# - name: Clone
# uses: actions/checkout@v3
#
# - name: Build
# uses: cross-platform-actions/action@v0.19.0
# with:
# operating_system: freebsd
# version: '13.2'
# hypervisor: 'qemu'
# run: |
# sudo pkg update
# sudo pkg install -y gmake automake autoconf pkgconf llvm15 clinfo clover opencl clblast openblas
# gmake CC=/usr/local/bin/clang15 CXX=/usr/local/bin/clang++15 -j `sysctl -n hw.ncpu`
release:
if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
runs-on: ubuntu-latest
needs:
- ubuntu-focal-make
- ubuntu-latest-cmake
- macOS-latest-make
- macOS-latest-cmake
- windows-latest-cmake
- windows-latest-cmake-cublas
steps:
- name: Clone
id: checkout
uses: actions/checkout@v3
with:
fetch-depth: 0
- name: Determine tag name
id: tag
shell: bash
run: |
BUILD_NUMBER="$(git rev-list --count HEAD)"
SHORT_HASH="$(git rev-parse --short=7 HEAD)"
if [[ "${{ env.BRANCH_NAME }}" == "master" ]]; then
echo "name=b${BUILD_NUMBER}" >> $GITHUB_OUTPUT
else
SAFE_NAME=$(echo "${{ env.BRANCH_NAME }}" | tr '/' '-')
echo "name=${SAFE_NAME}-b${BUILD_NUMBER}-${SHORT_HASH}" >> $GITHUB_OUTPUT
fi
- name: Download artifacts
id: download-artifact
uses: actions/download-artifact@v3
- name: Create release
id: create_release
uses: anzz1/action-create-release@v1
env:
GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
with:
tag_name: ${{ steps.tag.outputs.name }}
- name: Upload release
id: upload_release
uses: actions/github-script@v3
with:
github-token: ${{secrets.GITHUB_TOKEN}}
script: |
const path = require('path');
const fs = require('fs');
const release_id = '${{ steps.create_release.outputs.id }}';
for (let file of await fs.readdirSync('./artifact')) {
if (path.extname(file) === '.zip') {
console.log('uploadReleaseAsset', file);
await github.repos.uploadReleaseAsset({
owner: context.repo.owner,
repo: context.repo.repo,
release_id: release_id,
name: file,
data: await fs.readFileSync(`./artifact/${file}`)
});
}
}
# ubuntu-latest-gcc:
# runs-on: ubuntu-latest
#
# strategy:
# matrix:
# build: [Debug, Release]
#
# steps:
# - name: Clone
# uses: actions/checkout@v3
#
# - name: Dependencies
# run: |
# sudo apt-get update
# sudo apt-get install build-essential
# sudo apt-get install cmake
#
# - name: Configure
# run: cmake . -DCMAKE_BUILD_TYPE=${{ matrix.build }}
#
# - name: Build
# run: |
# make
#
# ubuntu-latest-clang:
# runs-on: ubuntu-latest
#
# strategy:
# matrix:
# build: [Debug, Release]
#
# steps:
# - name: Clone
# uses: actions/checkout@v3
#
# - name: Dependencies
# run: |
# sudo apt-get update
# sudo apt-get install build-essential
# sudo apt-get install cmake
#
# - name: Configure
# run: cmake . -DCMAKE_BUILD_TYPE=${{ matrix.build }} -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_C_COMPILER=clang
#
# - name: Build
# run: |
# make
#
# ubuntu-latest-gcc-sanitized:
# runs-on: ubuntu-latest
#
# strategy:
# matrix:
# sanitizer: [ADDRESS, THREAD, UNDEFINED]
#
# steps:
# - name: Clone
# uses: actions/checkout@v3
#
# - name: Dependencies
# run: |
# sudo apt-get update
# sudo apt-get install build-essential
# sudo apt-get install cmake
#
# - name: Configure
# run: cmake . -DCMAKE_BUILD_TYPE=Debug -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON
#
# - name: Build
# run: |
# make
#
# windows:
# runs-on: windows-latest
#
# strategy:
# matrix:
# build: [Release]
# arch: [Win32, x64]
# include:
# - arch: Win32
# s2arc: x86
# - arch: x64
# s2arc: x64
#
# steps:
# - name: Clone
# uses: actions/checkout@v3
#
# - name: Add msbuild to PATH
# uses: microsoft/setup-msbuild@v1
#
# - name: Configure
# run: >
# cmake -S . -B ./build -A ${{ matrix.arch }}
# -DCMAKE_BUILD_TYPE=${{ matrix.build }}
#
# - name: Build
# run: |
# cd ./build
# msbuild ALL_BUILD.vcxproj -t:build -p:configuration=${{ matrix.build }} -p:platform=${{ matrix.arch }}
#
# - name: Upload binaries
# uses: actions/upload-artifact@v1
# with:
# name: llama-bin-${{ matrix.arch }}
# path: build/bin/${{ matrix.build }}
#
# windows-blas:
# runs-on: windows-latest
#
# strategy:
# matrix:
# build: [Release]
# arch: [Win32, x64]
# blas: [ON]
# include:
# - arch: Win32
# obzip: https://github.com/xianyi/OpenBLAS/releases/download/v0.3.21/OpenBLAS-0.3.21-x86.zip
# s2arc: x86
# - arch: x64
# obzip: https://github.com/xianyi/OpenBLAS/releases/download/v0.3.21/OpenBLAS-0.3.21-x64.zip
# s2arc: x64
#
# steps:
# - name: Clone
# uses: actions/checkout@v3
#
# - name: Add msbuild to PATH
# uses: microsoft/setup-msbuild@v1
#
# - name: Fetch OpenBLAS
# if: matrix.blas == 'ON'
# run: |
# C:/msys64/usr/bin/wget.exe -qO blas.zip ${{ matrix.obzip }}
# 7z x blas.zip -oblas -y
# copy blas/include/cblas.h .
# copy blas/include/openblas_config.h .
# echo "blasdir=$env:GITHUB_WORKSPACE/blas" >> $env:GITHUB_ENV
#
# - name: Configure
# run: >
# cmake -S . -B ./build -A ${{ matrix.arch }}
# -DCMAKE_BUILD_TYPE=${{ matrix.build }}
# -DLLAMA_SUPPORT_OPENBLAS=${{ matrix.blas }}
# -DCMAKE_LIBRARY_PATH="$env:blasdir/lib"
#
# - name: Build
# run: |
# cd ./build
# msbuild ALL_BUILD.vcxproj -t:build -p:configuration=${{ matrix.build }} -p:platform=${{ matrix.arch }}
#
# - name: Copy libopenblas.dll
# if: matrix.blas == 'ON'
# run: copy "$env:blasdir/bin/libopenblas.dll" build/bin/${{ matrix.build }}
#
# - name: Upload binaries
# if: matrix.blas == 'ON'
# uses: actions/upload-artifact@v1
# with:
# name: llama-blas-bin-${{ matrix.arch }}
# path: build/bin/${{ matrix.build }}
#
# emscripten:
# runs-on: ubuntu-latest
#
# strategy:
# matrix:
# build: [Release]
#
# steps:
# - name: Clone
# uses: actions/checkout@v3
#
# - name: Dependencies
# run: |
# wget -q https://github.com/emscripten-core/emsdk/archive/master.tar.gz
# tar -xvf master.tar.gz
# emsdk-master/emsdk update
# emsdk-master/emsdk install latest
# emsdk-master/emsdk activate latest
#
# - name: Configure
# run: echo "tmp"
#
# - name: Build
# run: |
# pushd emsdk-master
# source ./emsdk_env.sh
# popd
# emcmake cmake . -DCMAKE_BUILD_TYPE=${{ matrix.build }}
# make
workspaceFile: 'llama.xcworkspace'
xcArchiveName: 'llama'
scheme: 'llama'
sdk: '["iphoneos", "iphonesimulator", "macosx", "xros", "xrsimulator"]'
version: ${{ inputs.version }}
configuration: 'Release'
runsonlabels: '["macOS", "self-hosted"]'

36
.github/workflows/code-coverage.yml vendored
View file

@ -1,36 +0,0 @@
name: Code Coverage
on: [push, pull_request]
env:
GGML_NLOOP: 3
GGML_N_THREADS: 1
jobs:
run:
runs-on: ubuntu-20.04
steps:
- name: Checkout
uses: actions/checkout@v3
- name: Dependencies
run: |
sudo apt-get update
sudo apt-get install build-essential gcc-8 lcov
- name: Build
run: CC=gcc-8 make -j LLAMA_CODE_COVERAGE=1 tests
- name: Run tests
run: CC=gcc-8 make test
- name: Generate coverage report
run: |
make coverage
make lcov-report
- name: Upload coverage to Codecov
uses: codecov/codecov-action@v3
env:
CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
with:
files: lcov-report/coverage.info

107
.github/workflows/docker.yml vendored
View file

@ -1,107 +0,0 @@
# This workflow uses actions that are not certified by GitHub.
# They are provided by a third-party and are governed by
# separate terms of service, privacy policy, and support
# documentation.
# GitHub recommends pinning actions to a commit SHA.
# To get a newer version, you will need to update the SHA.
# You can also reference a tag or branch, but the action may change without warning.
name: Publish Docker image
on:
pull_request:
push:
branches:
- master
jobs:
push_to_registry:
name: Push Docker image to Docker Hub
if: github.event.pull_request.draft == false
runs-on: ubuntu-latest
env:
COMMIT_SHA: ${{ github.sha }}
strategy:
matrix:
config:
- { tag: "light", dockerfile: ".devops/main.Dockerfile", platforms: "linux/amd64,linux/arm64" }
- { tag: "full", dockerfile: ".devops/full.Dockerfile", platforms: "linux/amd64,linux/arm64" }
- { tag: "server", dockerfile: ".devops/server.Dockerfile", platforms: "linux/amd64,linux/arm64" }
# NOTE(canardletter): The CUDA builds on arm64 are very slow, so I
# have disabled them for now until the reason why
# is understood.
- { tag: "light-cuda", dockerfile: ".devops/main-cuda.Dockerfile", platforms: "linux/amd64" }
- { tag: "full-cuda", dockerfile: ".devops/full-cuda.Dockerfile", platforms: "linux/amd64" }
- { tag: "server-cuda", dockerfile: ".devops/server-cuda.Dockerfile", platforms: "linux/amd64" }
- { tag: "light-rocm", dockerfile: ".devops/main-rocm.Dockerfile", platforms: "linux/amd64,linux/arm64" }
- { tag: "full-rocm", dockerfile: ".devops/full-rocm.Dockerfile", platforms: "linux/amd64,linux/arm64" }
- { tag: "server-rocm", dockerfile: ".devops/server-rocm.Dockerfile", platforms: "linux/amd64,linux/arm64" }
- { tag: "light-intel", dockerfile: ".devops/main-intel.Dockerfile", platforms: "linux/amd64" }
- { tag: "server-intel", dockerfile: ".devops/server-intel.Dockerfile", platforms: "linux/amd64" }
steps:
- name: Check out the repo
uses: actions/checkout@v3
- name: Set up QEMU
uses: docker/setup-qemu-action@v2
- name: Set up Docker Buildx
uses: docker/setup-buildx-action@v2
- name: Log in to Docker Hub
uses: docker/login-action@v2
with:
registry: ghcr.io
username: ${{ github.repository_owner }}
password: ${{ secrets.GITHUB_TOKEN }}
# https://github.com/jlumbroso/free-disk-space/tree/54081f138730dfa15788a46383842cd2f914a1be#example
- name: Free Disk Space (Ubuntu)
uses: jlumbroso/free-disk-space@main
with:
# this might remove tools that are actually needed,
# if set to "true" but frees about 6 GB
tool-cache: false
# all of these default to true, but feel free to set to
# "false" if necessary for your workflow
android: true
dotnet: true
haskell: true
large-packages: true
docker-images: true
swap-storage: true
- name: Determine tag name
id: tag
shell: bash
run: |
BUILD_NUMBER="$(git rev-list --count HEAD)"
SHORT_HASH="$(git rev-parse --short=7 HEAD)"
if [[ "${{ env.BRANCH_NAME }}" == "master" ]]; then
echo "name=b${BUILD_NUMBER}" >> $GITHUB_OUTPUT
else
SAFE_NAME=$(echo "${{ env.BRANCH_NAME }}" | tr '/' '-')
echo "name=${SAFE_NAME}-b${BUILD_NUMBER}-${SHORT_HASH}" >> $GITHUB_OUTPUT
fi
- name: Build and push Docker image (versioned)
if: github.event_name == 'push'
uses: docker/build-push-action@v4
with:
context: .
push: true
platforms: ${{ matrix.config.platforms }}
tags: "ghcr.io/${{ github.repository_owner }}/llama.cpp:${{ matrix.config.tag }}-${{ env.COMMIT_SHA }}"
file: ${{ matrix.config.dockerfile }}
- name: Build and push Docker image (tagged)
uses: docker/build-push-action@v4
with:
context: .
push: ${{ github.event_name == 'push' }}
platforms: ${{ matrix.config.platforms }}
tags: "ghcr.io/${{ github.repository_owner }}/llama.cpp:${{ matrix.config.tag }},ghcr.io/${{ github.repository_owner }}/llama.cpp:${{ matrix.config.tag }}-${{ steps.tag.outputs.name }}"
file: ${{ matrix.config.dockerfile }}

23
.github/workflows/editorconfig.yml vendored
View file

@ -1,23 +0,0 @@
name: EditorConfig Checker
on:
workflow_dispatch: # allows manual triggering
inputs:
create_release:
description: 'Create new release'
required: true
type: boolean
push:
branches:
- master
pull_request:
branches:
- master
jobs:
editorconfig:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v3
- uses: editorconfig-checker/action-editorconfig-checker@main
- run: editorconfig-checker

44
.github/workflows/gguf-publish.yml vendored
View file

@ -1,44 +0,0 @@
# This workflow will upload a Python Package using Twine when a GGUF release is created
# For more information see: https://help.github.com/en/actions/language-and-framework-guides/using-python-with-github-actions#publishing-to-package-registries
# See `gguf-py/README.md` for how to make a release.
# This workflow uses actions that are not certified by GitHub.
# They are provided by a third-party and are governed by
# separate terms of service, privacy policy, and support
# documentation.
name: Upload Python Package
on:
workflow_dispatch:
push:
# Pattern matched against refs/tags
tags:
- 'gguf-v*' # Push events to every version tag
jobs:
deploy:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v3
- name: Set up Python
uses: actions/setup-python@v2
with:
python-version: '3.9.x'
- name: Install dependencies
run: |
cd gguf-py
python -m pip install poetry
poetry install
- name: Build package
run: cd gguf-py && poetry build
- name: Publish package
uses: pypa/gh-action-pypi-publish@release/v1
with:
password: ${{ secrets.PYPI_API_TOKEN }}
packages-dir: gguf-py/dist

61
.github/workflows/nix-ci-aarch64.yml vendored
View file

@ -1,61 +0,0 @@
name: Nix aarch64 builds
on:
workflow_dispatch: # allows manual triggering
schedule:
# Rebuild daily rather than on every push because QEMU is expensive (e.g.
# 1.5h instead of minutes with the cold cache).
#
# randint(0, 59), randint(0, 23)
- cron: '26 12 * * *'
# But also rebuild if we touched any of the Nix expressions:
push:
branches:
- master
paths: ['**/*.nix', 'flake.lock']
pull_request:
types: [opened, synchronize, reopened]
paths: ['**/*.nix', 'flake.lock']
jobs:
nix-build-aarch64:
runs-on: ubuntu-latest
steps:
- name: Checkout repository
uses: actions/checkout@v4
- name: Install QEMU
# Copy-paste from https://github.com/orgs/community/discussions/8305#discussioncomment-5888654
run: |
sudo apt-get update
sudo apt-get install -y qemu-user-static qemu-system-aarch64
sudo usermod -a -G kvm $USER
- name: Install Nix
uses: DeterminateSystems/nix-installer-action@v9
with:
github-token: ${{ secrets.GITHUB_TOKEN }}
extra-conf: |
extra-platforms = aarch64-linux
extra-system-features = nixos-test kvm
extra-substituters = https://llama-cpp.cachix.org https://cuda-maintainers.cachix.org
extra-trusted-public-keys = llama-cpp.cachix.org-1:H75X+w83wUKTIPSO1KWy9ADUrzThyGs8P5tmAbkWhQc= cuda-maintainers.cachix.org-1:0dq3bujKpuEPMCX6U4WylrUDZ9JyUG0VpVZa7CNfq5E=
- uses: DeterminateSystems/magic-nix-cache-action@v2
with:
upstream-cache: https://${{ matrix.cachixName }}.cachix.org
- name: Set-up cachix to push the results to
uses: cachix/cachix-action@v13
with:
authToken: '${{ secrets.CACHIX_AUTH_TOKEN }}'
name: llama-cpp
- name: Show all output paths
run: >
nix run github:nix-community/nix-eval-jobs
-- --gc-roots-dir gcroot
--flake
".#packages.aarch64-linux"
- name: Build
run: >
nix run github:Mic92/nix-fast-build
-- --skip-cached --no-nom
--systems aarch64-linux
--flake
".#checks.aarch64-linux"

68
.github/workflows/nix-ci.yml vendored
View file

@ -1,68 +0,0 @@
name: Nix CI
on:
workflow_dispatch: # allows manual triggering
push:
branches:
- master
pull_request:
types: [opened, synchronize, reopened]
jobs:
nix-eval:
strategy:
fail-fast: false
matrix:
os: [ ubuntu-latest, macos-latest ]
runs-on: ${{ matrix.os }}
steps:
- name: Checkout repository
uses: actions/checkout@v4
- name: Install Nix
uses: DeterminateSystems/nix-installer-action@v9
with:
github-token: ${{ secrets.GITHUB_TOKEN }}
extra-conf: |
extra-substituters = https://llama-cpp.cachix.org https://cuda-maintainers.cachix.org
extra-trusted-public-keys = llama-cpp.cachix.org-1:H75X+w83wUKTIPSO1KWy9ADUrzThyGs8P5tmAbkWhQc= cuda-maintainers.cachix.org-1:0dq3bujKpuEPMCX6U4WylrUDZ9JyUG0VpVZa7CNfq5E=
- uses: DeterminateSystems/magic-nix-cache-action@v2
with:
upstream-cache: https://${{ matrix.cachixName }}.cachix.org
- name: List all flake outputs
run: nix flake show --all-systems
- name: Show all output paths
run: >
nix run github:nix-community/nix-eval-jobs
-- --gc-roots-dir gcroot
--flake
".#packages.$(nix eval --raw --impure --expr builtins.currentSystem)"
nix-build:
strategy:
fail-fast: false
matrix:
os: [ ubuntu-latest, macos-latest ]
runs-on: ${{ matrix.os }}
steps:
- name: Checkout repository
uses: actions/checkout@v4
- name: Install Nix
uses: DeterminateSystems/nix-installer-action@v9
with:
github-token: ${{ secrets.GITHUB_TOKEN }}
extra-conf: |
extra-substituters = https://llama-cpp.cachix.org https://cuda-maintainers.cachix.org
extra-trusted-public-keys = llama-cpp.cachix.org-1:H75X+w83wUKTIPSO1KWy9ADUrzThyGs8P5tmAbkWhQc= cuda-maintainers.cachix.org-1:0dq3bujKpuEPMCX6U4WylrUDZ9JyUG0VpVZa7CNfq5E=
- uses: DeterminateSystems/magic-nix-cache-action@v2
with:
upstream-cache: https://${{ matrix.cachixName }}.cachix.org
- name: Set-up cachix to push the results to
uses: cachix/cachix-action@v13
with:
authToken: '${{ secrets.CACHIX_AUTH_TOKEN }}'
name: llama-cpp
- name: Build
run: >
nix run github:Mic92/nix-fast-build
-- --skip-cached --no-nom
--flake
".#checks.$(nix eval --raw --impure --expr builtins.currentSystem)"

22
.github/workflows/nix-flake-update.yml vendored
View file

@ -1,22 +0,0 @@
name: update-flake-lock
on:
workflow_dispatch:
schedule:
- cron: '0 0 * * 0' # runs weekly on Sunday at 00:00
jobs:
lockfile:
runs-on: ubuntu-latest
steps:
- name: Checkout repository
uses: actions/checkout@v4
- name: Install Nix
uses: DeterminateSystems/nix-installer-action@main
- name: Update flake.lock
uses: DeterminateSystems/update-flake-lock@main
with:
pr-title: "nix: update flake.lock"
pr-labels: |
nix
pr-reviewers: philiptaron,SomeoneSerge
token: ${{ secrets.FLAKE_TOKEN }}

36
.github/workflows/nix-publish-flake.yml vendored
View file

@ -1,36 +0,0 @@
# Make the flake discoverable on https://flakestry.dev and https://flakehub.com/flakes
name: "Publish a flake to flakestry & flakehub"
on:
push:
tags:
- "*"
workflow_dispatch:
inputs:
tag:
description: "The existing tag to publish"
type: "string"
required: true
jobs:
flakestry-publish:
runs-on: ubuntu-latest
permissions:
id-token: "write"
contents: "read"
steps:
- uses: flakestry/flakestry-publish@main
with:
version: "${{ inputs.tag || github.ref_name }}"
flakehub-publish:
runs-on: "ubuntu-latest"
permissions:
id-token: "write"
contents: "read"
steps:
- uses: "actions/checkout@v4"
with:
ref: "${{ (inputs.tag != null) && format('refs/tags/{0}', inputs.tag) || '' }}"
- uses: "DeterminateSystems/nix-installer-action@main"
- uses: "DeterminateSystems/flakehub-push@main"
with:
visibility: "public"
tag: "${{ inputs.tag }}"

29
.github/workflows/python-check-requirements.yml vendored
View file

@ -1,29 +0,0 @@
name: Python check requirements.txt
on:
push:
paths:
- 'scripts/check-requirements.sh'
- 'convert*.py'
- 'requirements.txt'
- 'requirements/*.txt'
pull_request:
paths:
- 'scripts/check-requirements.sh'
- 'convert*.py'
- 'requirements.txt'
- 'requirements/*.txt'
jobs:
python-check-requirements:
runs-on: ubuntu-latest
name: check-requirements
steps:
- name: Check out source repository
uses: actions/checkout@v3
- name: Set up Python environment
uses: actions/setup-python@v4
with:
python-version: "3.11"
- name: Run check-requirements.sh script
run: bash scripts/check-requirements.sh nocleanup

20
.github/workflows/python-lint.yml vendored
View file

@ -1,20 +0,0 @@
name: flake8 Lint
on: [push, pull_request]
jobs:
flake8-lint:
runs-on: ubuntu-latest
name: Lint
steps:
- name: Check out source repository
uses: actions/checkout@v3
- name: Set up Python environment
uses: actions/setup-python@v4
with:
python-version: "3.11"
- name: flake8 Lint
uses: py-actions/flake8@v2
with:
ignore: "E203,E211,E221,E225,E231,E241,E251,E261,E266,E501,E701,E704,W503"
exclude: "examples/*,examples/*/**,*/**/__init__.py"

32
.github/workflows/release.yml vendored Normal file
View file

@ -0,0 +1,32 @@
#
# This source file is part of the Stanford Biodesign Digital Health Group open-source organization
#
# SPDX-FileCopyrightText: 2022 Stanford University and the project authors (see CONTRIBUTORS.md)
#
# SPDX-License-Identifier: MIT
#
name: Create XCFramework and Release
on:
workflow_dispatch:
inputs:
version:
description: 'The version number of the framework embedded in the XCArchives. This version number is also used as the release tag.'
type: string
required: true
jobs:
create-xcframework-and-release-workflow:
uses: StanfordBDHG/.github/.github/workflows/xcframework.yml@v2
with:
workspaceFile: llama.xcworkspace
xcFrameworkName: llama
scheme: llama
sdk: '["iphoneos", "iphonesimulator", "macosx", "xros", "xrsimulator"]'
version: ${{ inputs.version }}
configuration: Release
runsonlabels: '["macOS", "self-hosted"]'
user: PaulsAutomationBot
secrets:
access-token: ${{ secrets.PERSONAL_ACCESS_TOKEN }}

83
.github/workflows/server.yml vendored
View file

@ -1,83 +0,0 @@
# Server build and tests
name: Server
on:
workflow_dispatch: # allows manual triggering
push:
branches:
- master
paths: ['.github/workflows/server.yml', '**/CMakeLists.txt', '**/Makefile', '**/*.h', '**/*.hpp', '**/*.c', '**/*.cpp', '**/*.cu', '**/*.swift', '**/*.m', 'examples/server/tests/**.*']
pull_request:
types: [opened, synchronize, reopened]
paths: ['.github/workflows/server.yml', '**/CMakeLists.txt', '**/Makefile', '**/*.h', '**/*.hpp', '**/*.c', '**/*.cpp', '**/*.cu', '**/*.swift', '**/*.m', 'examples/server/tests/**.*']
jobs:
server:
runs-on: ubuntu-latest
strategy:
matrix:
sanitizer: [ADDRESS, THREAD, UNDEFINED]
build_type: [Debug, Release]
include:
- build_type: Release
sanitizer: ""
exclude:
- build_type: Release
sanitizer: ADDRESS
- build_type: Release
sanitizer: THREAD
- build_type: Release
sanitizer: UNDEFINED
container:
image: ubuntu:latest
ports:
- 8888
options: --cpus 4
steps:
- name: Clone
id: checkout
uses: actions/checkout@v3
- name: Dependencies
id: depends
run: |
apt-get update
apt-get -y install \
build-essential \
git \
cmake \
python3-pip \
wget \
psmisc
- name: Build
id: cmake_build
run: |
mkdir build
cd build
cmake .. \
-DLLAMA_NATIVE=OFF \
-DLLAMA_BUILD_SERVER=ON \
-DCMAKE_BUILD_TYPE=${{ matrix.build_type }} \
-DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON ;
cmake --build . --config ${{ matrix.build_type }} -j $(nproc) --target server
- name: Tests dependencies
id: test_dependencies
run: |
pip install -r examples/server/tests/requirements.txt
- name: Download models
id: download_models
run: |
cd examples/server/tests
../../../scripts/hf.sh --repo ggml-org/models --file tinyllamas/stories260K.gguf
- name: Tests
id: server_integration_test
run: |
cd examples/server/tests
PORT=8888 ./tests.sh

20
.github/workflows/tidy-post.yml vendored
View file

@ -1,20 +0,0 @@
name: clang-tidy review post comments
on:
workflow_dispatch:
workflows: ["clang-tidy-review"]
types:
- completed
jobs:
build:
runs-on: ubuntu-latest
steps:
- uses: ZedThree/clang-tidy-review/post@v0.13.0
# lgtm_comment_body, max_comments, and annotations need to be set on the posting workflow in a split setup
with:
# adjust options as necessary
lgtm_comment_body: ''
annotations: false
max_comments: 25

23
.github/workflows/tidy-review.yml vendored
View file

@ -1,23 +0,0 @@
name: clang-tidy-review
on:
pull_request:
branches:
- master
jobs:
clang-tidy-review:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v3
- uses: ZedThree/clang-tidy-review@v0.13.0
id: review
with:
lgtm_comment_body: ''
build_dir: build
cmake_command: cmake . -B build -DCMAKE_EXPORT_COMPILE_COMMANDS=on
split_workflow: true
- uses: ZedThree/clang-tidy-review/upload@v0.13.0

25
.github/workflows/zig-build.yml vendored
View file

@ -1,25 +0,0 @@
name: Zig CI
on:
pull_request:
push:
branches:
- master
jobs:
build:
strategy:
fail-fast: false
matrix:
runs-on: [ubuntu-latest, macos-latest, windows-latest]
runs-on: ${{ matrix.runs-on }}
steps:
- uses: actions/checkout@v3
with:
submodules: recursive
fetch-depth: 0
- uses: goto-bus-stop/setup-zig@v2
with:
version: 0.11.0
- name: Build Summary
run: zig build --summary all -freference-trace

23
.gitignore vendored
View file

@ -11,7 +11,6 @@
*.gcda
*.dot
*.bat
*.metallib
.DS_Store
.build/
.cache/
@ -92,3 +91,25 @@ examples/jeopardy/results.txt
poetry.lock
poetry.toml
nppBackup
.swiftpm
.build
*.pyc
.DS_Store
# Xcode
build/
*.pbxuser
!default.pbxuser
*.mode1v3
!default.mode1v3
*.mode2v3
!default.mode2v3
*.perspectivev3
!default.perspectivev3
xcuserdata
*.xccheckout
*.moved-aside
DerivedData
*.hmap
*.ipa
*.xcuserstate

67
Package.swift
View file

@ -1,60 +1,33 @@
// swift-tools-version:5.5
// swift-tools-version:5.9
//
// This source file is part of the TemplatePackage open source project
//
// SPDX-FileCopyrightText: 2022 Stanford University and the project authors (see CONTRIBUTORS.md)
//
// SPDX-License-Identifier: MIT
//
import PackageDescription
let package = Package(
name: "llama",
platforms: [
.macOS(.v12),
.iOS(.v14),
.watchOS(.v4),
.tvOS(.v14)
.iOS(.v14)
],
products: [
.library(name: "llama", targets: ["llama"]),
],
targets: [
.target(
.library(
name: "llama",
path: ".",
exclude: [
"cmake",
"examples",
"scripts",
"models",
"tests",
"CMakeLists.txt",
"ggml-cuda.cu",
"ggml-cuda.h",
"Makefile"
],
sources: [
"ggml.c",
"llama.cpp",
"ggml-alloc.c",
"ggml-backend.c",
"ggml-quants.c",
"ggml-metal.m",
],
resources: [
.process("ggml-metal.metal")
],
publicHeadersPath: "spm-headers",
cSettings: [
.unsafeFlags(["-Wno-shorten-64-to-32", "-O3", "-DNDEBUG"]),
.define("GGML_USE_ACCELERATE"),
.unsafeFlags(["-fno-objc-arc"]),
.define("GGML_USE_METAL"),
// NOTE: NEW_LAPACK will required iOS version 16.4+
// We should consider add this in the future when we drop support for iOS 14
// (ref: ref: https://developer.apple.com/documentation/accelerate/1513264-cblas_sgemm?language=objc)
// .define("ACCELERATE_NEW_LAPACK"),
// .define("ACCELERATE_LAPACK_ILP64")
],
linkerSettings: [
.linkedFramework("Accelerate")
targets: [
"llama"
]
)
],
cxxLanguageStandard: .cxx11
targets: [
.binaryTarget(
name: "llama",
path: "./llama.xcframework"
)
]
)

1104
README.md
View file

File diff suppressed because it is too large Load diff

2
kompute

@ -1 +1 @@
Subproject commit 4565194ed7c32d1d2efa32ceab4d3c6cae006306
Subproject commit d1e3b0953cf66acc94b2e29693e221427b2c1f3f

21
llama-module.modulemap Normal file
View file

@ -0,0 +1,21 @@
//
// This source file is part of the Stanford Biodesign Digital Health Group open-source organization
//
// SPDX-FileCopyrightText: 2022 Stanford University and the project authors (see CONTRIBUTORS.md)
//
// SPDX-License-Identifier: MIT
//
framework module llama {
umbrella header "llama.h"
header "common.h"
header "sampling.h"
header "log.h"
header "grammar-parser.h"
header "tokenize.h"
header "vector.h"
export *
module * { export * }
}

44
llama.xcframework/Info.plist Normal file
View file

@ -0,0 +1,44 @@
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
<key>AvailableLibraries</key>
<array>
<dict>
<key>BinaryPath</key>
<string>llama.framework/llama</string>
<key>LibraryIdentifier</key>
<string>ios-arm64</string>
<key>LibraryPath</key>
<string>llama.framework</string>
<key>SupportedArchitectures</key>
<array>
<string>arm64</string>
</array>
<key>SupportedPlatform</key>
<string>ios</string>
</dict>
<dict>
<key>BinaryPath</key>
<string>llama.framework/llama</string>
<key>LibraryIdentifier</key>
<string>ios-arm64_x86_64-simulator</string>
<key>LibraryPath</key>
<string>llama.framework</string>
<key>SupportedArchitectures</key>
<array>
<string>arm64</string>
<string>x86_64</string>
</array>
<key>SupportedPlatform</key>
<string>ios</string>
<key>SupportedPlatformVariant</key>
<string>simulator</string>
</dict>
</array>
<key>CFBundlePackageType</key>
<string>XFWK</string>
<key>XCFrameworkFormatVersion</key>
<string>1.0</string>
</dict>
</plist>

270
llama.xcframework/ios-arm64/llama.framework/Headers/common.h Normal file
View file

@ -0,0 +1,270 @@
// Various helper functions and utilities
#pragma once
#include "llama.h"
#include "sampling.h"
#define LOG_NO_FILE_LINE_FUNCTION
#include "log.h"
#include <cmath>
#include <string>
#include <vector>
#include <random>
#include <thread>
#include <unordered_map>
#include <tuple>
#ifdef _WIN32
#define DIRECTORY_SEPARATOR '\\'
#else
#define DIRECTORY_SEPARATOR '/'
#endif // _WIN32
#define die(msg) do { fputs("error: " msg "\n", stderr); exit(1); } while (0)
#define die_fmt(fmt, ...) do { fprintf(stderr, "error: " fmt "\n", __VA_ARGS__); exit(1); } while (0)
#define print_build_info() do { \
fprintf(stderr, "%s: build = %d (%s)\n", __func__, LLAMA_BUILD_NUMBER, LLAMA_COMMIT); \
fprintf(stderr, "%s: built with %s for %s\n", __func__, LLAMA_COMPILER, LLAMA_BUILD_TARGET); \
} while(0)
// build info
// If macro is not defined (passed via Xcode), set the corresponding variable as defined via external variable.
#ifndef LLAMA_BUILD_NUMBER
extern int LLAMA_BUILD_NUMBER;
#endif
#ifndef LLAMA_COMMIT
extern char const *LLAMA_COMMIT;
#endif
#ifndef LLAMA_COMPILER
extern char const *LLAMA_COMPILER;
#endif
#ifndef LLAMA_BUILD_TARGET
extern char const *LLAMA_BUILD_TARGET;
#endif
//
// CLI argument parsing
//
int32_t get_num_physical_cores();
struct gpt_params {
uint32_t seed = -1; // RNG seed
int32_t n_threads = get_num_physical_cores();
int32_t n_threads_draft = -1;
int32_t n_threads_batch = -1; // number of threads to use for batch processing (-1 = use n_threads)
int32_t n_threads_batch_draft = -1;
int32_t n_predict = -1; // new tokens to predict
int32_t n_ctx = 512; // context size
int32_t n_batch = 512; // batch size for prompt processing (must be >=32 to use BLAS)
int32_t n_keep = 0; // number of tokens to keep from initial prompt
int32_t n_draft = 8; // number of tokens to draft during speculative decoding
int32_t n_chunks = -1; // max number of chunks to process (-1 = unlimited)
int32_t n_parallel = 1; // number of parallel sequences to decode
int32_t n_sequences = 1; // number of sequences to decode
float p_accept = 0.5f; // speculative decoding accept probability
float p_split = 0.1f; // speculative decoding split probability
int32_t n_gpu_layers = -1; // number of layers to store in VRAM (-1 - use default)
int32_t n_gpu_layers_draft = -1; // number of layers to store in VRAM for the draft model (-1 - use default)
llama_split_mode split_mode = LLAMA_SPLIT_LAYER; // how to split the model across GPUs
int32_t main_gpu = 0; // the GPU that is used for scratch and small tensors
float tensor_split[LLAMA_MAX_DEVICES] = {0}; // how split tensors should be distributed across GPUs
int32_t n_beams = 0; // if non-zero then use beam search of given width.
int32_t grp_attn_n = 1; // group-attention factor
int32_t grp_attn_w = 512; // group-attention width
int32_t n_print = -1; // print token count every n tokens (-1 = disabled)
float rope_freq_base = 0.0f; // RoPE base frequency
float rope_freq_scale = 0.0f; // RoPE frequency scaling factor
float yarn_ext_factor = -1.0f; // YaRN extrapolation mix factor
float yarn_attn_factor = 1.0f; // YaRN magnitude scaling factor
float yarn_beta_fast = 32.0f; // YaRN low correction dim
float yarn_beta_slow = 1.0f; // YaRN high correction dim
int32_t yarn_orig_ctx = 0; // YaRN original context length
int8_t rope_scaling_type = LLAMA_ROPE_SCALING_UNSPECIFIED; // TODO: better to be int32_t for alignment
// pinging @cebtenzzre
// // sampling parameters
struct llama_sampling_params sparams;
std::string model = "models/7B/ggml-model-f16.gguf"; // model path
std::string model_draft = ""; // draft model for speculative decoding
std::string model_alias = "unknown"; // model alias
std::string prompt = "";
std::string prompt_file = ""; // store the external prompt file name
std::string path_prompt_cache = ""; // path to file for saving/loading prompt eval state
std::string input_prefix = ""; // string to prefix user inputs with
std::string input_suffix = ""; // string to suffix user inputs with
std::vector<std::string> antiprompt; // string upon seeing which more user input is prompted
std::string logdir = ""; // directory in which to save YAML log files
std::string logits_file = ""; // file for saving *all* logits
std::vector<llama_model_kv_override> kv_overrides;
// TODO: avoid tuple, use struct
std::vector<std::tuple<std::string, float>> lora_adapter; // lora adapter path with user defined scale
std::string lora_base = ""; // base model path for the lora adapter
int ppl_stride = 0; // stride for perplexity calculations. If left at 0, the pre-existing approach will be used.
int ppl_output_type = 0; // = 0 -> ppl output is as usual, = 1 -> ppl output is num_tokens, ppl, one per line
// (which is more convenient to use for plotting)
//
bool hellaswag = false; // compute HellaSwag score over random tasks from datafile supplied in prompt
size_t hellaswag_tasks = 400; // number of tasks to use when computing the HellaSwag score
bool winogrande = false; // compute Winogrande score over random tasks from datafile supplied in prompt
size_t winogrande_tasks= 0; // number of tasks to use when computing the Winogrande score. If 0, all tasks will be computed
bool multiple_choice = false; // compute TruthfulQA score over random tasks from datafile supplied in prompt
size_t multiple_choice_tasks = 0; // number of tasks to use when computing the TruthfulQA score. If 0, all tasks will be computed
bool kl_divergence = false; // compute KL-divergence
bool mul_mat_q = true; // if true, use mul_mat_q kernels instead of cuBLAS
bool random_prompt = false; // do not randomize prompt if none provided
bool use_color = false; // use color to distinguish generations and inputs
bool interactive = false; // interactive mode
bool chatml = false; // chatml mode (used for models trained on chatml syntax)
bool prompt_cache_all = false; // save user input and generations to prompt cache
bool prompt_cache_ro = false; // open the prompt cache read-only and do not update it
bool embedding = false; // get only sentence embedding
bool escape = false; // escape "\n", "\r", "\t", "\'", "\"", and "\\"
bool interactive_first = false; // wait for user input immediately
bool multiline_input = false; // reverse the usage of `\`
bool simple_io = false; // improves compatibility with subprocesses and limited consoles
bool cont_batching = false; // insert new sequences for decoding on-the-fly
bool input_prefix_bos = false; // prefix BOS to user inputs, preceding input_prefix
bool ignore_eos = false; // ignore generated EOS tokens
bool instruct = false; // instruction mode (used for Alpaca models)
bool logits_all = false; // return logits for all tokens in the batch
bool use_mmap = true; // use mmap for faster loads
bool use_mlock = false; // use mlock to keep model in memory
bool numa = false; // attempt optimizations that help on some NUMA systems
bool verbose_prompt = false; // print prompt tokens before generation
bool display_prompt = true; // print prompt before generation
bool infill = false; // use infill mode
bool dump_kv_cache = false; // dump the KV cache contents for debugging purposes
bool no_kv_offload = false; // disable KV offloading
std::string cache_type_k = "f16"; // KV cache data type for the K
std::string cache_type_v = "f16"; // KV cache data type for the V
// multimodal models (see examples/llava)
std::string mmproj = ""; // path to multimodal projector
std::string image = ""; // path to an image file
};
bool gpt_params_parse_ex(int argc, char ** argv, gpt_params & params);
bool gpt_params_parse(int argc, char ** argv, gpt_params & params);
void gpt_print_usage(int argc, char ** argv, const gpt_params & params);
std::string get_system_info(const gpt_params & params);
std::string gpt_random_prompt(std::mt19937 & rng);
void process_escapes(std::string& input);
//
// String parsing
//
std::string parse_samplers_input(std::string input);
//
// Model utils
//
// TODO: avoid tuplue, use struct
std::tuple<struct llama_model *, struct llama_context *> llama_init_from_gpt_params(gpt_params & params);
struct llama_model_params llama_model_params_from_gpt_params (const gpt_params & params);
struct llama_context_params llama_context_params_from_gpt_params(const gpt_params & params);
// Batch utils
void llama_batch_clear(struct llama_batch & batch);
void llama_batch_add(
struct llama_batch & batch,
llama_token id,
llama_pos pos,
const std::vector<llama_seq_id> & seq_ids,
bool logits);
//
// Vocab utils
//
// tokenizes a string into a vector of tokens
// should work similar to Python's `tokenizer.encode`
std::vector<llama_token> llama_tokenize(
const struct llama_context * ctx,
const std::string & text,
bool add_bos,
bool special = false);
std::vector<llama_token> llama_tokenize(
const struct llama_model * model,
const std::string & text,
bool add_bos,
bool special = false);
// tokenizes a token into a piece
// should work similar to Python's `tokenizer.id_to_piece`
std::string llama_token_to_piece(
const struct llama_context * ctx,
llama_token token);
// TODO: these should be moved in llama.h C-style API under single `llama_detokenize` function
// that takes into account the tokenizer type and decides how to handle the leading space
//
// detokenizes a vector of tokens into a string
// should work similar to Python's `tokenizer.decode`
// removes the leading space from the first non-BOS token
std::string llama_detokenize_spm(
llama_context * ctx,
const std::vector<llama_token> & tokens);
// detokenizes a vector of tokens into a string
// should work similar to Python's `tokenizer.decode`
std::string llama_detokenize_bpe(
llama_context * ctx,
const std::vector<llama_token> & tokens);
// Uses the value from the model metadata if possible, otherwise
// defaults to true when model type is SPM, otherwise false.
bool llama_should_add_bos_token(const llama_model * model);
//
// YAML utils
//
bool create_directory_with_parents(const std::string & path);
void dump_vector_float_yaml(FILE * stream, const char * prop_name, const std::vector<float> & data);
void dump_vector_int_yaml(FILE * stream, const char * prop_name, const std::vector<int> & data);
void dump_string_yaml_multiline(FILE * stream, const char * prop_name, const char * data);
std::string get_sortable_timestamp();
void dump_non_result_info_yaml(
FILE * stream, const gpt_params & params, const llama_context * lctx,
const std::string & timestamp, const std::vector<int> & prompt_tokens, const char * model_desc);
//
// KV cache utils
//
// Dump the KV cache view with the number of sequences per cell.
void dump_kv_cache_view(const llama_kv_cache_view & view, int row_size = 80);
// Dump the KV cache view showing individual sequences in each cell (long output).
void dump_kv_cache_view_seqs(const llama_kv_cache_view & view, int row_size = 40);

94
llama.xcframework/ios-arm64/llama.framework/Headers/ggml-alloc.h Normal file
View file

@ -0,0 +1,94 @@
#pragma once
#include "ggml.h"
#ifdef __cplusplus
extern "C" {
#endif
struct ggml_backend;
struct ggml_backend_buffer;
struct ggml_backend_buffer_type;
//
// Legacy API
//
typedef struct ggml_allocr * ggml_allocr_t;
// initialize allocator for use with CPU backend only
GGML_API ggml_allocr_t ggml_allocr_new(void * data, size_t size, size_t alignment);
GGML_API ggml_allocr_t ggml_allocr_new_measure(size_t alignment);
// initialize allocator for use with ggml-backend
GGML_API ggml_allocr_t ggml_allocr_new_from_buffer(struct ggml_backend_buffer * buffer);
GGML_API ggml_allocr_t ggml_allocr_new_from_backend(struct ggml_backend * backend, size_t size); // allocates an owned buffer
GGML_API ggml_allocr_t ggml_allocr_new_measure_from_backend(struct ggml_backend * backend);
GGML_API struct ggml_backend_buffer * ggml_allocr_get_buffer(ggml_allocr_t alloc);
// tell the allocator to parse nodes following the order described in the list
// you should call this if your graph are optimized to execute out-of-order
GGML_API void ggml_allocr_set_parse_seq(ggml_allocr_t alloc, const int * list, int n);
GGML_API void ggml_allocr_free (ggml_allocr_t alloc);
GGML_API bool ggml_allocr_is_measure (ggml_allocr_t alloc);
GGML_API void ggml_allocr_reset (ggml_allocr_t alloc);
GGML_API void ggml_allocr_alloc (ggml_allocr_t alloc, struct ggml_tensor * tensor);
GGML_API size_t ggml_allocr_max_size (ggml_allocr_t alloc);
GGML_API size_t ggml_allocr_alloc_graph(ggml_allocr_t alloc, struct ggml_cgraph * graph);
//
// ggml-backend v2 API
//
// Separate tensor and graph allocator objects
// This is necessary for multi-backend allocation because the graph allocator needs to use multiple tensor allocators
// The original API is kept as a wrapper around the new API
// Tensor allocator
typedef struct ggml_tallocr * ggml_tallocr_t;
GGML_API ggml_tallocr_t ggml_tallocr_new(void * data, size_t size, size_t alignment);
GGML_API ggml_tallocr_t ggml_tallocr_new_measure(size_t alignment);
GGML_API ggml_tallocr_t ggml_tallocr_new_from_buft(struct ggml_backend_buffer_type * buft, size_t size);
GGML_API ggml_tallocr_t ggml_tallocr_new_from_backend(struct ggml_backend * backend, size_t size); // allocates an owned buffer
GGML_API ggml_tallocr_t ggml_tallocr_new_from_buffer(struct ggml_backend_buffer * buffer);
GGML_API ggml_tallocr_t ggml_tallocr_new_measure_from_buft(struct ggml_backend_buffer_type * buft);
GGML_API ggml_tallocr_t ggml_tallocr_new_measure_from_backend(struct ggml_backend * backend);
GGML_API struct ggml_backend_buffer * ggml_tallocr_get_buffer(ggml_tallocr_t talloc);
GGML_API void ggml_tallocr_free (ggml_tallocr_t talloc);
GGML_API bool ggml_tallocr_is_measure (ggml_tallocr_t talloc);
GGML_API void ggml_tallocr_reset (ggml_tallocr_t talloc);
GGML_API void ggml_tallocr_alloc (ggml_tallocr_t talloc, struct ggml_tensor * tensor);
GGML_API size_t ggml_tallocr_max_size (ggml_tallocr_t talloc);
// Graph allocator
typedef struct ggml_gallocr * ggml_gallocr_t;
GGML_API ggml_gallocr_t ggml_gallocr_new(void);
GGML_API void ggml_gallocr_free(ggml_gallocr_t galloc);
GGML_API void ggml_gallocr_set_parse_seq(ggml_gallocr_t galloc, const int * list, int n);
GGML_API size_t ggml_gallocr_alloc_graph(ggml_gallocr_t galloc, ggml_tallocr_t talloc, struct ggml_cgraph * graph);
// Allocate tensors from the allocators given by the hash table
GGML_API void ggml_gallocr_alloc_graph_n(
ggml_gallocr_t galloc,
struct ggml_cgraph * graph,
struct ggml_hash_set hash_set,
ggml_tallocr_t * hash_node_talloc);
// Utils
// Create a buffer and allocate all the tensors in a ggml_context
GGML_API struct ggml_backend_buffer * ggml_backend_alloc_ctx_tensors_from_buft(struct ggml_context * ctx, struct ggml_backend_buffer_type * buft);
GGML_API struct ggml_backend_buffer * ggml_backend_alloc_ctx_tensors(struct ggml_context * ctx, struct ggml_backend * backend);
#ifdef __cplusplus
}
#endif

209
llama.xcframework/ios-arm64/llama.framework/Headers/ggml-backend.h Normal file
View file

@ -0,0 +1,209 @@
#pragma once
#include "ggml.h"
#include "ggml-alloc.h"
#ifdef __cplusplus
extern "C" {
#endif
typedef struct ggml_backend_buffer_type * ggml_backend_buffer_type_t;
typedef struct ggml_backend_buffer * ggml_backend_buffer_t;
typedef struct ggml_backend * ggml_backend_t;
typedef void * ggml_backend_graph_plan_t;
//
// Backend buffer
//
// buffer type
GGML_API const char * ggml_backend_buft_name (ggml_backend_buffer_type_t buft);
GGML_API GGML_CALL ggml_backend_buffer_t ggml_backend_buft_alloc_buffer (ggml_backend_buffer_type_t buft, size_t size);
GGML_API size_t ggml_backend_buft_get_alignment (ggml_backend_buffer_type_t buft);
GGML_API GGML_CALL size_t ggml_backend_buft_get_alloc_size (ggml_backend_buffer_type_t buft, struct ggml_tensor * tensor);
GGML_API bool ggml_backend_buft_supports_backend(ggml_backend_buffer_type_t buft, ggml_backend_t backend);
GGML_API bool ggml_backend_buft_is_host (ggml_backend_buffer_type_t buft);
// buffer
enum ggml_backend_buffer_usage {
GGML_BACKEND_BUFFER_USAGE_ANY = 0,
GGML_BACKEND_BUFFER_USAGE_WEIGHTS = 1,
};
GGML_API const char * ggml_backend_buffer_name (ggml_backend_buffer_t buffer);
GGML_API void ggml_backend_buffer_free (ggml_backend_buffer_t buffer);
GGML_API void * ggml_backend_buffer_get_base (ggml_backend_buffer_t buffer);
GGML_API size_t ggml_backend_buffer_get_size (ggml_backend_buffer_t buffer);
GGML_API GGML_CALL void ggml_backend_buffer_init_tensor (ggml_backend_buffer_t buffer, struct ggml_tensor * tensor);
GGML_API size_t ggml_backend_buffer_get_alignment (ggml_backend_buffer_t buffer);
GGML_API size_t ggml_backend_buffer_get_alloc_size(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor);
GGML_API void ggml_backend_buffer_clear (ggml_backend_buffer_t buffer, uint8_t value);
GGML_API bool ggml_backend_buffer_is_host (ggml_backend_buffer_t buffer);
GGML_API void ggml_backend_buffer_set_usage (ggml_backend_buffer_t buffer, enum ggml_backend_buffer_usage usage);
GGML_API ggml_backend_buffer_type_t ggml_backend_buffer_get_type (ggml_backend_buffer_t buffer);
GGML_API void ggml_backend_buffer_reset (ggml_backend_buffer_t buffer);
//
// Backend
//
GGML_API const char * ggml_backend_name(ggml_backend_t backend);
GGML_API void ggml_backend_free(ggml_backend_t backend);
GGML_API ggml_backend_buffer_type_t ggml_backend_get_default_buffer_type(ggml_backend_t backend);
GGML_API ggml_backend_buffer_t ggml_backend_alloc_buffer(ggml_backend_t backend, size_t size);
GGML_API size_t ggml_backend_get_alignment(ggml_backend_t backend);
GGML_API void ggml_backend_tensor_set_async(ggml_backend_t backend, struct ggml_tensor * tensor, const void * data, size_t offset, size_t size);
GGML_API void ggml_backend_tensor_get_async(ggml_backend_t backend, const struct ggml_tensor * tensor, void * data, size_t offset, size_t size);
GGML_API GGML_CALL void ggml_backend_tensor_set( struct ggml_tensor * tensor, const void * data, size_t offset, size_t size);
GGML_API GGML_CALL void ggml_backend_tensor_get(const struct ggml_tensor * tensor, void * data, size_t offset, size_t size);
GGML_API void ggml_backend_synchronize(ggml_backend_t backend);
GGML_API ggml_backend_graph_plan_t ggml_backend_graph_plan_create (ggml_backend_t backend, struct ggml_cgraph * cgraph);
GGML_API void ggml_backend_graph_plan_free (ggml_backend_t backend, ggml_backend_graph_plan_t plan);
GGML_API void ggml_backend_graph_plan_compute(ggml_backend_t backend, ggml_backend_graph_plan_t plan);
GGML_API bool ggml_backend_graph_compute (ggml_backend_t backend, struct ggml_cgraph * cgraph);
GGML_API bool ggml_backend_supports_op (ggml_backend_t backend, const struct ggml_tensor * op);
// tensor copy between different backends
GGML_API void ggml_backend_tensor_copy(struct ggml_tensor * src, struct ggml_tensor * dst);
GGML_API void ggml_backend_tensor_copy_async(ggml_backend_t backend, struct ggml_tensor * src, struct ggml_tensor * dst); // automatic fallback to sync copy
//
// CPU backend
//
GGML_API ggml_backend_t ggml_backend_cpu_init(void);
GGML_API GGML_CALL bool ggml_backend_is_cpu (ggml_backend_t backend);
GGML_API void ggml_backend_cpu_set_n_threads(ggml_backend_t backend_cpu, int n_threads);
// Create a backend buffer from an existing pointer
GGML_API GGML_CALL ggml_backend_buffer_t ggml_backend_cpu_buffer_from_ptr(void * ptr, size_t size);
GGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_cpu_buffer_type(void);
#ifdef GGML_USE_CPU_HBM
GGML_API ggml_backend_buffer_type_t ggml_backend_cpu_hbm_buffer_type(void);
#endif
//
// Backend registry
//
// The backend registry is a registry of all the available backends, and allows initializing backends in a generic way
GGML_API size_t ggml_backend_reg_get_count(void);
GGML_API size_t ggml_backend_reg_find_by_name(const char * name);
GGML_API ggml_backend_t ggml_backend_reg_init_backend_from_str(const char * backend_str); // str is name[:params]
GGML_API const char * ggml_backend_reg_get_name(size_t i);
GGML_API ggml_backend_t ggml_backend_reg_init_backend(size_t i, const char * params); // params is backend-specific
GGML_API ggml_backend_buffer_type_t ggml_backend_reg_get_default_buffer_type(size_t i);
GGML_API ggml_backend_buffer_t ggml_backend_reg_alloc_buffer(size_t i, size_t size);
//
// Backend scheduler
//
// The backend scheduler allows for multiple backends to be used together
// Handles compute buffer allocation, assignment of tensors to backends, and copying of tensors between backends
// The backends are selected based on:
// - the backend that supports the operation
// - the location of the pre-allocated tensors (e.g. the weights)
/*
Example usage:
sched = ggml_backend_sched_new({backend_gpu, backend_gpu2, backend_cpu}, num_backends);
// sched is initialized with measure allocators and cannot be used until allocated with a measure graph
// initialize buffers from a measure graph
measure_graph = build_graph(sched); // use the allocr to allocate inputs as needed
// in build_graph:
build_graph(...) {
// allocating tensors in a specific backend (optional, recommended: pre-allocate inputs in a different buffer)
alloc_cpu = ggml_backend_sched_get_allocr(sched, backend_cpu);
ggml_allocr_alloc(alloc_cpu, tensor);
// manually assigning nodes to a backend (optional, shouldn't be needed in most cases)
struct ggml_tensor * node = ggml_mul_mat(ctx, ...);
ggml_backend_sched_set_node_backend(sched, node, backend_gpu);
}
// allocate backend buffers from measure graph
ggml_backend_sched_init_measure(sched, measure_graph);
// the scheduler is now ready to compute graphs
// compute
graph = build_graph(sched);
ggml_backend_sched_graph_compute(sched, graph);
*/
struct ggml_backend_sched;
typedef struct ggml_backend_sched * ggml_backend_sched_t;
// when ask == true, the scheduler wants to know if the user wants to observe this node
// this allows the scheduler to batch nodes together in order to evaluate them in a single call
//
// when ask == false, the scheduler is passing the node tensor to the user for observation
// if the user returns false, the scheduler will cancel the graph compute
//
typedef bool (*ggml_backend_sched_eval_callback)(struct ggml_tensor * t, bool ask, void * user_data);
// Initialize a backend scheduler
GGML_API ggml_backend_sched_t ggml_backend_sched_new(ggml_backend_t * backends, ggml_backend_buffer_type_t * bufts, int n_backends, size_t graph_size);
GGML_API void ggml_backend_sched_free(ggml_backend_sched_t sched);
// Initialize backend buffers from a measure graph
GGML_API void ggml_backend_sched_init_measure(ggml_backend_sched_t sched, struct ggml_cgraph * measure_graph);
// Get the number of splits of the last graph
GGML_API int ggml_backend_sched_get_n_splits(ggml_backend_sched_t sched);
GGML_API ggml_tallocr_t ggml_backend_sched_get_tallocr(ggml_backend_sched_t sched, ggml_backend_t backend);
GGML_API ggml_backend_buffer_t ggml_backend_sched_get_buffer (ggml_backend_sched_t sched, ggml_backend_t backend);
GGML_API void ggml_backend_sched_set_node_backend(ggml_backend_sched_t sched, struct ggml_tensor * node, ggml_backend_t backend);
GGML_API ggml_backend_t ggml_backend_sched_get_node_backend(ggml_backend_sched_t sched, struct ggml_tensor * node);
// Allocate and compute graph on the backend scheduler
GGML_API void ggml_backend_sched_graph_compute(ggml_backend_sched_t sched, struct ggml_cgraph * graph);
// Reset all assignments and allocators - must be called before using the sched allocators to allocate inputs
GGML_API void ggml_backend_sched_reset(ggml_backend_sched_t sched);
// Set a callback to be called for each resulting node during graph compute
GGML_API void ggml_backend_sched_set_eval_callback(ggml_backend_sched_t sched, ggml_backend_sched_eval_callback callback, void * user_data);
//
// Utils
//
struct ggml_backend_graph_copy {
ggml_backend_buffer_t buffer;
struct ggml_context * ctx_allocated;
struct ggml_context * ctx_unallocated;
struct ggml_cgraph * graph;
};
// Copy a graph to a different backend
GGML_API struct ggml_backend_graph_copy ggml_backend_graph_copy(ggml_backend_t backend, struct ggml_cgraph * graph);
GGML_API void ggml_backend_graph_copy_free(struct ggml_backend_graph_copy copy);
typedef bool (*GGML_CALL ggml_backend_eval_callback)(int node_index, struct ggml_tensor * t1, struct ggml_tensor * t2, void * user_data);
// Compare the output of two backends
GGML_API bool ggml_backend_compare_graph_backend(ggml_backend_t backend1, ggml_backend_t backend2, struct ggml_cgraph * graph, ggml_backend_eval_callback callback, void * user_data);
// Tensor initialization
GGML_API void ggml_backend_tensor_alloc(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor, void * addr);
GGML_API void ggml_backend_view_init(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor);
#ifdef __cplusplus
}
#endif

2301
llama.xcframework/ios-arm64/llama.framework/Headers/ggml.h Normal file
View file

File diff suppressed because it is too large Load diff

29
llama.xcframework/ios-arm64/llama.framework/Headers/grammar-parser.h Normal file
View file

@ -0,0 +1,29 @@
// Implements a parser for an extended Backus-Naur form (BNF), producing the
// binary context-free grammar format specified by llama.h. Supports character
// ranges, grouping, and repetition operators. As an example, a grammar for
// arithmetic might look like:
//
// root ::= expr
// expr ::= term ([-+*/] term)*
// term ::= num | "(" space expr ")" space
// num ::= [0-9]+ space
// space ::= [ \t\n]*
#pragma once
#include "llama.h"
#include <vector>
#include <map>
#include <cstdint>
#include <string>
namespace grammar_parser {
struct parse_state {
std::map<std::string, uint32_t> symbol_ids;
std::vector<std::vector<llama_grammar_element>> rules;
std::vector<const llama_grammar_element *> c_rules();
};
parse_state parse(const char * src);
void print_grammar(FILE * file, const parse_state & state);
}

915
llama.xcframework/ios-arm64/llama.framework/Headers/llama.h Normal file
View file

@ -0,0 +1,915 @@
#ifndef LLAMA_H
#define LLAMA_H
#include "ggml.h"
#include "ggml-backend.h"
#ifdef GGML_USE_CUBLAS
#include "ggml-cuda.h"
#define LLAMA_MAX_DEVICES GGML_CUDA_MAX_DEVICES
#else
#define LLAMA_MAX_DEVICES 1
#endif // GGML_USE_CUBLAS
#include <stddef.h>
#include <stdint.h>
#include <stdio.h>
#include <stdbool.h>
#ifdef LLAMA_SHARED
# if defined(_WIN32) && !defined(__MINGW32__)
# ifdef LLAMA_BUILD
# define LLAMA_API __declspec(dllexport)
# else
# define LLAMA_API __declspec(dllimport)
# endif
# else
# define LLAMA_API __attribute__ ((visibility ("default")))
# endif
#else
# define LLAMA_API
#endif
#ifdef __GNUC__
# define DEPRECATED(func, hint) func __attribute__((deprecated(hint)))
#elif defined(_MSC_VER)
# define DEPRECATED(func, hint) __declspec(deprecated(hint)) func
#else
# define DEPRECATED(func, hint) func
#endif
#define LLAMA_DEFAULT_SEED 0xFFFFFFFF
#define LLAMA_MAX_RNG_STATE (64*1024)
#define LLAMA_FILE_MAGIC_GGLA 0x67676c61u // 'ggla'
#define LLAMA_FILE_MAGIC_GGSN 0x6767736eu // 'ggsn'
#define LLAMA_SESSION_MAGIC LLAMA_FILE_MAGIC_GGSN
#define LLAMA_SESSION_VERSION 4
#if defined(GGML_USE_CUBLAS) || defined(GGML_USE_CLBLAST) || defined(GGML_USE_METAL)
// Defined when llama.cpp is compiled with support for offloading model layers to GPU.
#define LLAMA_SUPPORTS_GPU_OFFLOAD
#endif
#ifdef __cplusplus
extern "C" {
#endif
//
// C interface
//
// TODO: show sample usage
//
struct llama_model;
struct llama_context;
typedef int32_t llama_pos;
typedef int32_t llama_token;
typedef int32_t llama_seq_id;
enum llama_vocab_type {
LLAMA_VOCAB_TYPE_SPM = 0, // SentencePiece
LLAMA_VOCAB_TYPE_BPE = 1, // Byte Pair Encoding
};
enum llama_token_type {
LLAMA_TOKEN_TYPE_UNDEFINED = 0,
LLAMA_TOKEN_TYPE_NORMAL = 1,
LLAMA_TOKEN_TYPE_UNKNOWN = 2,
LLAMA_TOKEN_TYPE_CONTROL = 3,
LLAMA_TOKEN_TYPE_USER_DEFINED = 4,
LLAMA_TOKEN_TYPE_UNUSED = 5,
LLAMA_TOKEN_TYPE_BYTE = 6,
};
// model file types
enum llama_ftype {
LLAMA_FTYPE_ALL_F32 = 0,
LLAMA_FTYPE_MOSTLY_F16 = 1, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q4_0 = 2, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q4_1 = 3, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q4_1_SOME_F16 = 4, // tok_embeddings.weight and output.weight are F16
// LLAMA_FTYPE_MOSTLY_Q4_2 = 5, // support has been removed
// LLAMA_FTYPE_MOSTLY_Q4_3 = 6, // support has been removed
LLAMA_FTYPE_MOSTLY_Q8_0 = 7, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q5_0 = 8, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q5_1 = 9, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q2_K = 10, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q3_K_S = 11, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q3_K_M = 12, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q3_K_L = 13, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q4_K_S = 14, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q4_K_M = 15, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q5_K_S = 16, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q5_K_M = 17, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q6_K = 18, // except 1d tensors
LLAMA_FTYPE_MOSTLY_IQ2_XXS = 19, // except 1d tensors
LLAMA_FTYPE_MOSTLY_IQ2_XS = 20, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q2_K_S = 21, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q3_K_XS = 22, // except 1d tensors
LLAMA_FTYPE_GUESSED = 1024, // not specified in the model file
};
enum llama_rope_scaling_type {
LLAMA_ROPE_SCALING_UNSPECIFIED = -1,
LLAMA_ROPE_SCALING_NONE = 0,
LLAMA_ROPE_SCALING_LINEAR = 1,
LLAMA_ROPE_SCALING_YARN = 2,
LLAMA_ROPE_SCALING_MAX_VALUE = LLAMA_ROPE_SCALING_YARN,
};
enum llama_split_mode {
LLAMA_SPLIT_NONE = 0, // single GPU
LLAMA_SPLIT_LAYER = 1, // split layers and KV across GPUs
LLAMA_SPLIT_ROW = 2, // split rows across GPUs
};
typedef struct llama_token_data {
llama_token id; // token id
float logit; // log-odds of the token
float p; // probability of the token
} llama_token_data;
typedef struct llama_token_data_array {
llama_token_data * data;
size_t size;
bool sorted;
} llama_token_data_array;
typedef bool (*llama_progress_callback)(float progress, void *ctx);
// Input data for llama_decode
// A llama_batch object can contain input about one or many sequences
// The provided arrays (i.e. token, embd, pos, etc.) must have size of n_tokens
//
// - token : the token ids of the input (used when embd is NULL)
// - embd : token embeddings (i.e. float vector of size n_embd) (used when token is NULL)
// - pos : the positions of the respective token in the sequence
// - seq_id : the sequence to which the respective token belongs
// - logits : if zero, the logits for the respective token will not be output
//
typedef struct llama_batch {
int32_t n_tokens;
llama_token * token;
float * embd;
llama_pos * pos;
int32_t * n_seq_id;
llama_seq_id ** seq_id;
int8_t * logits;
// NOTE: helpers for smooth API transition - can be deprecated in the future
// for future-proof code, use the above fields instead and ignore everything below
//
// pos[i] = all_pos_0 + i*all_pos_1
//
llama_pos all_pos_0; // used if pos == NULL
llama_pos all_pos_1; // used if pos == NULL
llama_seq_id all_seq_id; // used if seq_id == NULL
} llama_batch;
enum llama_model_kv_override_type {
LLAMA_KV_OVERRIDE_INT,
LLAMA_KV_OVERRIDE_FLOAT,
LLAMA_KV_OVERRIDE_BOOL,
};
struct llama_model_kv_override {
char key[128];
enum llama_model_kv_override_type tag;
union {
int64_t int_value;
double float_value;
bool bool_value;
};
};
struct llama_model_params {
int32_t n_gpu_layers; // number of layers to store in VRAM
enum llama_split_mode split_mode; // how to split the model across multiple GPUs
// main_gpu interpretation depends on split_mode:
// LLAMA_SPLIT_NONE: the GPU that is used for the entire model
// LLAMA_SPLIT_ROW: the GPU that is used for small tensors and intermediate results
// LLAMA_SPLIT_LAYER: ignored
int32_t main_gpu;
// proportion of the model (layers or rows) to offload to each GPU, size: LLAMA_MAX_DEVICES
const float * tensor_split;
// Called with a progress value between 0.0 and 1.0. Pass NULL to disable.
// If the provided progress_callback returns true, model loading continues.
// If it returns false, model loading is immediately aborted.
llama_progress_callback progress_callback;
// context pointer passed to the progress callback
void * progress_callback_user_data;
// override key-value pairs of the model meta data
const struct llama_model_kv_override * kv_overrides;
// Keep the booleans together to avoid misalignment during copy-by-value.
bool vocab_only; // only load the vocabulary, no weights
bool use_mmap; // use mmap if possible
bool use_mlock; // force system to keep model in RAM
};
struct llama_context_params {
uint32_t seed; // RNG seed, -1 for random
uint32_t n_ctx; // text context, 0 = from model
uint32_t n_batch; // prompt processing maximum batch size
uint32_t n_threads; // number of threads to use for generation
uint32_t n_threads_batch; // number of threads to use for batch processing
int8_t rope_scaling_type; // RoPE scaling type, from `enum llama_rope_scaling_type`
// ref: https://github.com/ggerganov/llama.cpp/pull/2054
float rope_freq_base; // RoPE base frequency, 0 = from model
float rope_freq_scale; // RoPE frequency scaling factor, 0 = from model
float yarn_ext_factor; // YaRN extrapolation mix factor, negative = from model
float yarn_attn_factor; // YaRN magnitude scaling factor
float yarn_beta_fast; // YaRN low correction dim
float yarn_beta_slow; // YaRN high correction dim
uint32_t yarn_orig_ctx; // YaRN original context size
ggml_backend_sched_eval_callback cb_eval;
void * cb_eval_user_data;
enum ggml_type type_k; // data type for K cache
enum ggml_type type_v; // data type for V cache
// Keep the booleans together to avoid misalignment during copy-by-value.
bool mul_mat_q; // if true, use experimental mul_mat_q kernels (DEPRECATED - always true)
bool logits_all; // the llama_eval() call computes all logits, not just the last one (DEPRECATED - set llama_batch.logits instead)
bool embedding; // embedding mode only
bool offload_kqv; // whether to offload the KQV ops (including the KV cache) to GPU
};
// model quantization parameters
typedef struct llama_model_quantize_params {
int32_t nthread; // number of threads to use for quantizing, if <=0 will use std::thread::hardware_concurrency()
enum llama_ftype ftype; // quantize to this llama_ftype
bool allow_requantize; // allow quantizing non-f32/f16 tensors
bool quantize_output_tensor; // quantize output.weight
bool only_copy; // only copy tensors - ftype, allow_requantize and quantize_output_tensor are ignored
bool pure; // disable k-quant mixtures and quantize all tensors to the same type
void * imatrix; // pointer to importance matrix data
} llama_model_quantize_params;
// grammar types
struct llama_grammar;
// grammar element type
enum llama_gretype {
// end of rule definition
LLAMA_GRETYPE_END = 0,
// start of alternate definition for rule
LLAMA_GRETYPE_ALT = 1,
// non-terminal element: reference to rule
LLAMA_GRETYPE_RULE_REF = 2,
// terminal element: character (code point)
LLAMA_GRETYPE_CHAR = 3,
// inverse char(s) ([^a], [^a-b] [^abc])
LLAMA_GRETYPE_CHAR_NOT = 4,
// modifies a preceding LLAMA_GRETYPE_CHAR or LLAMA_GRETYPE_CHAR_ALT to
// be an inclusive range ([a-z])
LLAMA_GRETYPE_CHAR_RNG_UPPER = 5,
// modifies a preceding LLAMA_GRETYPE_CHAR or
// LLAMA_GRETYPE_CHAR_RNG_UPPER to add an alternate char to match ([ab], [a-zA])
LLAMA_GRETYPE_CHAR_ALT = 6,
};
typedef struct llama_grammar_element {
enum llama_gretype type;
uint32_t value; // Unicode code point or rule ID
} llama_grammar_element;
// performance timing information
struct llama_timings {
double t_start_ms;
double t_end_ms;
double t_load_ms;
double t_sample_ms;
double t_p_eval_ms;
double t_eval_ms;
int32_t n_sample;
int32_t n_p_eval;
int32_t n_eval;
};
// Helpers for getting default parameters
LLAMA_API struct llama_model_params llama_model_default_params(void);
LLAMA_API struct llama_context_params llama_context_default_params(void);
LLAMA_API struct llama_model_quantize_params llama_model_quantize_default_params(void);
// Initialize the llama + ggml backend
// If numa is true, use NUMA optimizations
// Call once at the start of the program
LLAMA_API void llama_backend_init(bool numa);
// Call once at the end of the program - currently only used for MPI
LLAMA_API void llama_backend_free(void);
LLAMA_API struct llama_model * llama_load_model_from_file(
const char * path_model,
struct llama_model_params params);
LLAMA_API void llama_free_model(struct llama_model * model);
LLAMA_API struct llama_context * llama_new_context_with_model(
struct llama_model * model,
struct llama_context_params params);
// Frees all allocated memory
LLAMA_API void llama_free(struct llama_context * ctx);
LLAMA_API int64_t llama_time_us(void);
LLAMA_API int32_t llama_max_devices(void);
LLAMA_API bool llama_mmap_supported (void);
LLAMA_API bool llama_mlock_supported(void);
LLAMA_API const struct llama_model * llama_get_model(const struct llama_context * ctx);
LLAMA_API uint32_t llama_n_ctx (const struct llama_context * ctx);
LLAMA_API uint32_t llama_n_batch (const struct llama_context * ctx);
LLAMA_API enum llama_vocab_type llama_vocab_type(const struct llama_model * model);
LLAMA_API int32_t llama_n_vocab (const struct llama_model * model);
LLAMA_API int32_t llama_n_ctx_train(const struct llama_model * model);
LLAMA_API int32_t llama_n_embd (const struct llama_model * model);
// Get the model's RoPE frequency scaling factor
LLAMA_API float llama_rope_freq_scale_train(const struct llama_model * model);
// Functions to access the model's GGUF metadata scalar values
// - The functions return the length of the string on success, or -1 on failure
// - The output string is always null-terminated and cleared on failure
// - GGUF array values are not supported by these functions
// Get metadata value as a string by key name
LLAMA_API int32_t llama_model_meta_val_str(const struct llama_model * model, const char * key, char * buf, size_t buf_size);
// Get the number of metadata key/value pairs
LLAMA_API int32_t llama_model_meta_count(const struct llama_model * model);
// Get metadata key name by index
LLAMA_API int32_t llama_model_meta_key_by_index(const struct llama_model * model, int32_t i, char * buf, size_t buf_size);
// Get metadata value as a string by index
LLAMA_API int32_t llama_model_meta_val_str_by_index(const struct llama_model * model, int32_t i, char * buf, size_t buf_size);
// Get a string describing the model type
LLAMA_API int32_t llama_model_desc(const struct llama_model * model, char * buf, size_t buf_size);
// Returns the total size of all the tensors in the model in bytes
LLAMA_API uint64_t llama_model_size(const struct llama_model * model);
// Returns the total number of parameters in the model
LLAMA_API uint64_t llama_model_n_params(const struct llama_model * model);
// Get a llama model tensor
LLAMA_API struct ggml_tensor * llama_get_model_tensor(struct llama_model * model, const char * name);
// Returns 0 on success
LLAMA_API uint32_t llama_model_quantize(
const char * fname_inp,
const char * fname_out,
const llama_model_quantize_params * params);
// Apply a LoRA adapter to a loaded model
// path_base_model is the path to a higher quality model to use as a base for
// the layers modified by the adapter. Can be NULL to use the current loaded model.
// The model needs to be reloaded before applying a new adapter, otherwise the adapter
// will be applied on top of the previous one
// Returns 0 on success
LLAMA_API DEPRECATED(int32_t llama_apply_lora_from_file(
struct llama_context * ctx,
const char * path_lora,
float scale,
const char * path_base_model,
int32_t n_threads),
"use llama_model_apply_lora_from_file instead");
LLAMA_API int32_t llama_model_apply_lora_from_file(
const struct llama_model * model,
const char * path_lora,
float scale,
const char * path_base_model,
int32_t n_threads);
//
// KV cache
//
// Information associated with an individual cell in the KV cache view.
struct llama_kv_cache_view_cell {
// The position for this cell. Takes KV cache shifts into account.
// May be negative if the cell is not populated.
llama_pos pos;
};
// An updateable view of the KV cache.
struct llama_kv_cache_view {
// Number of KV cache cells. This will be the same as the context size.
int32_t n_cells;
// Maximum number of sequences that can exist in a cell. It's not an error
// if there are more sequences in a cell than this value, however they will
// not be visible in the view cells_sequences.
int32_t n_max_seq;
// Number of tokens in the cache. For example, if there are two populated
// cells, the first with 1 sequence id in it and the second with 2 sequence
// ids then you'll have 3 tokens.
int32_t token_count;
// Number of populated cache cells.
int32_t used_cells;
// Maximum contiguous empty slots in the cache.
int32_t max_contiguous;
// Index to the start of the max_contiguous slot range. Can be negative
// when cache is full.
int32_t max_contiguous_idx;
// Information for an individual cell.
struct llama_kv_cache_view_cell * cells;
// The sequences for each cell. There will be n_max_seq items per cell.
llama_seq_id * cells_sequences;
};
// Create an empty KV cache view. (use only for debugging purposes)
LLAMA_API struct llama_kv_cache_view llama_kv_cache_view_init(const struct llama_context * ctx, int32_t n_max_seq);
// Free a KV cache view. (use only for debugging purposes)
LLAMA_API void llama_kv_cache_view_free(struct llama_kv_cache_view * view);
// Update the KV cache view structure with the current state of the KV cache. (use only for debugging purposes)
LLAMA_API void llama_kv_cache_view_update(const struct llama_context * ctx, struct llama_kv_cache_view * view);
// Returns the number of tokens in the KV cache (slow, use only for debug)
// If a KV cell has multiple sequences assigned to it, it will be counted multiple times
LLAMA_API int32_t llama_get_kv_cache_token_count(const struct llama_context * ctx);
// Returns the number of used KV cells (i.e. have at least one sequence assigned to them)
LLAMA_API int32_t llama_get_kv_cache_used_cells(const struct llama_context * ctx);
// Clear the KV cache
LLAMA_API void llama_kv_cache_clear(
struct llama_context * ctx);
// Removes all tokens that belong to the specified sequence and have positions in [p0, p1)
// seq_id < 0 : match any sequence
// p0 < 0 : [0, p1]
// p1 < 0 : [p0, inf)
LLAMA_API void llama_kv_cache_seq_rm(
struct llama_context * ctx,
llama_seq_id seq_id,
llama_pos p0,
llama_pos p1);
// Copy all tokens that belong to the specified sequence to another sequence
// Note that this does not allocate extra KV cache memory - it simply assigns the tokens to the new sequence
// p0 < 0 : [0, p1]
// p1 < 0 : [p0, inf)
LLAMA_API void llama_kv_cache_seq_cp(
struct llama_context * ctx,
llama_seq_id seq_id_src,
llama_seq_id seq_id_dst,
llama_pos p0,
llama_pos p1);
// Removes all tokens that do not belong to the specified sequence
LLAMA_API void llama_kv_cache_seq_keep(
struct llama_context * ctx,
llama_seq_id seq_id);
// Adds relative position "delta" to all tokens that belong to the specified sequence and have positions in [p0, p1)
// If the KV cache is RoPEd, the KV data is updated accordingly
// p0 < 0 : [0, p1]
// p1 < 0 : [p0, inf)
LLAMA_API void llama_kv_cache_seq_shift(
struct llama_context * ctx,
llama_seq_id seq_id,
llama_pos p0,
llama_pos p1,
llama_pos delta);
// Integer division of the positions by factor of `d > 1`
// If the KV cache is RoPEd, the KV data is updated accordingly
// p0 < 0 : [0, p1]
// p1 < 0 : [p0, inf)
LLAMA_API void llama_kv_cache_seq_div(
struct llama_context * ctx,
llama_seq_id seq_id,
llama_pos p0,
llama_pos p1,
int d);
//
// State / sessions
//
// Returns the maximum size in bytes of the state (rng, logits, embedding
// and kv_cache) - will often be smaller after compacting tokens
LLAMA_API size_t llama_get_state_size(const struct llama_context * ctx);
// Copies the state to the specified destination address.
// Destination needs to have allocated enough memory.
// Returns the number of bytes copied
LLAMA_API size_t llama_copy_state_data(
struct llama_context * ctx,
uint8_t * dst);
// Set the state reading from the specified address
// Returns the number of bytes read
LLAMA_API size_t llama_set_state_data(
struct llama_context * ctx,
uint8_t * src);
// Save/load session file
LLAMA_API bool llama_load_session_file(
struct llama_context * ctx,
const char * path_session,
llama_token * tokens_out,
size_t n_token_capacity,
size_t * n_token_count_out);
LLAMA_API bool llama_save_session_file(
struct llama_context * ctx,
const char * path_session,
const llama_token * tokens,
size_t n_token_count);
//
// Decoding
//
// Run the llama inference to obtain the logits and probabilities for the next token(s).
// tokens + n_tokens is the provided batch of new tokens to process
// n_past is the number of tokens to use from previous eval calls
// Returns 0 on success
// DEPRECATED: use llama_decode() instead
LLAMA_API DEPRECATED(int llama_eval(
struct llama_context * ctx,
llama_token * tokens,
int32_t n_tokens,
int32_t n_past),
"use llama_decode() instead");
// Same as llama_eval, but use float matrix input directly.
// DEPRECATED: use llama_decode() instead
LLAMA_API DEPRECATED(int llama_eval_embd(
struct llama_context * ctx,
float * embd,
int32_t n_tokens,
int32_t n_past),
"use llama_decode() instead");
// Return batch for single sequence of tokens starting at pos_0
//
// NOTE: this is a helper function to facilitate transition to the new batch API - avoid using it
//
LLAMA_API struct llama_batch llama_batch_get_one(
llama_token * tokens,
int32_t n_tokens,
llama_pos pos_0,
llama_seq_id seq_id);
// Allocates a batch of tokens on the heap that can hold a maximum of n_tokens
// Each token can be assigned up to n_seq_max sequence ids
// The batch has to be freed with llama_batch_free()
// If embd != 0, llama_batch.embd will be allocated with size of n_tokens * embd * sizeof(float)
// Otherwise, llama_batch.token will be allocated to store n_tokens llama_token
// The rest of the llama_batch members are allocated with size n_tokens
// All members are left uninitialized
LLAMA_API struct llama_batch llama_batch_init(
int32_t n_tokens,
int32_t embd,
int32_t n_seq_max);
// Frees a batch of tokens allocated with llama_batch_init()
LLAMA_API void llama_batch_free(struct llama_batch batch);
// Positive return values does not mean a fatal error, but rather a warning.
// 0 - success
// 1 - could not find a KV slot for the batch (try reducing the size of the batch or increase the context)
// < 0 - error
LLAMA_API int32_t llama_decode(
struct llama_context * ctx,
struct llama_batch batch);
// Set the number of threads used for decoding
// n_threads is the number of threads used for generation (single token)
// n_threads_batch is the number of threads used for prompt and batch processing (multiple tokens)
LLAMA_API void llama_set_n_threads(struct llama_context * ctx, uint32_t n_threads, uint32_t n_threads_batch);
// Token logits obtained from the last call to llama_eval()
// The logits for the last token are stored in the last row
// Logits for which llama_batch.logits[i] == 0 are undefined
// Rows: n_tokens provided with llama_batch
// Cols: n_vocab
LLAMA_API float * llama_get_logits(struct llama_context * ctx);
// Logits for the ith token. Equivalent to:
// llama_get_logits(ctx) + i*n_vocab
LLAMA_API float * llama_get_logits_ith(struct llama_context * ctx, int32_t i);
// Get the embeddings for the input
// shape: [n_embd] (1-dimensional)
LLAMA_API float * llama_get_embeddings(struct llama_context * ctx);
//
// Vocab
//
LLAMA_API const char * llama_token_get_text(const struct llama_model * model, llama_token token);
LLAMA_API float llama_token_get_score(const struct llama_model * model, llama_token token);
LLAMA_API enum llama_token_type llama_token_get_type(const struct llama_model * model, llama_token token);
// Special tokens
LLAMA_API llama_token llama_token_bos(const struct llama_model * model); // beginning-of-sentence
LLAMA_API llama_token llama_token_eos(const struct llama_model * model); // end-of-sentence
LLAMA_API llama_token llama_token_nl (const struct llama_model * model); // next-line
// Returns -1 if unknown, 1 for true or 0 for false.
LLAMA_API int32_t llama_add_bos_token(const struct llama_model * model);
// Returns -1 if unknown, 1 for true or 0 for false.
LLAMA_API int32_t llama_add_eos_token(const struct llama_model * model);
// codellama infill tokens
LLAMA_API llama_token llama_token_prefix(const struct llama_model * model); // Beginning of infill prefix
LLAMA_API llama_token llama_token_middle(const struct llama_model * model); // Beginning of infill middle
LLAMA_API llama_token llama_token_suffix(const struct llama_model * model); // Beginning of infill suffix
LLAMA_API llama_token llama_token_eot (const struct llama_model * model); // End of infill middle
//
// Tokenization
//
/// @details Convert the provided text into tokens.
/// @param tokens The tokens pointer must be large enough to hold the resulting tokens.
/// @return Returns the number of tokens on success, no more than n_max_tokens
/// @return Returns a negative number on failure - the number of tokens that would have been returned
/// @param special Allow tokenizing special and/or control tokens which otherwise are not exposed and treated as plaintext.
/// Does not insert a leading space.
LLAMA_API int32_t llama_tokenize(
const struct llama_model * model,
const char * text,
int32_t text_len,
llama_token * tokens,
int32_t n_max_tokens,
bool add_bos,
bool special);
// Token Id -> Piece.
// Uses the vocabulary in the provided context.
// Does not write null terminator to the buffer.
// User code is responsible to remove the leading whitespace of the first non-BOS token when decoding multiple tokens.
LLAMA_API int32_t llama_token_to_piece(
const struct llama_model * model,
llama_token token,
char * buf,
int32_t length);
//
// Grammar
//
LLAMA_API struct llama_grammar * llama_grammar_init(
const llama_grammar_element ** rules,
size_t n_rules,
size_t start_rule_index);
LLAMA_API void llama_grammar_free(struct llama_grammar * grammar);
LLAMA_API struct llama_grammar * llama_grammar_copy(const struct llama_grammar * grammar);
//
// Sampling functions
//
// Sets the current rng seed.
LLAMA_API void llama_set_rng_seed(struct llama_context * ctx, uint32_t seed);
/// @details Repetition penalty described in CTRL academic paper https://arxiv.org/abs/1909.05858, with negative logit fix.
/// @details Frequency and presence penalties described in OpenAI API https://platform.openai.com/docs/api-reference/parameter-details.
LLAMA_API void llama_sample_repetition_penalties(
struct llama_context * ctx,
llama_token_data_array * candidates,
const llama_token * last_tokens,
size_t penalty_last_n,
float penalty_repeat,
float penalty_freq,
float penalty_present);
/// @details Apply classifier-free guidance to the logits as described in academic paper "Stay on topic with Classifier-Free Guidance" https://arxiv.org/abs/2306.17806
/// @param logits Logits extracted from the original generation context.
/// @param logits_guidance Logits extracted from a separate context from the same model. Other than a negative prompt at the beginning, it should have all generated and user input tokens copied from the main context.
/// @param scale Guidance strength. 1.0f means no guidance. Higher values mean stronger guidance.
LLAMA_API void llama_sample_apply_guidance(
struct llama_context * ctx,
float * logits,
float * logits_guidance,
float scale);
LLAMA_API DEPRECATED(void llama_sample_classifier_free_guidance(
struct llama_context * ctx,
llama_token_data_array * candidates,
struct llama_context * guidance_ctx,
float scale),
"use llama_sample_apply_guidance() instead");
/// @details Sorts candidate tokens by their logits in descending order and calculate probabilities based on logits.
LLAMA_API void llama_sample_softmax(
struct llama_context * ctx,
llama_token_data_array * candidates);
/// @details Top-K sampling described in academic paper "The Curious Case of Neural Text Degeneration" https://arxiv.org/abs/1904.09751
LLAMA_API void llama_sample_top_k(
struct llama_context * ctx,
llama_token_data_array * candidates,
int32_t k,
size_t min_keep);
/// @details Nucleus sampling described in academic paper "The Curious Case of Neural Text Degeneration" https://arxiv.org/abs/1904.09751
LLAMA_API void llama_sample_top_p(
struct llama_context * ctx,
llama_token_data_array * candidates,
float p,
size_t min_keep);
/// @details Minimum P sampling as described in https://github.com/ggerganov/llama.cpp/pull/3841
LLAMA_API void llama_sample_min_p(
struct llama_context * ctx,
llama_token_data_array * candidates,
float p,
size_t min_keep);
/// @details Tail Free Sampling described in https://www.trentonbricken.com/Tail-Free-Sampling/.
LLAMA_API void llama_sample_tail_free(
struct llama_context * ctx,
llama_token_data_array * candidates,
float z,
size_t min_keep);
/// @details Locally Typical Sampling implementation described in the paper https://arxiv.org/abs/2202.00666.
LLAMA_API void llama_sample_typical(
struct llama_context * ctx,
llama_token_data_array * candidates,
float p,
size_t min_keep);
LLAMA_API void llama_sample_temp(
struct llama_context * ctx,
llama_token_data_array * candidates,
float temp);
LLAMA_API DEPRECATED(void llama_sample_temperature(
struct llama_context * ctx,
llama_token_data_array * candidates,
float temp),
"use llama_sample_temp instead");
/// @details Apply constraints from grammar
LLAMA_API void llama_sample_grammar(
struct llama_context * ctx,
llama_token_data_array * candidates,
const struct llama_grammar * grammar);
/// @details Mirostat 1.0 algorithm described in the paper https://arxiv.org/abs/2007.14966. Uses tokens instead of words.
/// @param candidates A vector of `llama_token_data` containing the candidate tokens, their probabilities (p), and log-odds (logit) for the current position in the generated text.
/// @param tau The target cross-entropy (or surprise) value you want to achieve for the generated text. A higher value corresponds to more surprising or less predictable text, while a lower value corresponds to less surprising or more predictable text.
/// @param eta The learning rate used to update `mu` based on the error between the target and observed surprisal of the sampled word. A larger learning rate will cause `mu` to be updated more quickly, while a smaller learning rate will result in slower updates.
/// @param m The number of tokens considered in the estimation of `s_hat`. This is an arbitrary value that is used to calculate `s_hat`, which in turn helps to calculate the value of `k`. In the paper, they use `m = 100`, but you can experiment with different values to see how it affects the performance of the algorithm.
/// @param mu Maximum cross-entropy. This value is initialized to be twice the target cross-entropy (`2 * tau`) and is updated in the algorithm based on the error between the target and observed surprisal.
LLAMA_API llama_token llama_sample_token_mirostat(
struct llama_context * ctx,
llama_token_data_array * candidates,
float tau,
float eta,
int32_t m,
float * mu);
/// @details Mirostat 2.0 algorithm described in the paper https://arxiv.org/abs/2007.14966. Uses tokens instead of words.
/// @param candidates A vector of `llama_token_data` containing the candidate tokens, their probabilities (p), and log-odds (logit) for the current position in the generated text.
/// @param tau The target cross-entropy (or surprise) value you want to achieve for the generated text. A higher value corresponds to more surprising or less predictable text, while a lower value corresponds to less surprising or more predictable text.
/// @param eta The learning rate used to update `mu` based on the error between the target and observed surprisal of the sampled word. A larger learning rate will cause `mu` to be updated more quickly, while a smaller learning rate will result in slower updates.
/// @param mu Maximum cross-entropy. This value is initialized to be twice the target cross-entropy (`2 * tau`) and is updated in the algorithm based on the error between the target and observed surprisal.
LLAMA_API llama_token llama_sample_token_mirostat_v2(
struct llama_context * ctx,
llama_token_data_array * candidates,
float tau,
float eta,
float * mu);
/// @details Selects the token with the highest probability.
/// Does not compute the token probabilities. Use llama_sample_softmax() instead.
LLAMA_API llama_token llama_sample_token_greedy(
struct llama_context * ctx,
llama_token_data_array * candidates);
/// @details Randomly selects a token from the candidates based on their probabilities.
LLAMA_API llama_token llama_sample_token(
struct llama_context * ctx,
llama_token_data_array * candidates);
/// @details Accepts the sampled token into the grammar
LLAMA_API void llama_grammar_accept_token(
struct llama_context * ctx,
struct llama_grammar * grammar,
llama_token token);
//
// Beam search
//
struct llama_beam_view {
const llama_token * tokens;
size_t n_tokens;
float p; // Cumulative beam probability (renormalized relative to all beams)
bool eob; // Callback should set this to true when a beam is at end-of-beam.
};
// Passed to beam_search_callback function.
// Whenever 0 < common_prefix_length, this number of tokens should be copied from any of the beams
// (e.g. beams[0]) as they will be removed (shifted) from all beams in all subsequent callbacks.
// These pointers are valid only during the synchronous callback, so should not be saved.
struct llama_beams_state {
struct llama_beam_view * beam_views;
size_t n_beams; // Number of elements in beam_views[].
size_t common_prefix_length; // Current max length of prefix tokens shared by all beams.
bool last_call; // True iff this is the last callback invocation.
};
// Type of pointer to the beam_search_callback function.
// void* callback_data is any custom data passed to llama_beam_search, that is subsequently
// passed back to beam_search_callback. This avoids having to use global variables in the callback.
typedef void (*llama_beam_search_callback_fn_t)(void * callback_data, struct llama_beams_state);
/// @details Deterministically returns entire sentence constructed by a beam search.
/// @param ctx Pointer to the llama_context.
/// @param callback Invoked for each iteration of the beam_search loop, passing in beams_state.
/// @param callback_data A pointer that is simply passed back to callback.
/// @param n_beams Number of beams to use.
/// @param n_past Number of tokens already evaluated.
/// @param n_predict Maximum number of tokens to predict. EOS may occur earlier.
LLAMA_API void llama_beam_search(
struct llama_context * ctx,
llama_beam_search_callback_fn_t callback,
void * callback_data,
size_t n_beams,
int32_t n_past,
int32_t n_predict);
// Performance information
LLAMA_API struct llama_timings llama_get_timings(struct llama_context * ctx);
LLAMA_API void llama_print_timings(struct llama_context * ctx);
LLAMA_API void llama_reset_timings(struct llama_context * ctx);
// Print system information
LLAMA_API const char * llama_print_system_info(void);
// Set callback for all future logging events.
// If this is not called, or NULL is supplied, everything is output on stderr.
LLAMA_API void llama_log_set(ggml_log_callback log_callback, void * user_data);
LLAMA_API void llama_dump_timing_info_yaml(FILE * stream, const struct llama_context * ctx);
#ifdef __cplusplus
}
#endif
// Internal API to be implemented by llama.cpp and used by tests/benchmarks only
#ifdef LLAMA_API_INTERNAL
#include <vector>
#include <string>
struct ggml_tensor;
const std::vector<std::pair<std::string, struct ggml_tensor *>> & llama_internal_get_tensor_map(
struct llama_context * ctx
);
#endif // LLAMA_API_INTERNAL
#endif // LLAMA_H

723
llama.xcframework/ios-arm64/llama.framework/Headers/log.h Normal file
View file

@ -0,0 +1,723 @@
#pragma once
#include <chrono>
#include <cstring>
#include <sstream>
#include <iostream>
#include <thread>
#include <vector>
#include <algorithm>
#include <cinttypes>
// --------------------------------
//
// Basic usage:
//
// --------
//
// The LOG() and LOG_TEE() macros are ready to go by default
// they do not require any initialization.
//
// LOGLN() and LOG_TEELN() are variants which automatically
// include \n character at the end of the log string.
//
// LOG() behaves exactly like printf, by default writing to a logfile.
// LOG_TEE() additionally, prints to the screen too ( mimics Unix tee command ).
//
// Default logfile is named
// "llama.<threadID>.log"
// Default LOG_TEE() secondary output target is
// stderr
//
// Logs can be dynamically disabled or enabled using functions:
// log_disable()
// and
// log_enable()
//
// A log target can be changed with:
// log_set_target( string )
// creating and opening, or re-opening a file by string filename
// or
// log_set_target( FILE* )
// allowing to point at stderr, stdout, or any valid FILE* file handler.
//
// --------
//
// End of Basic usage.
//
// --------------------------------
// Specifies a log target.
// default uses log_handler() with "llama.log" log file
// this can be changed, by defining LOG_TARGET
// like so:
//
// #define LOG_TARGET (a valid FILE*)
// #include "log.h"
//
// or it can be simply redirected to stdout or stderr
// like so:
//
// #define LOG_TARGET stderr
// #include "log.h"
//
// The log target can also be redirected to a different function
// like so:
//
// #define LOG_TARGET log_handler_different()
// #include "log.h"
//
// FILE* log_handler_different()
// {
// return stderr;
// }
//
// or:
//
// #define LOG_TARGET log_handler_another_one("somelog.log")
// #include "log.h"
//
// FILE* log_handler_another_one(char*filename)
// {
// static FILE* logfile = nullptr;
// (...)
// if( !logfile )
// {
// fopen(...)
// }
// (...)
// return logfile
// }
//
#ifndef LOG_TARGET
#define LOG_TARGET log_handler()
#endif
#ifndef LOG_TEE_TARGET
#define LOG_TEE_TARGET stderr
#endif
// Utility for synchronizing log configuration state
// since std::optional was introduced only in c++17
enum LogTriState
{
LogTriStateSame,
LogTriStateFalse,
LogTriStateTrue
};
// Utility to obtain "pid" like unique process id and use it when creating log files.
inline std::string log_get_pid()
{
static std::string pid;
if (pid.empty())
{
// std::this_thread::get_id() is the most portable way of obtaining a "process id"
// it's not the same as "pid" but is unique enough to solve multiple instances
// trying to write to the same log.
std::stringstream ss;
ss << std::this_thread::get_id();
pid = ss.str();
}
return pid;
}
// Utility function for generating log file names with unique id based on thread id.
// invocation with log_filename_generator( "llama", "log" ) creates a string "llama.<number>.log"
// where the number is a runtime id of the current thread.
#define log_filename_generator(log_file_basename, log_file_extension) log_filename_generator_impl(LogTriStateSame, log_file_basename, log_file_extension)
// INTERNAL, DO NOT USE
inline std::string log_filename_generator_impl(LogTriState multilog, const std::string & log_file_basename, const std::string & log_file_extension)
{
static bool _multilog = false;
if (multilog != LogTriStateSame)
{
_multilog = multilog == LogTriStateTrue;
}
std::stringstream buf;
buf << log_file_basename;
if (_multilog)
{
buf << ".";
buf << log_get_pid();
}
buf << ".";
buf << log_file_extension;
return buf.str();
}
#ifndef LOG_DEFAULT_FILE_NAME
#define LOG_DEFAULT_FILE_NAME log_filename_generator("llama", "log")
#endif
// Utility for turning #define values into string literals
// so we can have a define for stderr and
// we can print "stderr" instead of literal stderr, etc.
#define LOG_STRINGIZE1(s) #s
#define LOG_STRINGIZE(s) LOG_STRINGIZE1(s)
#define LOG_TEE_TARGET_STRING LOG_STRINGIZE(LOG_TEE_TARGET)
// Allows disabling timestamps.
// in order to disable, define LOG_NO_TIMESTAMPS
// like so:
//
// #define LOG_NO_TIMESTAMPS
// #include "log.h"
//
#ifndef LOG_NO_TIMESTAMPS
#ifndef _MSC_VER
#define LOG_TIMESTAMP_FMT "[%" PRIu64 "] "
#define LOG_TIMESTAMP_VAL , (std::chrono::duration_cast<std::chrono::duration<std::uint64_t>>(std::chrono::system_clock::now().time_since_epoch())).count()
#else
#define LOG_TIMESTAMP_FMT "[%" PRIu64 "] "
#define LOG_TIMESTAMP_VAL , (std::chrono::duration_cast<std::chrono::duration<std::uint64_t>>(std::chrono::system_clock::now().time_since_epoch())).count()
#endif
#else
#define LOG_TIMESTAMP_FMT "%s"
#define LOG_TIMESTAMP_VAL ,""
#endif
#ifdef LOG_TEE_TIMESTAMPS
#ifndef _MSC_VER
#define LOG_TEE_TIMESTAMP_FMT "[%" PRIu64 "] "
#define LOG_TEE_TIMESTAMP_VAL , (std::chrono::duration_cast<std::chrono::duration<std::uint64_t>>(std::chrono::system_clock::now().time_since_epoch())).count()
#else
#define LOG_TEE_TIMESTAMP_FMT "[%" PRIu64 "] "
#define LOG_TEE_TIMESTAMP_VAL , (std::chrono::duration_cast<std::chrono::duration<std::uint64_t>>(std::chrono::system_clock::now().time_since_epoch())).count()
#endif
#else
#define LOG_TEE_TIMESTAMP_FMT "%s"
#define LOG_TEE_TIMESTAMP_VAL ,""
#endif
// Allows disabling file/line/function prefix
// in order to disable, define LOG_NO_FILE_LINE_FUNCTION
// like so:
//
// #define LOG_NO_FILE_LINE_FUNCTION
// #include "log.h"
//
#ifndef LOG_NO_FILE_LINE_FUNCTION
#ifndef _MSC_VER
#define LOG_FLF_FMT "[%24s:%5d][%24s] "
#define LOG_FLF_VAL , __FILE__, __LINE__, __FUNCTION__
#else
#define LOG_FLF_FMT "[%24s:%5ld][%24s] "
#define LOG_FLF_VAL , __FILE__, __LINE__, __FUNCTION__
#endif
#else
#define LOG_FLF_FMT "%s"
#define LOG_FLF_VAL ,""
#endif
#ifdef LOG_TEE_FILE_LINE_FUNCTION
#ifndef _MSC_VER
#define LOG_TEE_FLF_FMT "[%24s:%5d][%24s] "
#define LOG_TEE_FLF_VAL , __FILE__, __LINE__, __FUNCTION__
#else
#define LOG_TEE_FLF_FMT "[%24s:%5ld][%24s] "
#define LOG_TEE_FLF_VAL , __FILE__, __LINE__, __FUNCTION__
#endif
#else
#define LOG_TEE_FLF_FMT "%s"
#define LOG_TEE_FLF_VAL ,""
#endif
// INTERNAL, DO NOT USE
// USE LOG() INSTEAD
//
#ifndef _MSC_VER
#define LOG_IMPL(str, ...) \
do { \
if (LOG_TARGET != nullptr) \
{ \
fprintf(LOG_TARGET, LOG_TIMESTAMP_FMT LOG_FLF_FMT str "%s" LOG_TIMESTAMP_VAL LOG_FLF_VAL, __VA_ARGS__); \
fflush(LOG_TARGET); \
} \
} while (0)
#else
#define LOG_IMPL(str, ...) \
do { \
if (LOG_TARGET != nullptr) \
{ \
fprintf(LOG_TARGET, LOG_TIMESTAMP_FMT LOG_FLF_FMT str "%s" LOG_TIMESTAMP_VAL LOG_FLF_VAL "", ##__VA_ARGS__); \
fflush(LOG_TARGET); \
} \
} while (0)
#endif
// INTERNAL, DO NOT USE
// USE LOG_TEE() INSTEAD
//
#ifndef _MSC_VER
#define LOG_TEE_IMPL(str, ...) \
do { \
if (LOG_TARGET != nullptr) \
{ \
fprintf(LOG_TARGET, LOG_TIMESTAMP_FMT LOG_FLF_FMT str "%s" LOG_TIMESTAMP_VAL LOG_FLF_VAL, __VA_ARGS__); \
fflush(LOG_TARGET); \
} \
if (LOG_TARGET != nullptr && LOG_TARGET != stdout && LOG_TARGET != stderr && LOG_TEE_TARGET != nullptr) \
{ \
fprintf(LOG_TEE_TARGET, LOG_TEE_TIMESTAMP_FMT LOG_TEE_FLF_FMT str "%s" LOG_TEE_TIMESTAMP_VAL LOG_TEE_FLF_VAL, __VA_ARGS__); \
fflush(LOG_TEE_TARGET); \
} \
} while (0)
#else
#define LOG_TEE_IMPL(str, ...) \
do { \
if (LOG_TARGET != nullptr) \
{ \
fprintf(LOG_TARGET, LOG_TIMESTAMP_FMT LOG_FLF_FMT str "%s" LOG_TIMESTAMP_VAL LOG_FLF_VAL "", ##__VA_ARGS__); \
fflush(LOG_TARGET); \
} \
if (LOG_TARGET != nullptr && LOG_TARGET != stdout && LOG_TARGET != stderr && LOG_TEE_TARGET != nullptr) \
{ \
fprintf(LOG_TEE_TARGET, LOG_TEE_TIMESTAMP_FMT LOG_TEE_FLF_FMT str "%s" LOG_TEE_TIMESTAMP_VAL LOG_TEE_FLF_VAL "", ##__VA_ARGS__); \
fflush(LOG_TEE_TARGET); \
} \
} while (0)
#endif
// The '\0' as a last argument, is a trick to bypass the silly
// "warning: ISO C++11 requires at least one argument for the "..." in a variadic macro"
// so we can have a single macro which can be called just like printf.
// Main LOG macro.
// behaves like printf, and supports arguments the exact same way.
//
#ifndef _MSC_VER
#define LOG(...) LOG_IMPL(__VA_ARGS__, "")
#else
#define LOG(str, ...) LOG_IMPL("%s" str, "", __VA_ARGS__, "")
#endif
// Main TEE macro.
// does the same as LOG
// and
// simultaneously writes stderr.
//
// Secondary target can be changed just like LOG_TARGET
// by defining LOG_TEE_TARGET
//
#ifndef _MSC_VER
#define LOG_TEE(...) LOG_TEE_IMPL(__VA_ARGS__, "")
#else
#define LOG_TEE(str, ...) LOG_TEE_IMPL("%s" str, "", __VA_ARGS__, "")
#endif
// LOG macro variants with auto endline.
#ifndef _MSC_VER
#define LOGLN(...) LOG_IMPL(__VA_ARGS__, "\n")
#define LOG_TEELN(...) LOG_TEE_IMPL(__VA_ARGS__, "\n")
#else
#define LOGLN(str, ...) LOG_IMPL("%s" str, "", __VA_ARGS__, "\n")
#define LOG_TEELN(str, ...) LOG_TEE_IMPL("%s" str, "", __VA_ARGS__, "\n")
#endif
// INTERNAL, DO NOT USE
inline FILE *log_handler1_impl(bool change = false, LogTriState append = LogTriStateSame, LogTriState disable = LogTriStateSame, const std::string & filename = LOG_DEFAULT_FILE_NAME, FILE *target = nullptr)
{
static bool _initialized = false;
static bool _append = false;
static bool _disabled = filename.empty() && target == nullptr;
static std::string log_current_filename{filename};
static FILE *log_current_target{target};
static FILE *logfile = nullptr;
if (change)
{
if (append != LogTriStateSame)
{
_append = append == LogTriStateTrue;
return logfile;
}
if (disable == LogTriStateTrue)
{
// Disable primary target
_disabled = true;
}
// If previously disabled, only enable, and keep previous target
else if (disable == LogTriStateFalse)
{
_disabled = false;
}
// Otherwise, process the arguments
else if (log_current_filename != filename || log_current_target != target)
{
_initialized = false;
}
}
if (_disabled)
{
// Log is disabled
return nullptr;
}
if (_initialized)
{
// with fallback in case something went wrong
return logfile ? logfile : stderr;
}
// do the (re)initialization
if (target != nullptr)
{
if (logfile != nullptr && logfile != stdout && logfile != stderr)
{
fclose(logfile);
}
log_current_filename = LOG_DEFAULT_FILE_NAME;
log_current_target = target;
logfile = target;
}
else
{
if (log_current_filename != filename)
{
if (logfile != nullptr && logfile != stdout && logfile != stderr)
{
fclose(logfile);
}
}
logfile = fopen(filename.c_str(), _append ? "a" : "w");
}
if (!logfile)
{
// Verify whether the file was opened, otherwise fallback to stderr
logfile = stderr;
fprintf(stderr, "Failed to open logfile '%s' with error '%s'\n", filename.c_str(), std::strerror(errno));
fflush(stderr);
// At this point we let the init flag be to true below, and let the target fallback to stderr
// otherwise we would repeatedly fopen() which was already unsuccessful
}
_initialized = true;
return logfile ? logfile : stderr;
}
// INTERNAL, DO NOT USE
inline FILE *log_handler2_impl(bool change = false, LogTriState append = LogTriStateSame, LogTriState disable = LogTriStateSame, FILE *target = nullptr, const std::string & filename = LOG_DEFAULT_FILE_NAME)
{
return log_handler1_impl(change, append, disable, filename, target);
}
// Disables logs entirely at runtime.
// Makes LOG() and LOG_TEE() produce no output,
// until enabled back.
#define log_disable() log_disable_impl()
// INTERNAL, DO NOT USE
inline FILE *log_disable_impl()
{
return log_handler1_impl(true, LogTriStateSame, LogTriStateTrue);
}
// Enables logs at runtime.
#define log_enable() log_enable_impl()
// INTERNAL, DO NOT USE
inline FILE *log_enable_impl()
{
return log_handler1_impl(true, LogTriStateSame, LogTriStateFalse);
}
// Sets target fir logs, either by a file name or FILE* pointer (stdout, stderr, or any valid FILE*)
#define log_set_target(target) log_set_target_impl(target)
// INTERNAL, DO NOT USE
inline FILE *log_set_target_impl(const std::string & filename) { return log_handler1_impl(true, LogTriStateSame, LogTriStateSame, filename); }
inline FILE *log_set_target_impl(FILE *target) { return log_handler2_impl(true, LogTriStateSame, LogTriStateSame, target); }
// INTERNAL, DO NOT USE
inline FILE *log_handler() { return log_handler1_impl(); }
// Enable or disable creating separate log files for each run.
// can ONLY be invoked BEFORE first log use.
#define log_multilog(enable) log_filename_generator_impl((enable) ? LogTriStateTrue : LogTriStateFalse, "", "")
// Enable or disable append mode for log file.
// can ONLY be invoked BEFORE first log use.
#define log_append(enable) log_append_impl(enable)
// INTERNAL, DO NOT USE
inline FILE *log_append_impl(bool enable)
{
return log_handler1_impl(true, enable ? LogTriStateTrue : LogTriStateFalse, LogTriStateSame);
}
inline void log_test()
{
log_disable();
LOG("01 Hello World to nobody, because logs are disabled!\n");
log_enable();
LOG("02 Hello World to default output, which is \"%s\" ( Yaaay, arguments! )!\n", LOG_STRINGIZE(LOG_TARGET));
LOG_TEE("03 Hello World to **both** default output and " LOG_TEE_TARGET_STRING "!\n");
log_set_target(stderr);
LOG("04 Hello World to stderr!\n");
LOG_TEE("05 Hello World TEE with double printing to stderr prevented!\n");
log_set_target(LOG_DEFAULT_FILE_NAME);
LOG("06 Hello World to default log file!\n");
log_set_target(stdout);
LOG("07 Hello World to stdout!\n");
log_set_target(LOG_DEFAULT_FILE_NAME);
LOG("08 Hello World to default log file again!\n");
log_disable();
LOG("09 Hello World _1_ into the void!\n");
log_enable();
LOG("10 Hello World back from the void ( you should not see _1_ in the log or the output )!\n");
log_disable();
log_set_target("llama.anotherlog.log");
LOG("11 Hello World _2_ to nobody, new target was selected but logs are still disabled!\n");
log_enable();
LOG("12 Hello World this time in a new file ( you should not see _2_ in the log or the output )?\n");
log_set_target("llama.yetanotherlog.log");
LOG("13 Hello World this time in yet new file?\n");
log_set_target(log_filename_generator("llama_autonamed", "log"));
LOG("14 Hello World in log with generated filename!\n");
#ifdef _MSC_VER
LOG_TEE("15 Hello msvc TEE without arguments\n");
LOG_TEE("16 Hello msvc TEE with (%d)(%s) arguments\n", 1, "test");
LOG_TEELN("17 Hello msvc TEELN without arguments\n");
LOG_TEELN("18 Hello msvc TEELN with (%d)(%s) arguments\n", 1, "test");
LOG("19 Hello msvc LOG without arguments\n");
LOG("20 Hello msvc LOG with (%d)(%s) arguments\n", 1, "test");
LOGLN("21 Hello msvc LOGLN without arguments\n");
LOGLN("22 Hello msvc LOGLN with (%d)(%s) arguments\n", 1, "test");
#endif
}
inline bool log_param_single_parse(const std::string & param)
{
if ( param == "--log-test")
{
log_test();
return true;
}
if ( param == "--log-disable")
{
log_disable();
return true;
}
if ( param == "--log-enable")
{
log_enable();
return true;
}
if (param == "--log-new")
{
log_multilog(true);
return true;
}
if (param == "--log-append")
{
log_append(true);
return true;
}
return false;
}
inline bool log_param_pair_parse(bool check_but_dont_parse, const std::string & param, const std::string & next = std::string())
{
if ( param == "--log-file")
{
if (!check_but_dont_parse)
{
log_set_target(log_filename_generator(next.empty() ? "unnamed" : next, "log"));
}
return true;
}
return false;
}
inline void log_print_usage()
{
printf("log options:\n");
/* format
printf(" -h, --help show this help message and exit\n");*/
/* spacing
printf("__-param----------------Description\n");*/
printf(" --log-test Run simple logging test\n");
printf(" --log-disable Disable trace logs\n");
printf(" --log-enable Enable trace logs\n");
printf(" --log-file Specify a log filename (without extension)\n");
printf(" --log-new Create a separate new log file on start. "
"Each log file will have unique name: \"<name>.<ID>.log\"\n");
printf(" --log-append Don't truncate the old log file.\n");
}
#define log_dump_cmdline(argc, argv) log_dump_cmdline_impl(argc, argv)
// INTERNAL, DO NOT USE
inline void log_dump_cmdline_impl(int argc, char **argv)
{
std::stringstream buf;
for (int i = 0; i < argc; ++i)
{
if (std::string(argv[i]).find(' ') != std::string::npos)
{
buf << " \"" << argv[i] <<"\"";
}
else
{
buf << " " << argv[i];
}
}
LOGLN("Cmd:%s", buf.str().c_str());
}
#define log_tostr(var) log_var_to_string_impl(var).c_str()
inline std::string log_var_to_string_impl(bool var)
{
return var ? "true" : "false";
}
inline std::string log_var_to_string_impl(std::string var)
{
return var;
}
inline std::string log_var_to_string_impl(const std::vector<int> & var)
{
std::stringstream buf;
buf << "[ ";
bool first = true;
for (auto e : var)
{
if (first)
{
first = false;
}
else
{
buf << ", ";
}
buf << std::to_string(e);
}
buf << " ]";
return buf.str();
}
template <typename C, typename T>
inline std::string LOG_TOKENS_TOSTR_PRETTY(const C & ctx, const T & tokens)
{
std::stringstream buf;
buf << "[ ";
bool first = true;
for (const auto &token : tokens)
{
if (!first) {
buf << ", ";
} else {
first = false;
}
auto detokenized = llama_token_to_piece(ctx, token);
detokenized.erase(
std::remove_if(
detokenized.begin(),
detokenized.end(),
[](const unsigned char c) { return !std::isprint(c); }),
detokenized.end());
buf
<< "'" << detokenized << "'"
<< ":" << std::to_string(token);
}
buf << " ]";
return buf.str();
}
template <typename C, typename B>
inline std::string LOG_BATCH_TOSTR_PRETTY(const C & ctx, const B & batch)
{
std::stringstream buf;
buf << "[ ";
bool first = true;
for (int i = 0; i < batch.n_tokens; ++i)
{
if (!first) {
buf << ", ";
} else {
first = false;
}
auto detokenized = llama_token_to_piece(ctx, batch.token[i]);
detokenized.erase(
std::remove_if(
detokenized.begin(),
detokenized.end(),
[](const unsigned char c) { return !std::isprint(c); }),
detokenized.end());
buf
<< "\n" << std::to_string(i)
<< ":token '" << detokenized << "'"
<< ":pos " << std::to_string(batch.pos[i])
<< ":n_seq_id " << std::to_string(batch.n_seq_id[i])
<< ":seq_id " << std::to_string(batch.seq_id[i][0])
<< ":logits " << std::to_string(batch.logits[i]);
}
buf << " ]";
return buf.str();
}
#ifdef LOG_DISABLE_LOGS
#undef LOG
#define LOG(...) // dummy stub
#undef LOGLN
#define LOGLN(...) // dummy stub
#undef LOG_TEE
#define LOG_TEE(...) fprintf(stderr, __VA_ARGS__) // convert to normal fprintf
#undef LOG_TEELN
#define LOG_TEELN(...) fprintf(stderr, __VA_ARGS__) // convert to normal fprintf
#undef LOG_DISABLE
#define LOG_DISABLE() // dummy stub
#undef LOG_ENABLE
#define LOG_ENABLE() // dummy stub
#undef LOG_ENABLE
#define LOG_ENABLE() // dummy stub
#undef LOG_SET_TARGET
#define LOG_SET_TARGET(...) // dummy stub
#undef LOG_DUMP_CMDLINE
#define LOG_DUMP_CMDLINE(...) // dummy stub
#endif // LOG_DISABLE_LOGS

117
llama.xcframework/ios-arm64/llama.framework/Headers/sampling.h Normal file
View file

@ -0,0 +1,117 @@
#pragma once
#include "llama.h"
#include "grammar-parser.h"
#include <string>
#include <vector>
#include <unordered_map>
// sampling parameters
typedef struct llama_sampling_params {
int32_t n_prev = 64; // number of previous tokens to remember
int32_t n_probs = 0; // if greater than 0, output the probabilities of top n_probs tokens.
int32_t top_k = 40; // <= 0 to use vocab size
float top_p = 0.95f; // 1.0 = disabled
float min_p = 0.05f; // 0.0 = disabled
float tfs_z = 1.00f; // 1.0 = disabled
float typical_p = 1.00f; // 1.0 = disabled
float temp = 0.80f; // <= 0.0 to sample greedily, 0.0 to not output probabilities
int32_t penalty_last_n = 64; // last n tokens to penalize (0 = disable penalty, -1 = context size)
float penalty_repeat = 1.10f; // 1.0 = disabled
float penalty_freq = 0.00f; // 0.0 = disabled
float penalty_present = 0.00f; // 0.0 = disabled
int32_t mirostat = 0; // 0 = disabled, 1 = mirostat, 2 = mirostat 2.0
float mirostat_tau = 5.00f; // target entropy
float mirostat_eta = 0.10f; // learning rate
bool penalize_nl = true; // consider newlines as a repeatable token
std::string samplers_sequence = "kfypmt"; // top_k, tail_free, typical_p, top_p, min_p, temp
std::string grammar; // optional BNF-like grammar to constrain sampling
// Classifier-Free Guidance
// https://arxiv.org/abs/2306.17806
std::string cfg_negative_prompt; // string to help guidance
float cfg_scale = 1.f; // how strong is guidance
std::unordered_map<llama_token, float> logit_bias; // logit bias for specific tokens
std::vector<llama_token> penalty_prompt_tokens;
bool use_penalty_prompt_tokens = false;
} llama_sampling_params;
// general sampler context
// TODO: move to llama.h
struct llama_sampling_context {
// parameters that will be used for sampling
llama_sampling_params params;
// mirostat sampler state
float mirostat_mu;
llama_grammar * grammar;
// internal
grammar_parser::parse_state parsed_grammar;
// TODO: replace with ring-buffer
std::vector<llama_token> prev;
std::vector<llama_token_data> cur;
};
#include "common.h"
// Create a new sampling context instance.
struct llama_sampling_context * llama_sampling_init(const struct llama_sampling_params & params);
void llama_sampling_free(struct llama_sampling_context * ctx);
// Reset the sampler context
// - clear prev tokens
// - reset grammar
void llama_sampling_reset(llama_sampling_context * ctx);
// Copy the sampler context
void llama_sampling_cp(llama_sampling_context * src, llama_sampling_context * dst);
// Get the last sampled token
llama_token llama_sampling_last(llama_sampling_context * ctx);
// Get a string representation of the last sampled tokens
std::string llama_sampling_prev_str(llama_sampling_context * ctx_sampling, llama_context * ctx_main, int n);
// Print sampling parameters into a string
std::string llama_sampling_print(const llama_sampling_params & params);
// Print sampling order into a string
std::string llama_sampling_order_print(const llama_sampling_params & params);
// this is a common sampling function used across the examples for convenience
// it can serve as a starting point for implementing your own sampling function
// Note: When using multiple sequences, it is the caller's responsibility to call
// llama_sampling_reset when a sequence ends
//
// required:
// - ctx_main: context to use for sampling
// - ctx_sampling: sampling-specific context
//
// optional:
// - ctx_cfg: context to use for classifier-free guidance
// - idx: sample from llama_get_logits_ith(ctx, idx)
//
// returns:
// - token: sampled token
// - candidates: vector of candidate tokens
//
llama_token llama_sampling_sample(
struct llama_sampling_context * ctx_sampling,
struct llama_context * ctx_main,
struct llama_context * ctx_cfg,
int idx = 0);
void llama_sampling_accept(
struct llama_sampling_context * ctx_sampling,
struct llama_context * ctx_main,
llama_token id,
bool apply_grammar);

31
llama.xcframework/ios-arm64/llama.framework/Headers/tokenize.h Normal file
View file

@ -0,0 +1,31 @@
//
// This source file is part of the Stanford Spezi open source project
//
// SPDX-FileCopyrightText: 2022 Stanford University and the project authors (see CONTRIBUTORS.md)
//
// SPDX-License-Identifier: MIT
//
#ifndef tokenize_hpp
#define tokenize_hpp
#include <vector>
#include "common.h"
/// Tokenize a `String` via a given `llama_context`.
std::vector<llama_token> llama_tokenize_with_context(
const struct llama_context * ctx,
const std::string & text,
bool add_bos,
bool special = false);
/// Tokenize a `String` via a given `llama_model`.
std::vector<llama_token> llama_tokenize_with_model(
const struct llama_model * model,
const std::string & text,
bool add_bos,
bool special = false);
#endif

20
llama.xcframework/ios-arm64/llama.framework/Headers/vector.h Normal file
View file

@ -0,0 +1,20 @@
//
// This source file is part of the Stanford Spezi open source project
//
// SPDX-FileCopyrightText: 2022 Stanford University and the project authors (see CONTRIBUTORS.md)
//
// SPDX-License-Identifier: MIT
//
#ifndef vector_hpp
#define vector_hpp
#include <vector>
#include "common.h"
/// Create an empty `vector` of `llama_seq_id`s that serve as a buffer for batch processing.
const std::vector<llama_seq_id> getLlamaSeqIdVector();
#endif

BIN
llama.xcframework/ios-arm64/llama.framework/Info.plist Normal file
View file

Binary file not shown.

21
llama.xcframework/ios-arm64/llama.framework/Modules/module.modulemap Normal file
View file

@ -0,0 +1,21 @@
//
// This source file is part of the Stanford Biodesign Digital Health Group open-source organization
//
// SPDX-FileCopyrightText: 2022 Stanford University and the project authors (see CONTRIBUTORS.md)
//
// SPDX-License-Identifier: MIT
//
framework module llama {
umbrella header "llama.h"
header "common.h"
header "sampling.h"
header "log.h"
header "grammar-parser.h"
header "tokenize.h"
header "vector.h"
export *
module * { export * }
}

BIN
llama.xcframework/ios-arm64/llama.framework/default.metallib Normal file
View file

Binary file not shown.

3
llama.xcframework/ios-arm64/llama.framework/ggml-metal.dat Normal file
View file

@ -0,0 +1,3 @@
dependencies: \
/Users/githubaction/runner/_work/llama.cpp/llama.cpp/ggml-metal.metal \
/Applications/Xcode-15.2.0.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/metal/ios/lib/clang/32023.98/include/metal/module.modulemap

BIN
llama.xcframework/ios-arm64/llama.framework/llama Normal file
View file

Binary file not shown.

270
llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/common.h Normal file
View file

@ -0,0 +1,270 @@
// Various helper functions and utilities
#pragma once
#include "llama.h"
#include "sampling.h"
#define LOG_NO_FILE_LINE_FUNCTION
#include "log.h"
#include <cmath>
#include <string>
#include <vector>
#include <random>
#include <thread>
#include <unordered_map>
#include <tuple>
#ifdef _WIN32
#define DIRECTORY_SEPARATOR '\\'
#else
#define DIRECTORY_SEPARATOR '/'
#endif // _WIN32
#define die(msg) do { fputs("error: " msg "\n", stderr); exit(1); } while (0)
#define die_fmt(fmt, ...) do { fprintf(stderr, "error: " fmt "\n", __VA_ARGS__); exit(1); } while (0)
#define print_build_info() do { \
fprintf(stderr, "%s: build = %d (%s)\n", __func__, LLAMA_BUILD_NUMBER, LLAMA_COMMIT); \
fprintf(stderr, "%s: built with %s for %s\n", __func__, LLAMA_COMPILER, LLAMA_BUILD_TARGET); \
} while(0)
// build info
// If macro is not defined (passed via Xcode), set the corresponding variable as defined via external variable.
#ifndef LLAMA_BUILD_NUMBER
extern int LLAMA_BUILD_NUMBER;
#endif
#ifndef LLAMA_COMMIT
extern char const *LLAMA_COMMIT;
#endif
#ifndef LLAMA_COMPILER
extern char const *LLAMA_COMPILER;
#endif
#ifndef LLAMA_BUILD_TARGET
extern char const *LLAMA_BUILD_TARGET;
#endif
//
// CLI argument parsing
//
int32_t get_num_physical_cores();
struct gpt_params {
uint32_t seed = -1; // RNG seed
int32_t n_threads = get_num_physical_cores();
int32_t n_threads_draft = -1;
int32_t n_threads_batch = -1; // number of threads to use for batch processing (-1 = use n_threads)
int32_t n_threads_batch_draft = -1;
int32_t n_predict = -1; // new tokens to predict
int32_t n_ctx = 512; // context size
int32_t n_batch = 512; // batch size for prompt processing (must be >=32 to use BLAS)
int32_t n_keep = 0; // number of tokens to keep from initial prompt
int32_t n_draft = 8; // number of tokens to draft during speculative decoding
int32_t n_chunks = -1; // max number of chunks to process (-1 = unlimited)
int32_t n_parallel = 1; // number of parallel sequences to decode
int32_t n_sequences = 1; // number of sequences to decode
float p_accept = 0.5f; // speculative decoding accept probability
float p_split = 0.1f; // speculative decoding split probability
int32_t n_gpu_layers = -1; // number of layers to store in VRAM (-1 - use default)
int32_t n_gpu_layers_draft = -1; // number of layers to store in VRAM for the draft model (-1 - use default)
llama_split_mode split_mode = LLAMA_SPLIT_LAYER; // how to split the model across GPUs
int32_t main_gpu = 0; // the GPU that is used for scratch and small tensors
float tensor_split[LLAMA_MAX_DEVICES] = {0}; // how split tensors should be distributed across GPUs
int32_t n_beams = 0; // if non-zero then use beam search of given width.
int32_t grp_attn_n = 1; // group-attention factor
int32_t grp_attn_w = 512; // group-attention width
int32_t n_print = -1; // print token count every n tokens (-1 = disabled)
float rope_freq_base = 0.0f; // RoPE base frequency
float rope_freq_scale = 0.0f; // RoPE frequency scaling factor
float yarn_ext_factor = -1.0f; // YaRN extrapolation mix factor
float yarn_attn_factor = 1.0f; // YaRN magnitude scaling factor
float yarn_beta_fast = 32.0f; // YaRN low correction dim
float yarn_beta_slow = 1.0f; // YaRN high correction dim
int32_t yarn_orig_ctx = 0; // YaRN original context length
int8_t rope_scaling_type = LLAMA_ROPE_SCALING_UNSPECIFIED; // TODO: better to be int32_t for alignment
// pinging @cebtenzzre
// // sampling parameters
struct llama_sampling_params sparams;
std::string model = "models/7B/ggml-model-f16.gguf"; // model path
std::string model_draft = ""; // draft model for speculative decoding
std::string model_alias = "unknown"; // model alias
std::string prompt = "";
std::string prompt_file = ""; // store the external prompt file name
std::string path_prompt_cache = ""; // path to file for saving/loading prompt eval state
std::string input_prefix = ""; // string to prefix user inputs with
std::string input_suffix = ""; // string to suffix user inputs with
std::vector<std::string> antiprompt; // string upon seeing which more user input is prompted
std::string logdir = ""; // directory in which to save YAML log files
std::string logits_file = ""; // file for saving *all* logits
std::vector<llama_model_kv_override> kv_overrides;
// TODO: avoid tuple, use struct
std::vector<std::tuple<std::string, float>> lora_adapter; // lora adapter path with user defined scale
std::string lora_base = ""; // base model path for the lora adapter
int ppl_stride = 0; // stride for perplexity calculations. If left at 0, the pre-existing approach will be used.
int ppl_output_type = 0; // = 0 -> ppl output is as usual, = 1 -> ppl output is num_tokens, ppl, one per line
// (which is more convenient to use for plotting)
//
bool hellaswag = false; // compute HellaSwag score over random tasks from datafile supplied in prompt
size_t hellaswag_tasks = 400; // number of tasks to use when computing the HellaSwag score
bool winogrande = false; // compute Winogrande score over random tasks from datafile supplied in prompt
size_t winogrande_tasks= 0; // number of tasks to use when computing the Winogrande score. If 0, all tasks will be computed
bool multiple_choice = false; // compute TruthfulQA score over random tasks from datafile supplied in prompt
size_t multiple_choice_tasks = 0; // number of tasks to use when computing the TruthfulQA score. If 0, all tasks will be computed
bool kl_divergence = false; // compute KL-divergence
bool mul_mat_q = true; // if true, use mul_mat_q kernels instead of cuBLAS
bool random_prompt = false; // do not randomize prompt if none provided
bool use_color = false; // use color to distinguish generations and inputs
bool interactive = false; // interactive mode
bool chatml = false; // chatml mode (used for models trained on chatml syntax)
bool prompt_cache_all = false; // save user input and generations to prompt cache
bool prompt_cache_ro = false; // open the prompt cache read-only and do not update it
bool embedding = false; // get only sentence embedding
bool escape = false; // escape "\n", "\r", "\t", "\'", "\"", and "\\"
bool interactive_first = false; // wait for user input immediately
bool multiline_input = false; // reverse the usage of `\`
bool simple_io = false; // improves compatibility with subprocesses and limited consoles
bool cont_batching = false; // insert new sequences for decoding on-the-fly
bool input_prefix_bos = false; // prefix BOS to user inputs, preceding input_prefix
bool ignore_eos = false; // ignore generated EOS tokens
bool instruct = false; // instruction mode (used for Alpaca models)
bool logits_all = false; // return logits for all tokens in the batch
bool use_mmap = true; // use mmap for faster loads
bool use_mlock = false; // use mlock to keep model in memory
bool numa = false; // attempt optimizations that help on some NUMA systems
bool verbose_prompt = false; // print prompt tokens before generation
bool display_prompt = true; // print prompt before generation
bool infill = false; // use infill mode
bool dump_kv_cache = false; // dump the KV cache contents for debugging purposes
bool no_kv_offload = false; // disable KV offloading
std::string cache_type_k = "f16"; // KV cache data type for the K
std::string cache_type_v = "f16"; // KV cache data type for the V
// multimodal models (see examples/llava)
std::string mmproj = ""; // path to multimodal projector
std::string image = ""; // path to an image file
};
bool gpt_params_parse_ex(int argc, char ** argv, gpt_params & params);
bool gpt_params_parse(int argc, char ** argv, gpt_params & params);
void gpt_print_usage(int argc, char ** argv, const gpt_params & params);
std::string get_system_info(const gpt_params & params);
std::string gpt_random_prompt(std::mt19937 & rng);
void process_escapes(std::string& input);
//
// String parsing
//
std::string parse_samplers_input(std::string input);
//
// Model utils
//
// TODO: avoid tuplue, use struct
std::tuple<struct llama_model *, struct llama_context *> llama_init_from_gpt_params(gpt_params & params);
struct llama_model_params llama_model_params_from_gpt_params (const gpt_params & params);
struct llama_context_params llama_context_params_from_gpt_params(const gpt_params & params);
// Batch utils
void llama_batch_clear(struct llama_batch & batch);
void llama_batch_add(
struct llama_batch & batch,
llama_token id,
llama_pos pos,
const std::vector<llama_seq_id> & seq_ids,
bool logits);
//
// Vocab utils
//
// tokenizes a string into a vector of tokens
// should work similar to Python's `tokenizer.encode`
std::vector<llama_token> llama_tokenize(
const struct llama_context * ctx,
const std::string & text,
bool add_bos,
bool special = false);
std::vector<llama_token> llama_tokenize(
const struct llama_model * model,
const std::string & text,
bool add_bos,
bool special = false);
// tokenizes a token into a piece
// should work similar to Python's `tokenizer.id_to_piece`
std::string llama_token_to_piece(
const struct llama_context * ctx,
llama_token token);
// TODO: these should be moved in llama.h C-style API under single `llama_detokenize` function
// that takes into account the tokenizer type and decides how to handle the leading space
//
// detokenizes a vector of tokens into a string
// should work similar to Python's `tokenizer.decode`
// removes the leading space from the first non-BOS token
std::string llama_detokenize_spm(
llama_context * ctx,
const std::vector<llama_token> & tokens);
// detokenizes a vector of tokens into a string
// should work similar to Python's `tokenizer.decode`
std::string llama_detokenize_bpe(
llama_context * ctx,
const std::vector<llama_token> & tokens);
// Uses the value from the model metadata if possible, otherwise
// defaults to true when model type is SPM, otherwise false.
bool llama_should_add_bos_token(const llama_model * model);
//
// YAML utils
//
bool create_directory_with_parents(const std::string & path);
void dump_vector_float_yaml(FILE * stream, const char * prop_name, const std::vector<float> & data);
void dump_vector_int_yaml(FILE * stream, const char * prop_name, const std::vector<int> & data);
void dump_string_yaml_multiline(FILE * stream, const char * prop_name, const char * data);
std::string get_sortable_timestamp();
void dump_non_result_info_yaml(
FILE * stream, const gpt_params & params, const llama_context * lctx,
const std::string & timestamp, const std::vector<int> & prompt_tokens, const char * model_desc);
//
// KV cache utils
//
// Dump the KV cache view with the number of sequences per cell.
void dump_kv_cache_view(const llama_kv_cache_view & view, int row_size = 80);
// Dump the KV cache view showing individual sequences in each cell (long output).
void dump_kv_cache_view_seqs(const llama_kv_cache_view & view, int row_size = 40);

94
llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/ggml-alloc.h Normal file
View file

@ -0,0 +1,94 @@
#pragma once
#include "ggml.h"
#ifdef __cplusplus
extern "C" {
#endif
struct ggml_backend;
struct ggml_backend_buffer;
struct ggml_backend_buffer_type;
//
// Legacy API
//
typedef struct ggml_allocr * ggml_allocr_t;
// initialize allocator for use with CPU backend only
GGML_API ggml_allocr_t ggml_allocr_new(void * data, size_t size, size_t alignment);
GGML_API ggml_allocr_t ggml_allocr_new_measure(size_t alignment);
// initialize allocator for use with ggml-backend
GGML_API ggml_allocr_t ggml_allocr_new_from_buffer(struct ggml_backend_buffer * buffer);
GGML_API ggml_allocr_t ggml_allocr_new_from_backend(struct ggml_backend * backend, size_t size); // allocates an owned buffer
GGML_API ggml_allocr_t ggml_allocr_new_measure_from_backend(struct ggml_backend * backend);
GGML_API struct ggml_backend_buffer * ggml_allocr_get_buffer(ggml_allocr_t alloc);
// tell the allocator to parse nodes following the order described in the list
// you should call this if your graph are optimized to execute out-of-order
GGML_API void ggml_allocr_set_parse_seq(ggml_allocr_t alloc, const int * list, int n);
GGML_API void ggml_allocr_free (ggml_allocr_t alloc);
GGML_API bool ggml_allocr_is_measure (ggml_allocr_t alloc);
GGML_API void ggml_allocr_reset (ggml_allocr_t alloc);
GGML_API void ggml_allocr_alloc (ggml_allocr_t alloc, struct ggml_tensor * tensor);
GGML_API size_t ggml_allocr_max_size (ggml_allocr_t alloc);
GGML_API size_t ggml_allocr_alloc_graph(ggml_allocr_t alloc, struct ggml_cgraph * graph);
//
// ggml-backend v2 API
//
// Separate tensor and graph allocator objects
// This is necessary for multi-backend allocation because the graph allocator needs to use multiple tensor allocators
// The original API is kept as a wrapper around the new API
// Tensor allocator
typedef struct ggml_tallocr * ggml_tallocr_t;
GGML_API ggml_tallocr_t ggml_tallocr_new(void * data, size_t size, size_t alignment);
GGML_API ggml_tallocr_t ggml_tallocr_new_measure(size_t alignment);
GGML_API ggml_tallocr_t ggml_tallocr_new_from_buft(struct ggml_backend_buffer_type * buft, size_t size);
GGML_API ggml_tallocr_t ggml_tallocr_new_from_backend(struct ggml_backend * backend, size_t size); // allocates an owned buffer
GGML_API ggml_tallocr_t ggml_tallocr_new_from_buffer(struct ggml_backend_buffer * buffer);
GGML_API ggml_tallocr_t ggml_tallocr_new_measure_from_buft(struct ggml_backend_buffer_type * buft);
GGML_API ggml_tallocr_t ggml_tallocr_new_measure_from_backend(struct ggml_backend * backend);
GGML_API struct ggml_backend_buffer * ggml_tallocr_get_buffer(ggml_tallocr_t talloc);
GGML_API void ggml_tallocr_free (ggml_tallocr_t talloc);
GGML_API bool ggml_tallocr_is_measure (ggml_tallocr_t talloc);
GGML_API void ggml_tallocr_reset (ggml_tallocr_t talloc);
GGML_API void ggml_tallocr_alloc (ggml_tallocr_t talloc, struct ggml_tensor * tensor);
GGML_API size_t ggml_tallocr_max_size (ggml_tallocr_t talloc);
// Graph allocator
typedef struct ggml_gallocr * ggml_gallocr_t;
GGML_API ggml_gallocr_t ggml_gallocr_new(void);
GGML_API void ggml_gallocr_free(ggml_gallocr_t galloc);
GGML_API void ggml_gallocr_set_parse_seq(ggml_gallocr_t galloc, const int * list, int n);
GGML_API size_t ggml_gallocr_alloc_graph(ggml_gallocr_t galloc, ggml_tallocr_t talloc, struct ggml_cgraph * graph);
// Allocate tensors from the allocators given by the hash table
GGML_API void ggml_gallocr_alloc_graph_n(
ggml_gallocr_t galloc,
struct ggml_cgraph * graph,
struct ggml_hash_set hash_set,
ggml_tallocr_t * hash_node_talloc);
// Utils
// Create a buffer and allocate all the tensors in a ggml_context
GGML_API struct ggml_backend_buffer * ggml_backend_alloc_ctx_tensors_from_buft(struct ggml_context * ctx, struct ggml_backend_buffer_type * buft);
GGML_API struct ggml_backend_buffer * ggml_backend_alloc_ctx_tensors(struct ggml_context * ctx, struct ggml_backend * backend);
#ifdef __cplusplus
}
#endif

209
llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/ggml-backend.h Normal file
View file

@ -0,0 +1,209 @@
#pragma once
#include "ggml.h"
#include "ggml-alloc.h"
#ifdef __cplusplus
extern "C" {
#endif
typedef struct ggml_backend_buffer_type * ggml_backend_buffer_type_t;
typedef struct ggml_backend_buffer * ggml_backend_buffer_t;
typedef struct ggml_backend * ggml_backend_t;
typedef void * ggml_backend_graph_plan_t;
//
// Backend buffer
//
// buffer type
GGML_API const char * ggml_backend_buft_name (ggml_backend_buffer_type_t buft);
GGML_API GGML_CALL ggml_backend_buffer_t ggml_backend_buft_alloc_buffer (ggml_backend_buffer_type_t buft, size_t size);
GGML_API size_t ggml_backend_buft_get_alignment (ggml_backend_buffer_type_t buft);
GGML_API GGML_CALL size_t ggml_backend_buft_get_alloc_size (ggml_backend_buffer_type_t buft, struct ggml_tensor * tensor);
GGML_API bool ggml_backend_buft_supports_backend(ggml_backend_buffer_type_t buft, ggml_backend_t backend);
GGML_API bool ggml_backend_buft_is_host (ggml_backend_buffer_type_t buft);
// buffer
enum ggml_backend_buffer_usage {
GGML_BACKEND_BUFFER_USAGE_ANY = 0,
GGML_BACKEND_BUFFER_USAGE_WEIGHTS = 1,
};
GGML_API const char * ggml_backend_buffer_name (ggml_backend_buffer_t buffer);
GGML_API void ggml_backend_buffer_free (ggml_backend_buffer_t buffer);
GGML_API void * ggml_backend_buffer_get_base (ggml_backend_buffer_t buffer);
GGML_API size_t ggml_backend_buffer_get_size (ggml_backend_buffer_t buffer);
GGML_API GGML_CALL void ggml_backend_buffer_init_tensor (ggml_backend_buffer_t buffer, struct ggml_tensor * tensor);
GGML_API size_t ggml_backend_buffer_get_alignment (ggml_backend_buffer_t buffer);
GGML_API size_t ggml_backend_buffer_get_alloc_size(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor);
GGML_API void ggml_backend_buffer_clear (ggml_backend_buffer_t buffer, uint8_t value);
GGML_API bool ggml_backend_buffer_is_host (ggml_backend_buffer_t buffer);
GGML_API void ggml_backend_buffer_set_usage (ggml_backend_buffer_t buffer, enum ggml_backend_buffer_usage usage);
GGML_API ggml_backend_buffer_type_t ggml_backend_buffer_get_type (ggml_backend_buffer_t buffer);
GGML_API void ggml_backend_buffer_reset (ggml_backend_buffer_t buffer);
//
// Backend
//
GGML_API const char * ggml_backend_name(ggml_backend_t backend);
GGML_API void ggml_backend_free(ggml_backend_t backend);
GGML_API ggml_backend_buffer_type_t ggml_backend_get_default_buffer_type(ggml_backend_t backend);
GGML_API ggml_backend_buffer_t ggml_backend_alloc_buffer(ggml_backend_t backend, size_t size);
GGML_API size_t ggml_backend_get_alignment(ggml_backend_t backend);
GGML_API void ggml_backend_tensor_set_async(ggml_backend_t backend, struct ggml_tensor * tensor, const void * data, size_t offset, size_t size);
GGML_API void ggml_backend_tensor_get_async(ggml_backend_t backend, const struct ggml_tensor * tensor, void * data, size_t offset, size_t size);
GGML_API GGML_CALL void ggml_backend_tensor_set( struct ggml_tensor * tensor, const void * data, size_t offset, size_t size);
GGML_API GGML_CALL void ggml_backend_tensor_get(const struct ggml_tensor * tensor, void * data, size_t offset, size_t size);
GGML_API void ggml_backend_synchronize(ggml_backend_t backend);
GGML_API ggml_backend_graph_plan_t ggml_backend_graph_plan_create (ggml_backend_t backend, struct ggml_cgraph * cgraph);
GGML_API void ggml_backend_graph_plan_free (ggml_backend_t backend, ggml_backend_graph_plan_t plan);
GGML_API void ggml_backend_graph_plan_compute(ggml_backend_t backend, ggml_backend_graph_plan_t plan);
GGML_API bool ggml_backend_graph_compute (ggml_backend_t backend, struct ggml_cgraph * cgraph);
GGML_API bool ggml_backend_supports_op (ggml_backend_t backend, const struct ggml_tensor * op);
// tensor copy between different backends
GGML_API void ggml_backend_tensor_copy(struct ggml_tensor * src, struct ggml_tensor * dst);
GGML_API void ggml_backend_tensor_copy_async(ggml_backend_t backend, struct ggml_tensor * src, struct ggml_tensor * dst); // automatic fallback to sync copy
//
// CPU backend
//
GGML_API ggml_backend_t ggml_backend_cpu_init(void);
GGML_API GGML_CALL bool ggml_backend_is_cpu (ggml_backend_t backend);
GGML_API void ggml_backend_cpu_set_n_threads(ggml_backend_t backend_cpu, int n_threads);
// Create a backend buffer from an existing pointer
GGML_API GGML_CALL ggml_backend_buffer_t ggml_backend_cpu_buffer_from_ptr(void * ptr, size_t size);
GGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_cpu_buffer_type(void);
#ifdef GGML_USE_CPU_HBM
GGML_API ggml_backend_buffer_type_t ggml_backend_cpu_hbm_buffer_type(void);
#endif
//
// Backend registry
//
// The backend registry is a registry of all the available backends, and allows initializing backends in a generic way
GGML_API size_t ggml_backend_reg_get_count(void);
GGML_API size_t ggml_backend_reg_find_by_name(const char * name);
GGML_API ggml_backend_t ggml_backend_reg_init_backend_from_str(const char * backend_str); // str is name[:params]
GGML_API const char * ggml_backend_reg_get_name(size_t i);
GGML_API ggml_backend_t ggml_backend_reg_init_backend(size_t i, const char * params); // params is backend-specific
GGML_API ggml_backend_buffer_type_t ggml_backend_reg_get_default_buffer_type(size_t i);
GGML_API ggml_backend_buffer_t ggml_backend_reg_alloc_buffer(size_t i, size_t size);
//
// Backend scheduler
//
// The backend scheduler allows for multiple backends to be used together
// Handles compute buffer allocation, assignment of tensors to backends, and copying of tensors between backends
// The backends are selected based on:
// - the backend that supports the operation
// - the location of the pre-allocated tensors (e.g. the weights)
/*
Example usage:
sched = ggml_backend_sched_new({backend_gpu, backend_gpu2, backend_cpu}, num_backends);
// sched is initialized with measure allocators and cannot be used until allocated with a measure graph
// initialize buffers from a measure graph
measure_graph = build_graph(sched); // use the allocr to allocate inputs as needed
// in build_graph:
build_graph(...) {
// allocating tensors in a specific backend (optional, recommended: pre-allocate inputs in a different buffer)
alloc_cpu = ggml_backend_sched_get_allocr(sched, backend_cpu);
ggml_allocr_alloc(alloc_cpu, tensor);
// manually assigning nodes to a backend (optional, shouldn't be needed in most cases)
struct ggml_tensor * node = ggml_mul_mat(ctx, ...);
ggml_backend_sched_set_node_backend(sched, node, backend_gpu);
}
// allocate backend buffers from measure graph
ggml_backend_sched_init_measure(sched, measure_graph);
// the scheduler is now ready to compute graphs
// compute
graph = build_graph(sched);
ggml_backend_sched_graph_compute(sched, graph);
*/
struct ggml_backend_sched;
typedef struct ggml_backend_sched * ggml_backend_sched_t;
// when ask == true, the scheduler wants to know if the user wants to observe this node
// this allows the scheduler to batch nodes together in order to evaluate them in a single call
//
// when ask == false, the scheduler is passing the node tensor to the user for observation
// if the user returns false, the scheduler will cancel the graph compute
//
typedef bool (*ggml_backend_sched_eval_callback)(struct ggml_tensor * t, bool ask, void * user_data);
// Initialize a backend scheduler
GGML_API ggml_backend_sched_t ggml_backend_sched_new(ggml_backend_t * backends, ggml_backend_buffer_type_t * bufts, int n_backends, size_t graph_size);
GGML_API void ggml_backend_sched_free(ggml_backend_sched_t sched);
// Initialize backend buffers from a measure graph
GGML_API void ggml_backend_sched_init_measure(ggml_backend_sched_t sched, struct ggml_cgraph * measure_graph);
// Get the number of splits of the last graph
GGML_API int ggml_backend_sched_get_n_splits(ggml_backend_sched_t sched);
GGML_API ggml_tallocr_t ggml_backend_sched_get_tallocr(ggml_backend_sched_t sched, ggml_backend_t backend);
GGML_API ggml_backend_buffer_t ggml_backend_sched_get_buffer (ggml_backend_sched_t sched, ggml_backend_t backend);
GGML_API void ggml_backend_sched_set_node_backend(ggml_backend_sched_t sched, struct ggml_tensor * node, ggml_backend_t backend);
GGML_API ggml_backend_t ggml_backend_sched_get_node_backend(ggml_backend_sched_t sched, struct ggml_tensor * node);
// Allocate and compute graph on the backend scheduler
GGML_API void ggml_backend_sched_graph_compute(ggml_backend_sched_t sched, struct ggml_cgraph * graph);
// Reset all assignments and allocators - must be called before using the sched allocators to allocate inputs
GGML_API void ggml_backend_sched_reset(ggml_backend_sched_t sched);
// Set a callback to be called for each resulting node during graph compute
GGML_API void ggml_backend_sched_set_eval_callback(ggml_backend_sched_t sched, ggml_backend_sched_eval_callback callback, void * user_data);
//
// Utils
//
struct ggml_backend_graph_copy {
ggml_backend_buffer_t buffer;
struct ggml_context * ctx_allocated;
struct ggml_context * ctx_unallocated;
struct ggml_cgraph * graph;
};
// Copy a graph to a different backend
GGML_API struct ggml_backend_graph_copy ggml_backend_graph_copy(ggml_backend_t backend, struct ggml_cgraph * graph);
GGML_API void ggml_backend_graph_copy_free(struct ggml_backend_graph_copy copy);
typedef bool (*GGML_CALL ggml_backend_eval_callback)(int node_index, struct ggml_tensor * t1, struct ggml_tensor * t2, void * user_data);
// Compare the output of two backends
GGML_API bool ggml_backend_compare_graph_backend(ggml_backend_t backend1, ggml_backend_t backend2, struct ggml_cgraph * graph, ggml_backend_eval_callback callback, void * user_data);
// Tensor initialization
GGML_API void ggml_backend_tensor_alloc(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor, void * addr);
GGML_API void ggml_backend_view_init(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor);
#ifdef __cplusplus
}
#endif

2301
llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/ggml.h Normal file
View file

File diff suppressed because it is too large Load diff

29
llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/grammar-parser.h Normal file
View file

@ -0,0 +1,29 @@
// Implements a parser for an extended Backus-Naur form (BNF), producing the
// binary context-free grammar format specified by llama.h. Supports character
// ranges, grouping, and repetition operators. As an example, a grammar for
// arithmetic might look like:
//
// root ::= expr
// expr ::= term ([-+*/] term)*
// term ::= num | "(" space expr ")" space
// num ::= [0-9]+ space
// space ::= [ \t\n]*
#pragma once
#include "llama.h"
#include <vector>
#include <map>
#include <cstdint>
#include <string>
namespace grammar_parser {
struct parse_state {
std::map<std::string, uint32_t> symbol_ids;
std::vector<std::vector<llama_grammar_element>> rules;
std::vector<const llama_grammar_element *> c_rules();
};
parse_state parse(const char * src);
void print_grammar(FILE * file, const parse_state & state);
}

915
llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/llama.h Normal file
View file

@ -0,0 +1,915 @@
#ifndef LLAMA_H
#define LLAMA_H
#include "ggml.h"
#include "ggml-backend.h"
#ifdef GGML_USE_CUBLAS
#include "ggml-cuda.h"
#define LLAMA_MAX_DEVICES GGML_CUDA_MAX_DEVICES
#else
#define LLAMA_MAX_DEVICES 1
#endif // GGML_USE_CUBLAS
#include <stddef.h>
#include <stdint.h>
#include <stdio.h>
#include <stdbool.h>
#ifdef LLAMA_SHARED
# if defined(_WIN32) && !defined(__MINGW32__)
# ifdef LLAMA_BUILD
# define LLAMA_API __declspec(dllexport)
# else
# define LLAMA_API __declspec(dllimport)
# endif
# else
# define LLAMA_API __attribute__ ((visibility ("default")))
# endif
#else
# define LLAMA_API
#endif
#ifdef __GNUC__
# define DEPRECATED(func, hint) func __attribute__((deprecated(hint)))
#elif defined(_MSC_VER)
# define DEPRECATED(func, hint) __declspec(deprecated(hint)) func
#else
# define DEPRECATED(func, hint) func
#endif
#define LLAMA_DEFAULT_SEED 0xFFFFFFFF
#define LLAMA_MAX_RNG_STATE (64*1024)
#define LLAMA_FILE_MAGIC_GGLA 0x67676c61u // 'ggla'
#define LLAMA_FILE_MAGIC_GGSN 0x6767736eu // 'ggsn'
#define LLAMA_SESSION_MAGIC LLAMA_FILE_MAGIC_GGSN
#define LLAMA_SESSION_VERSION 4
#if defined(GGML_USE_CUBLAS) || defined(GGML_USE_CLBLAST) || defined(GGML_USE_METAL)
// Defined when llama.cpp is compiled with support for offloading model layers to GPU.
#define LLAMA_SUPPORTS_GPU_OFFLOAD
#endif
#ifdef __cplusplus
extern "C" {
#endif
//
// C interface
//
// TODO: show sample usage
//
struct llama_model;
struct llama_context;
typedef int32_t llama_pos;
typedef int32_t llama_token;
typedef int32_t llama_seq_id;
enum llama_vocab_type {
LLAMA_VOCAB_TYPE_SPM = 0, // SentencePiece
LLAMA_VOCAB_TYPE_BPE = 1, // Byte Pair Encoding
};
enum llama_token_type {
LLAMA_TOKEN_TYPE_UNDEFINED = 0,
LLAMA_TOKEN_TYPE_NORMAL = 1,
LLAMA_TOKEN_TYPE_UNKNOWN = 2,
LLAMA_TOKEN_TYPE_CONTROL = 3,
LLAMA_TOKEN_TYPE_USER_DEFINED = 4,
LLAMA_TOKEN_TYPE_UNUSED = 5,
LLAMA_TOKEN_TYPE_BYTE = 6,
};
// model file types
enum llama_ftype {
LLAMA_FTYPE_ALL_F32 = 0,
LLAMA_FTYPE_MOSTLY_F16 = 1, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q4_0 = 2, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q4_1 = 3, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q4_1_SOME_F16 = 4, // tok_embeddings.weight and output.weight are F16
// LLAMA_FTYPE_MOSTLY_Q4_2 = 5, // support has been removed
// LLAMA_FTYPE_MOSTLY_Q4_3 = 6, // support has been removed
LLAMA_FTYPE_MOSTLY_Q8_0 = 7, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q5_0 = 8, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q5_1 = 9, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q2_K = 10, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q3_K_S = 11, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q3_K_M = 12, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q3_K_L = 13, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q4_K_S = 14, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q4_K_M = 15, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q5_K_S = 16, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q5_K_M = 17, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q6_K = 18, // except 1d tensors
LLAMA_FTYPE_MOSTLY_IQ2_XXS = 19, // except 1d tensors
LLAMA_FTYPE_MOSTLY_IQ2_XS = 20, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q2_K_S = 21, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q3_K_XS = 22, // except 1d tensors
LLAMA_FTYPE_GUESSED = 1024, // not specified in the model file
};
enum llama_rope_scaling_type {
LLAMA_ROPE_SCALING_UNSPECIFIED = -1,
LLAMA_ROPE_SCALING_NONE = 0,
LLAMA_ROPE_SCALING_LINEAR = 1,
LLAMA_ROPE_SCALING_YARN = 2,
LLAMA_ROPE_SCALING_MAX_VALUE = LLAMA_ROPE_SCALING_YARN,
};
enum llama_split_mode {
LLAMA_SPLIT_NONE = 0, // single GPU
LLAMA_SPLIT_LAYER = 1, // split layers and KV across GPUs
LLAMA_SPLIT_ROW = 2, // split rows across GPUs
};
typedef struct llama_token_data {
llama_token id; // token id
float logit; // log-odds of the token
float p; // probability of the token
} llama_token_data;
typedef struct llama_token_data_array {
llama_token_data * data;
size_t size;
bool sorted;
} llama_token_data_array;
typedef bool (*llama_progress_callback)(float progress, void *ctx);
// Input data for llama_decode
// A llama_batch object can contain input about one or many sequences
// The provided arrays (i.e. token, embd, pos, etc.) must have size of n_tokens
//
// - token : the token ids of the input (used when embd is NULL)
// - embd : token embeddings (i.e. float vector of size n_embd) (used when token is NULL)
// - pos : the positions of the respective token in the sequence
// - seq_id : the sequence to which the respective token belongs
// - logits : if zero, the logits for the respective token will not be output
//
typedef struct llama_batch {
int32_t n_tokens;
llama_token * token;
float * embd;
llama_pos * pos;
int32_t * n_seq_id;
llama_seq_id ** seq_id;
int8_t * logits;
// NOTE: helpers for smooth API transition - can be deprecated in the future
// for future-proof code, use the above fields instead and ignore everything below
//
// pos[i] = all_pos_0 + i*all_pos_1
//
llama_pos all_pos_0; // used if pos == NULL
llama_pos all_pos_1; // used if pos == NULL
llama_seq_id all_seq_id; // used if seq_id == NULL
} llama_batch;
enum llama_model_kv_override_type {
LLAMA_KV_OVERRIDE_INT,
LLAMA_KV_OVERRIDE_FLOAT,
LLAMA_KV_OVERRIDE_BOOL,
};
struct llama_model_kv_override {
char key[128];
enum llama_model_kv_override_type tag;
union {
int64_t int_value;
double float_value;
bool bool_value;
};
};
struct llama_model_params {
int32_t n_gpu_layers; // number of layers to store in VRAM
enum llama_split_mode split_mode; // how to split the model across multiple GPUs
// main_gpu interpretation depends on split_mode:
// LLAMA_SPLIT_NONE: the GPU that is used for the entire model
// LLAMA_SPLIT_ROW: the GPU that is used for small tensors and intermediate results
// LLAMA_SPLIT_LAYER: ignored
int32_t main_gpu;
// proportion of the model (layers or rows) to offload to each GPU, size: LLAMA_MAX_DEVICES
const float * tensor_split;
// Called with a progress value between 0.0 and 1.0. Pass NULL to disable.
// If the provided progress_callback returns true, model loading continues.
// If it returns false, model loading is immediately aborted.
llama_progress_callback progress_callback;
// context pointer passed to the progress callback
void * progress_callback_user_data;
// override key-value pairs of the model meta data
const struct llama_model_kv_override * kv_overrides;
// Keep the booleans together to avoid misalignment during copy-by-value.
bool vocab_only; // only load the vocabulary, no weights
bool use_mmap; // use mmap if possible
bool use_mlock; // force system to keep model in RAM
};
struct llama_context_params {
uint32_t seed; // RNG seed, -1 for random
uint32_t n_ctx; // text context, 0 = from model
uint32_t n_batch; // prompt processing maximum batch size
uint32_t n_threads; // number of threads to use for generation
uint32_t n_threads_batch; // number of threads to use for batch processing
int8_t rope_scaling_type; // RoPE scaling type, from `enum llama_rope_scaling_type`
// ref: https://github.com/ggerganov/llama.cpp/pull/2054
float rope_freq_base; // RoPE base frequency, 0 = from model
float rope_freq_scale; // RoPE frequency scaling factor, 0 = from model
float yarn_ext_factor; // YaRN extrapolation mix factor, negative = from model
float yarn_attn_factor; // YaRN magnitude scaling factor
float yarn_beta_fast; // YaRN low correction dim
float yarn_beta_slow; // YaRN high correction dim
uint32_t yarn_orig_ctx; // YaRN original context size
ggml_backend_sched_eval_callback cb_eval;
void * cb_eval_user_data;
enum ggml_type type_k; // data type for K cache
enum ggml_type type_v; // data type for V cache
// Keep the booleans together to avoid misalignment during copy-by-value.
bool mul_mat_q; // if true, use experimental mul_mat_q kernels (DEPRECATED - always true)
bool logits_all; // the llama_eval() call computes all logits, not just the last one (DEPRECATED - set llama_batch.logits instead)
bool embedding; // embedding mode only
bool offload_kqv; // whether to offload the KQV ops (including the KV cache) to GPU
};
// model quantization parameters
typedef struct llama_model_quantize_params {
int32_t nthread; // number of threads to use for quantizing, if <=0 will use std::thread::hardware_concurrency()
enum llama_ftype ftype; // quantize to this llama_ftype
bool allow_requantize; // allow quantizing non-f32/f16 tensors
bool quantize_output_tensor; // quantize output.weight
bool only_copy; // only copy tensors - ftype, allow_requantize and quantize_output_tensor are ignored
bool pure; // disable k-quant mixtures and quantize all tensors to the same type
void * imatrix; // pointer to importance matrix data
} llama_model_quantize_params;
// grammar types
struct llama_grammar;
// grammar element type
enum llama_gretype {
// end of rule definition
LLAMA_GRETYPE_END = 0,
// start of alternate definition for rule
LLAMA_GRETYPE_ALT = 1,
// non-terminal element: reference to rule
LLAMA_GRETYPE_RULE_REF = 2,
// terminal element: character (code point)
LLAMA_GRETYPE_CHAR = 3,
// inverse char(s) ([^a], [^a-b] [^abc])
LLAMA_GRETYPE_CHAR_NOT = 4,
// modifies a preceding LLAMA_GRETYPE_CHAR or LLAMA_GRETYPE_CHAR_ALT to
// be an inclusive range ([a-z])
LLAMA_GRETYPE_CHAR_RNG_UPPER = 5,
// modifies a preceding LLAMA_GRETYPE_CHAR or
// LLAMA_GRETYPE_CHAR_RNG_UPPER to add an alternate char to match ([ab], [a-zA])
LLAMA_GRETYPE_CHAR_ALT = 6,
};
typedef struct llama_grammar_element {
enum llama_gretype type;
uint32_t value; // Unicode code point or rule ID
} llama_grammar_element;
// performance timing information
struct llama_timings {
double t_start_ms;
double t_end_ms;
double t_load_ms;
double t_sample_ms;
double t_p_eval_ms;
double t_eval_ms;
int32_t n_sample;
int32_t n_p_eval;
int32_t n_eval;
};
// Helpers for getting default parameters
LLAMA_API struct llama_model_params llama_model_default_params(void);
LLAMA_API struct llama_context_params llama_context_default_params(void);
LLAMA_API struct llama_model_quantize_params llama_model_quantize_default_params(void);
// Initialize the llama + ggml backend
// If numa is true, use NUMA optimizations
// Call once at the start of the program
LLAMA_API void llama_backend_init(bool numa);
// Call once at the end of the program - currently only used for MPI
LLAMA_API void llama_backend_free(void);
LLAMA_API struct llama_model * llama_load_model_from_file(
const char * path_model,
struct llama_model_params params);
LLAMA_API void llama_free_model(struct llama_model * model);
LLAMA_API struct llama_context * llama_new_context_with_model(
struct llama_model * model,
struct llama_context_params params);
// Frees all allocated memory
LLAMA_API void llama_free(struct llama_context * ctx);
LLAMA_API int64_t llama_time_us(void);
LLAMA_API int32_t llama_max_devices(void);
LLAMA_API bool llama_mmap_supported (void);
LLAMA_API bool llama_mlock_supported(void);
LLAMA_API const struct llama_model * llama_get_model(const struct llama_context * ctx);
LLAMA_API uint32_t llama_n_ctx (const struct llama_context * ctx);
LLAMA_API uint32_t llama_n_batch (const struct llama_context * ctx);
LLAMA_API enum llama_vocab_type llama_vocab_type(const struct llama_model * model);
LLAMA_API int32_t llama_n_vocab (const struct llama_model * model);
LLAMA_API int32_t llama_n_ctx_train(const struct llama_model * model);
LLAMA_API int32_t llama_n_embd (const struct llama_model * model);
// Get the model's RoPE frequency scaling factor
LLAMA_API float llama_rope_freq_scale_train(const struct llama_model * model);
// Functions to access the model's GGUF metadata scalar values
// - The functions return the length of the string on success, or -1 on failure
// - The output string is always null-terminated and cleared on failure
// - GGUF array values are not supported by these functions
// Get metadata value as a string by key name
LLAMA_API int32_t llama_model_meta_val_str(const struct llama_model * model, const char * key, char * buf, size_t buf_size);
// Get the number of metadata key/value pairs
LLAMA_API int32_t llama_model_meta_count(const struct llama_model * model);
// Get metadata key name by index
LLAMA_API int32_t llama_model_meta_key_by_index(const struct llama_model * model, int32_t i, char * buf, size_t buf_size);
// Get metadata value as a string by index
LLAMA_API int32_t llama_model_meta_val_str_by_index(const struct llama_model * model, int32_t i, char * buf, size_t buf_size);
// Get a string describing the model type
LLAMA_API int32_t llama_model_desc(const struct llama_model * model, char * buf, size_t buf_size);
// Returns the total size of all the tensors in the model in bytes
LLAMA_API uint64_t llama_model_size(const struct llama_model * model);
// Returns the total number of parameters in the model
LLAMA_API uint64_t llama_model_n_params(const struct llama_model * model);
// Get a llama model tensor
LLAMA_API struct ggml_tensor * llama_get_model_tensor(struct llama_model * model, const char * name);
// Returns 0 on success
LLAMA_API uint32_t llama_model_quantize(
const char * fname_inp,
const char * fname_out,
const llama_model_quantize_params * params);
// Apply a LoRA adapter to a loaded model
// path_base_model is the path to a higher quality model to use as a base for
// the layers modified by the adapter. Can be NULL to use the current loaded model.
// The model needs to be reloaded before applying a new adapter, otherwise the adapter
// will be applied on top of the previous one
// Returns 0 on success
LLAMA_API DEPRECATED(int32_t llama_apply_lora_from_file(
struct llama_context * ctx,
const char * path_lora,
float scale,
const char * path_base_model,
int32_t n_threads),
"use llama_model_apply_lora_from_file instead");
LLAMA_API int32_t llama_model_apply_lora_from_file(
const struct llama_model * model,
const char * path_lora,
float scale,
const char * path_base_model,
int32_t n_threads);
//
// KV cache
//
// Information associated with an individual cell in the KV cache view.
struct llama_kv_cache_view_cell {
// The position for this cell. Takes KV cache shifts into account.
// May be negative if the cell is not populated.
llama_pos pos;
};
// An updateable view of the KV cache.
struct llama_kv_cache_view {
// Number of KV cache cells. This will be the same as the context size.
int32_t n_cells;
// Maximum number of sequences that can exist in a cell. It's not an error
// if there are more sequences in a cell than this value, however they will
// not be visible in the view cells_sequences.
int32_t n_max_seq;
// Number of tokens in the cache. For example, if there are two populated
// cells, the first with 1 sequence id in it and the second with 2 sequence
// ids then you'll have 3 tokens.
int32_t token_count;
// Number of populated cache cells.
int32_t used_cells;
// Maximum contiguous empty slots in the cache.
int32_t max_contiguous;
// Index to the start of the max_contiguous slot range. Can be negative
// when cache is full.
int32_t max_contiguous_idx;
// Information for an individual cell.
struct llama_kv_cache_view_cell * cells;
// The sequences for each cell. There will be n_max_seq items per cell.
llama_seq_id * cells_sequences;
};
// Create an empty KV cache view. (use only for debugging purposes)
LLAMA_API struct llama_kv_cache_view llama_kv_cache_view_init(const struct llama_context * ctx, int32_t n_max_seq);
// Free a KV cache view. (use only for debugging purposes)
LLAMA_API void llama_kv_cache_view_free(struct llama_kv_cache_view * view);
// Update the KV cache view structure with the current state of the KV cache. (use only for debugging purposes)
LLAMA_API void llama_kv_cache_view_update(const struct llama_context * ctx, struct llama_kv_cache_view * view);
// Returns the number of tokens in the KV cache (slow, use only for debug)
// If a KV cell has multiple sequences assigned to it, it will be counted multiple times
LLAMA_API int32_t llama_get_kv_cache_token_count(const struct llama_context * ctx);
// Returns the number of used KV cells (i.e. have at least one sequence assigned to them)
LLAMA_API int32_t llama_get_kv_cache_used_cells(const struct llama_context * ctx);
// Clear the KV cache
LLAMA_API void llama_kv_cache_clear(
struct llama_context * ctx);
// Removes all tokens that belong to the specified sequence and have positions in [p0, p1)
// seq_id < 0 : match any sequence
// p0 < 0 : [0, p1]
// p1 < 0 : [p0, inf)
LLAMA_API void llama_kv_cache_seq_rm(
struct llama_context * ctx,
llama_seq_id seq_id,
llama_pos p0,
llama_pos p1);
// Copy all tokens that belong to the specified sequence to another sequence
// Note that this does not allocate extra KV cache memory - it simply assigns the tokens to the new sequence
// p0 < 0 : [0, p1]
// p1 < 0 : [p0, inf)
LLAMA_API void llama_kv_cache_seq_cp(
struct llama_context * ctx,
llama_seq_id seq_id_src,
llama_seq_id seq_id_dst,
llama_pos p0,
llama_pos p1);
// Removes all tokens that do not belong to the specified sequence
LLAMA_API void llama_kv_cache_seq_keep(
struct llama_context * ctx,
llama_seq_id seq_id);
// Adds relative position "delta" to all tokens that belong to the specified sequence and have positions in [p0, p1)
// If the KV cache is RoPEd, the KV data is updated accordingly
// p0 < 0 : [0, p1]
// p1 < 0 : [p0, inf)
LLAMA_API void llama_kv_cache_seq_shift(
struct llama_context * ctx,
llama_seq_id seq_id,
llama_pos p0,
llama_pos p1,
llama_pos delta);
// Integer division of the positions by factor of `d > 1`
// If the KV cache is RoPEd, the KV data is updated accordingly
// p0 < 0 : [0, p1]
// p1 < 0 : [p0, inf)
LLAMA_API void llama_kv_cache_seq_div(
struct llama_context * ctx,
llama_seq_id seq_id,
llama_pos p0,
llama_pos p1,
int d);
//
// State / sessions
//
// Returns the maximum size in bytes of the state (rng, logits, embedding
// and kv_cache) - will often be smaller after compacting tokens
LLAMA_API size_t llama_get_state_size(const struct llama_context * ctx);
// Copies the state to the specified destination address.
// Destination needs to have allocated enough memory.
// Returns the number of bytes copied
LLAMA_API size_t llama_copy_state_data(
struct llama_context * ctx,
uint8_t * dst);
// Set the state reading from the specified address
// Returns the number of bytes read
LLAMA_API size_t llama_set_state_data(
struct llama_context * ctx,
uint8_t * src);
// Save/load session file
LLAMA_API bool llama_load_session_file(
struct llama_context * ctx,
const char * path_session,
llama_token * tokens_out,
size_t n_token_capacity,
size_t * n_token_count_out);
LLAMA_API bool llama_save_session_file(
struct llama_context * ctx,
const char * path_session,
const llama_token * tokens,
size_t n_token_count);
//
// Decoding
//
// Run the llama inference to obtain the logits and probabilities for the next token(s).
// tokens + n_tokens is the provided batch of new tokens to process
// n_past is the number of tokens to use from previous eval calls
// Returns 0 on success
// DEPRECATED: use llama_decode() instead
LLAMA_API DEPRECATED(int llama_eval(
struct llama_context * ctx,
llama_token * tokens,
int32_t n_tokens,
int32_t n_past),
"use llama_decode() instead");
// Same as llama_eval, but use float matrix input directly.
// DEPRECATED: use llama_decode() instead
LLAMA_API DEPRECATED(int llama_eval_embd(
struct llama_context * ctx,
float * embd,
int32_t n_tokens,
int32_t n_past),
"use llama_decode() instead");
// Return batch for single sequence of tokens starting at pos_0
//
// NOTE: this is a helper function to facilitate transition to the new batch API - avoid using it
//
LLAMA_API struct llama_batch llama_batch_get_one(
llama_token * tokens,
int32_t n_tokens,
llama_pos pos_0,
llama_seq_id seq_id);
// Allocates a batch of tokens on the heap that can hold a maximum of n_tokens
// Each token can be assigned up to n_seq_max sequence ids
// The batch has to be freed with llama_batch_free()
// If embd != 0, llama_batch.embd will be allocated with size of n_tokens * embd * sizeof(float)
// Otherwise, llama_batch.token will be allocated to store n_tokens llama_token
// The rest of the llama_batch members are allocated with size n_tokens
// All members are left uninitialized
LLAMA_API struct llama_batch llama_batch_init(
int32_t n_tokens,
int32_t embd,
int32_t n_seq_max);
// Frees a batch of tokens allocated with llama_batch_init()
LLAMA_API void llama_batch_free(struct llama_batch batch);
// Positive return values does not mean a fatal error, but rather a warning.
// 0 - success
// 1 - could not find a KV slot for the batch (try reducing the size of the batch or increase the context)
// < 0 - error
LLAMA_API int32_t llama_decode(
struct llama_context * ctx,
struct llama_batch batch);
// Set the number of threads used for decoding
// n_threads is the number of threads used for generation (single token)
// n_threads_batch is the number of threads used for prompt and batch processing (multiple tokens)
LLAMA_API void llama_set_n_threads(struct llama_context * ctx, uint32_t n_threads, uint32_t n_threads_batch);
// Token logits obtained from the last call to llama_eval()
// The logits for the last token are stored in the last row
// Logits for which llama_batch.logits[i] == 0 are undefined
// Rows: n_tokens provided with llama_batch
// Cols: n_vocab
LLAMA_API float * llama_get_logits(struct llama_context * ctx);
// Logits for the ith token. Equivalent to:
// llama_get_logits(ctx) + i*n_vocab
LLAMA_API float * llama_get_logits_ith(struct llama_context * ctx, int32_t i);
// Get the embeddings for the input
// shape: [n_embd] (1-dimensional)
LLAMA_API float * llama_get_embeddings(struct llama_context * ctx);
//
// Vocab
//
LLAMA_API const char * llama_token_get_text(const struct llama_model * model, llama_token token);
LLAMA_API float llama_token_get_score(const struct llama_model * model, llama_token token);
LLAMA_API enum llama_token_type llama_token_get_type(const struct llama_model * model, llama_token token);
// Special tokens
LLAMA_API llama_token llama_token_bos(const struct llama_model * model); // beginning-of-sentence
LLAMA_API llama_token llama_token_eos(const struct llama_model * model); // end-of-sentence
LLAMA_API llama_token llama_token_nl (const struct llama_model * model); // next-line
// Returns -1 if unknown, 1 for true or 0 for false.
LLAMA_API int32_t llama_add_bos_token(const struct llama_model * model);
// Returns -1 if unknown, 1 for true or 0 for false.
LLAMA_API int32_t llama_add_eos_token(const struct llama_model * model);
// codellama infill tokens
LLAMA_API llama_token llama_token_prefix(const struct llama_model * model); // Beginning of infill prefix
LLAMA_API llama_token llama_token_middle(const struct llama_model * model); // Beginning of infill middle
LLAMA_API llama_token llama_token_suffix(const struct llama_model * model); // Beginning of infill suffix
LLAMA_API llama_token llama_token_eot (const struct llama_model * model); // End of infill middle
//
// Tokenization
//
/// @details Convert the provided text into tokens.
/// @param tokens The tokens pointer must be large enough to hold the resulting tokens.
/// @return Returns the number of tokens on success, no more than n_max_tokens
/// @return Returns a negative number on failure - the number of tokens that would have been returned
/// @param special Allow tokenizing special and/or control tokens which otherwise are not exposed and treated as plaintext.
/// Does not insert a leading space.
LLAMA_API int32_t llama_tokenize(
const struct llama_model * model,
const char * text,
int32_t text_len,
llama_token * tokens,
int32_t n_max_tokens,
bool add_bos,
bool special);
// Token Id -> Piece.
// Uses the vocabulary in the provided context.
// Does not write null terminator to the buffer.
// User code is responsible to remove the leading whitespace of the first non-BOS token when decoding multiple tokens.
LLAMA_API int32_t llama_token_to_piece(
const struct llama_model * model,
llama_token token,
char * buf,
int32_t length);
//
// Grammar
//
LLAMA_API struct llama_grammar * llama_grammar_init(
const llama_grammar_element ** rules,
size_t n_rules,
size_t start_rule_index);
LLAMA_API void llama_grammar_free(struct llama_grammar * grammar);
LLAMA_API struct llama_grammar * llama_grammar_copy(const struct llama_grammar * grammar);
//
// Sampling functions
//
// Sets the current rng seed.
LLAMA_API void llama_set_rng_seed(struct llama_context * ctx, uint32_t seed);
/// @details Repetition penalty described in CTRL academic paper https://arxiv.org/abs/1909.05858, with negative logit fix.
/// @details Frequency and presence penalties described in OpenAI API https://platform.openai.com/docs/api-reference/parameter-details.
LLAMA_API void llama_sample_repetition_penalties(
struct llama_context * ctx,
llama_token_data_array * candidates,
const llama_token * last_tokens,
size_t penalty_last_n,
float penalty_repeat,
float penalty_freq,
float penalty_present);
/// @details Apply classifier-free guidance to the logits as described in academic paper "Stay on topic with Classifier-Free Guidance" https://arxiv.org/abs/2306.17806
/// @param logits Logits extracted from the original generation context.
/// @param logits_guidance Logits extracted from a separate context from the same model. Other than a negative prompt at the beginning, it should have all generated and user input tokens copied from the main context.
/// @param scale Guidance strength. 1.0f means no guidance. Higher values mean stronger guidance.
LLAMA_API void llama_sample_apply_guidance(
struct llama_context * ctx,
float * logits,
float * logits_guidance,
float scale);
LLAMA_API DEPRECATED(void llama_sample_classifier_free_guidance(
struct llama_context * ctx,
llama_token_data_array * candidates,
struct llama_context * guidance_ctx,
float scale),
"use llama_sample_apply_guidance() instead");
/// @details Sorts candidate tokens by their logits in descending order and calculate probabilities based on logits.
LLAMA_API void llama_sample_softmax(
struct llama_context * ctx,
llama_token_data_array * candidates);
/// @details Top-K sampling described in academic paper "The Curious Case of Neural Text Degeneration" https://arxiv.org/abs/1904.09751
LLAMA_API void llama_sample_top_k(
struct llama_context * ctx,
llama_token_data_array * candidates,
int32_t k,
size_t min_keep);
/// @details Nucleus sampling described in academic paper "The Curious Case of Neural Text Degeneration" https://arxiv.org/abs/1904.09751
LLAMA_API void llama_sample_top_p(
struct llama_context * ctx,
llama_token_data_array * candidates,
float p,
size_t min_keep);
/// @details Minimum P sampling as described in https://github.com/ggerganov/llama.cpp/pull/3841
LLAMA_API void llama_sample_min_p(
struct llama_context * ctx,
llama_token_data_array * candidates,
float p,
size_t min_keep);
/// @details Tail Free Sampling described in https://www.trentonbricken.com/Tail-Free-Sampling/.
LLAMA_API void llama_sample_tail_free(
struct llama_context * ctx,
llama_token_data_array * candidates,
float z,
size_t min_keep);
/// @details Locally Typical Sampling implementation described in the paper https://arxiv.org/abs/2202.00666.
LLAMA_API void llama_sample_typical(
struct llama_context * ctx,
llama_token_data_array * candidates,
float p,
size_t min_keep);
LLAMA_API void llama_sample_temp(
struct llama_context * ctx,
llama_token_data_array * candidates,
float temp);
LLAMA_API DEPRECATED(void llama_sample_temperature(
struct llama_context * ctx,
llama_token_data_array * candidates,
float temp),
"use llama_sample_temp instead");
/// @details Apply constraints from grammar
LLAMA_API void llama_sample_grammar(
struct llama_context * ctx,
llama_token_data_array * candidates,
const struct llama_grammar * grammar);
/// @details Mirostat 1.0 algorithm described in the paper https://arxiv.org/abs/2007.14966. Uses tokens instead of words.
/// @param candidates A vector of `llama_token_data` containing the candidate tokens, their probabilities (p), and log-odds (logit) for the current position in the generated text.
/// @param tau The target cross-entropy (or surprise) value you want to achieve for the generated text. A higher value corresponds to more surprising or less predictable text, while a lower value corresponds to less surprising or more predictable text.
/// @param eta The learning rate used to update `mu` based on the error between the target and observed surprisal of the sampled word. A larger learning rate will cause `mu` to be updated more quickly, while a smaller learning rate will result in slower updates.
/// @param m The number of tokens considered in the estimation of `s_hat`. This is an arbitrary value that is used to calculate `s_hat`, which in turn helps to calculate the value of `k`. In the paper, they use `m = 100`, but you can experiment with different values to see how it affects the performance of the algorithm.
/// @param mu Maximum cross-entropy. This value is initialized to be twice the target cross-entropy (`2 * tau`) and is updated in the algorithm based on the error between the target and observed surprisal.
LLAMA_API llama_token llama_sample_token_mirostat(
struct llama_context * ctx,
llama_token_data_array * candidates,
float tau,
float eta,
int32_t m,
float * mu);
/// @details Mirostat 2.0 algorithm described in the paper https://arxiv.org/abs/2007.14966. Uses tokens instead of words.
/// @param candidates A vector of `llama_token_data` containing the candidate tokens, their probabilities (p), and log-odds (logit) for the current position in the generated text.
/// @param tau The target cross-entropy (or surprise) value you want to achieve for the generated text. A higher value corresponds to more surprising or less predictable text, while a lower value corresponds to less surprising or more predictable text.
/// @param eta The learning rate used to update `mu` based on the error between the target and observed surprisal of the sampled word. A larger learning rate will cause `mu` to be updated more quickly, while a smaller learning rate will result in slower updates.
/// @param mu Maximum cross-entropy. This value is initialized to be twice the target cross-entropy (`2 * tau`) and is updated in the algorithm based on the error between the target and observed surprisal.
LLAMA_API llama_token llama_sample_token_mirostat_v2(
struct llama_context * ctx,
llama_token_data_array * candidates,
float tau,
float eta,
float * mu);
/// @details Selects the token with the highest probability.
/// Does not compute the token probabilities. Use llama_sample_softmax() instead.
LLAMA_API llama_token llama_sample_token_greedy(
struct llama_context * ctx,
llama_token_data_array * candidates);
/// @details Randomly selects a token from the candidates based on their probabilities.
LLAMA_API llama_token llama_sample_token(
struct llama_context * ctx,
llama_token_data_array * candidates);
/// @details Accepts the sampled token into the grammar
LLAMA_API void llama_grammar_accept_token(
struct llama_context * ctx,
struct llama_grammar * grammar,
llama_token token);
//
// Beam search
//
struct llama_beam_view {
const llama_token * tokens;
size_t n_tokens;
float p; // Cumulative beam probability (renormalized relative to all beams)
bool eob; // Callback should set this to true when a beam is at end-of-beam.
};
// Passed to beam_search_callback function.
// Whenever 0 < common_prefix_length, this number of tokens should be copied from any of the beams
// (e.g. beams[0]) as they will be removed (shifted) from all beams in all subsequent callbacks.
// These pointers are valid only during the synchronous callback, so should not be saved.
struct llama_beams_state {
struct llama_beam_view * beam_views;
size_t n_beams; // Number of elements in beam_views[].
size_t common_prefix_length; // Current max length of prefix tokens shared by all beams.
bool last_call; // True iff this is the last callback invocation.
};
// Type of pointer to the beam_search_callback function.
// void* callback_data is any custom data passed to llama_beam_search, that is subsequently
// passed back to beam_search_callback. This avoids having to use global variables in the callback.
typedef void (*llama_beam_search_callback_fn_t)(void * callback_data, struct llama_beams_state);
/// @details Deterministically returns entire sentence constructed by a beam search.
/// @param ctx Pointer to the llama_context.
/// @param callback Invoked for each iteration of the beam_search loop, passing in beams_state.
/// @param callback_data A pointer that is simply passed back to callback.
/// @param n_beams Number of beams to use.
/// @param n_past Number of tokens already evaluated.
/// @param n_predict Maximum number of tokens to predict. EOS may occur earlier.
LLAMA_API void llama_beam_search(
struct llama_context * ctx,
llama_beam_search_callback_fn_t callback,
void * callback_data,
size_t n_beams,
int32_t n_past,
int32_t n_predict);
// Performance information
LLAMA_API struct llama_timings llama_get_timings(struct llama_context * ctx);
LLAMA_API void llama_print_timings(struct llama_context * ctx);
LLAMA_API void llama_reset_timings(struct llama_context * ctx);
// Print system information
LLAMA_API const char * llama_print_system_info(void);
// Set callback for all future logging events.
// If this is not called, or NULL is supplied, everything is output on stderr.
LLAMA_API void llama_log_set(ggml_log_callback log_callback, void * user_data);
LLAMA_API void llama_dump_timing_info_yaml(FILE * stream, const struct llama_context * ctx);
#ifdef __cplusplus
}
#endif
// Internal API to be implemented by llama.cpp and used by tests/benchmarks only
#ifdef LLAMA_API_INTERNAL
#include <vector>
#include <string>
struct ggml_tensor;
const std::vector<std::pair<std::string, struct ggml_tensor *>> & llama_internal_get_tensor_map(
struct llama_context * ctx
);
#endif // LLAMA_API_INTERNAL
#endif // LLAMA_H

723
llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/log.h Normal file
View file

@ -0,0 +1,723 @@
#pragma once
#include <chrono>
#include <cstring>
#include <sstream>
#include <iostream>
#include <thread>
#include <vector>
#include <algorithm>
#include <cinttypes>
// --------------------------------
//
// Basic usage:
//
// --------
//
// The LOG() and LOG_TEE() macros are ready to go by default
// they do not require any initialization.
//
// LOGLN() and LOG_TEELN() are variants which automatically
// include \n character at the end of the log string.
//
// LOG() behaves exactly like printf, by default writing to a logfile.
// LOG_TEE() additionally, prints to the screen too ( mimics Unix tee command ).
//
// Default logfile is named
// "llama.<threadID>.log"
// Default LOG_TEE() secondary output target is
// stderr
//
// Logs can be dynamically disabled or enabled using functions:
// log_disable()
// and
// log_enable()
//
// A log target can be changed with:
// log_set_target( string )
// creating and opening, or re-opening a file by string filename
// or
// log_set_target( FILE* )
// allowing to point at stderr, stdout, or any valid FILE* file handler.
//
// --------
//
// End of Basic usage.
//
// --------------------------------
// Specifies a log target.
// default uses log_handler() with "llama.log" log file
// this can be changed, by defining LOG_TARGET
// like so:
//
// #define LOG_TARGET (a valid FILE*)
// #include "log.h"
//
// or it can be simply redirected to stdout or stderr
// like so:
//
// #define LOG_TARGET stderr
// #include "log.h"
//
// The log target can also be redirected to a different function
// like so:
//
// #define LOG_TARGET log_handler_different()
// #include "log.h"
//
// FILE* log_handler_different()
// {
// return stderr;
// }
//
// or:
//
// #define LOG_TARGET log_handler_another_one("somelog.log")
// #include "log.h"
//
// FILE* log_handler_another_one(char*filename)
// {
// static FILE* logfile = nullptr;
// (...)
// if( !logfile )
// {
// fopen(...)
// }
// (...)
// return logfile
// }
//
#ifndef LOG_TARGET
#define LOG_TARGET log_handler()
#endif
#ifndef LOG_TEE_TARGET
#define LOG_TEE_TARGET stderr
#endif
// Utility for synchronizing log configuration state
// since std::optional was introduced only in c++17
enum LogTriState
{
LogTriStateSame,
LogTriStateFalse,
LogTriStateTrue
};
// Utility to obtain "pid" like unique process id and use it when creating log files.
inline std::string log_get_pid()
{
static std::string pid;
if (pid.empty())
{
// std::this_thread::get_id() is the most portable way of obtaining a "process id"
// it's not the same as "pid" but is unique enough to solve multiple instances
// trying to write to the same log.
std::stringstream ss;
ss << std::this_thread::get_id();
pid = ss.str();
}
return pid;
}
// Utility function for generating log file names with unique id based on thread id.
// invocation with log_filename_generator( "llama", "log" ) creates a string "llama.<number>.log"
// where the number is a runtime id of the current thread.
#define log_filename_generator(log_file_basename, log_file_extension) log_filename_generator_impl(LogTriStateSame, log_file_basename, log_file_extension)
// INTERNAL, DO NOT USE
inline std::string log_filename_generator_impl(LogTriState multilog, const std::string & log_file_basename, const std::string & log_file_extension)
{
static bool _multilog = false;
if (multilog != LogTriStateSame)
{
_multilog = multilog == LogTriStateTrue;
}
std::stringstream buf;
buf << log_file_basename;
if (_multilog)
{
buf << ".";
buf << log_get_pid();
}
buf << ".";
buf << log_file_extension;
return buf.str();
}
#ifndef LOG_DEFAULT_FILE_NAME
#define LOG_DEFAULT_FILE_NAME log_filename_generator("llama", "log")
#endif
// Utility for turning #define values into string literals
// so we can have a define for stderr and
// we can print "stderr" instead of literal stderr, etc.
#define LOG_STRINGIZE1(s) #s
#define LOG_STRINGIZE(s) LOG_STRINGIZE1(s)
#define LOG_TEE_TARGET_STRING LOG_STRINGIZE(LOG_TEE_TARGET)
// Allows disabling timestamps.
// in order to disable, define LOG_NO_TIMESTAMPS
// like so:
//
// #define LOG_NO_TIMESTAMPS
// #include "log.h"
//
#ifndef LOG_NO_TIMESTAMPS
#ifndef _MSC_VER
#define LOG_TIMESTAMP_FMT "[%" PRIu64 "] "
#define LOG_TIMESTAMP_VAL , (std::chrono::duration_cast<std::chrono::duration<std::uint64_t>>(std::chrono::system_clock::now().time_since_epoch())).count()
#else
#define LOG_TIMESTAMP_FMT "[%" PRIu64 "] "
#define LOG_TIMESTAMP_VAL , (std::chrono::duration_cast<std::chrono::duration<std::uint64_t>>(std::chrono::system_clock::now().time_since_epoch())).count()
#endif
#else
#define LOG_TIMESTAMP_FMT "%s"
#define LOG_TIMESTAMP_VAL ,""
#endif
#ifdef LOG_TEE_TIMESTAMPS
#ifndef _MSC_VER
#define LOG_TEE_TIMESTAMP_FMT "[%" PRIu64 "] "
#define LOG_TEE_TIMESTAMP_VAL , (std::chrono::duration_cast<std::chrono::duration<std::uint64_t>>(std::chrono::system_clock::now().time_since_epoch())).count()
#else
#define LOG_TEE_TIMESTAMP_FMT "[%" PRIu64 "] "
#define LOG_TEE_TIMESTAMP_VAL , (std::chrono::duration_cast<std::chrono::duration<std::uint64_t>>(std::chrono::system_clock::now().time_since_epoch())).count()
#endif
#else
#define LOG_TEE_TIMESTAMP_FMT "%s"
#define LOG_TEE_TIMESTAMP_VAL ,""
#endif
// Allows disabling file/line/function prefix
// in order to disable, define LOG_NO_FILE_LINE_FUNCTION
// like so:
//
// #define LOG_NO_FILE_LINE_FUNCTION
// #include "log.h"
//
#ifndef LOG_NO_FILE_LINE_FUNCTION
#ifndef _MSC_VER
#define LOG_FLF_FMT "[%24s:%5d][%24s] "
#define LOG_FLF_VAL , __FILE__, __LINE__, __FUNCTION__
#else
#define LOG_FLF_FMT "[%24s:%5ld][%24s] "
#define LOG_FLF_VAL , __FILE__, __LINE__, __FUNCTION__
#endif
#else
#define LOG_FLF_FMT "%s"
#define LOG_FLF_VAL ,""
#endif
#ifdef LOG_TEE_FILE_LINE_FUNCTION
#ifndef _MSC_VER
#define LOG_TEE_FLF_FMT "[%24s:%5d][%24s] "
#define LOG_TEE_FLF_VAL , __FILE__, __LINE__, __FUNCTION__
#else
#define LOG_TEE_FLF_FMT "[%24s:%5ld][%24s] "
#define LOG_TEE_FLF_VAL , __FILE__, __LINE__, __FUNCTION__
#endif
#else
#define LOG_TEE_FLF_FMT "%s"
#define LOG_TEE_FLF_VAL ,""
#endif
// INTERNAL, DO NOT USE
// USE LOG() INSTEAD
//
#ifndef _MSC_VER
#define LOG_IMPL(str, ...) \
do { \
if (LOG_TARGET != nullptr) \
{ \
fprintf(LOG_TARGET, LOG_TIMESTAMP_FMT LOG_FLF_FMT str "%s" LOG_TIMESTAMP_VAL LOG_FLF_VAL, __VA_ARGS__); \
fflush(LOG_TARGET); \
} \
} while (0)
#else
#define LOG_IMPL(str, ...) \
do { \
if (LOG_TARGET != nullptr) \
{ \
fprintf(LOG_TARGET, LOG_TIMESTAMP_FMT LOG_FLF_FMT str "%s" LOG_TIMESTAMP_VAL LOG_FLF_VAL "", ##__VA_ARGS__); \
fflush(LOG_TARGET); \
} \
} while (0)
#endif
// INTERNAL, DO NOT USE
// USE LOG_TEE() INSTEAD
//
#ifndef _MSC_VER
#define LOG_TEE_IMPL(str, ...) \
do { \
if (LOG_TARGET != nullptr) \
{ \
fprintf(LOG_TARGET, LOG_TIMESTAMP_FMT LOG_FLF_FMT str "%s" LOG_TIMESTAMP_VAL LOG_FLF_VAL, __VA_ARGS__); \
fflush(LOG_TARGET); \
} \
if (LOG_TARGET != nullptr && LOG_TARGET != stdout && LOG_TARGET != stderr && LOG_TEE_TARGET != nullptr) \
{ \
fprintf(LOG_TEE_TARGET, LOG_TEE_TIMESTAMP_FMT LOG_TEE_FLF_FMT str "%s" LOG_TEE_TIMESTAMP_VAL LOG_TEE_FLF_VAL, __VA_ARGS__); \
fflush(LOG_TEE_TARGET); \
} \
} while (0)
#else
#define LOG_TEE_IMPL(str, ...) \
do { \
if (LOG_TARGET != nullptr) \
{ \
fprintf(LOG_TARGET, LOG_TIMESTAMP_FMT LOG_FLF_FMT str "%s" LOG_TIMESTAMP_VAL LOG_FLF_VAL "", ##__VA_ARGS__); \
fflush(LOG_TARGET); \
} \
if (LOG_TARGET != nullptr && LOG_TARGET != stdout && LOG_TARGET != stderr && LOG_TEE_TARGET != nullptr) \
{ \
fprintf(LOG_TEE_TARGET, LOG_TEE_TIMESTAMP_FMT LOG_TEE_FLF_FMT str "%s" LOG_TEE_TIMESTAMP_VAL LOG_TEE_FLF_VAL "", ##__VA_ARGS__); \
fflush(LOG_TEE_TARGET); \
} \
} while (0)
#endif
// The '\0' as a last argument, is a trick to bypass the silly
// "warning: ISO C++11 requires at least one argument for the "..." in a variadic macro"
// so we can have a single macro which can be called just like printf.
// Main LOG macro.
// behaves like printf, and supports arguments the exact same way.
//
#ifndef _MSC_VER
#define LOG(...) LOG_IMPL(__VA_ARGS__, "")
#else
#define LOG(str, ...) LOG_IMPL("%s" str, "", __VA_ARGS__, "")
#endif
// Main TEE macro.
// does the same as LOG
// and
// simultaneously writes stderr.
//
// Secondary target can be changed just like LOG_TARGET
// by defining LOG_TEE_TARGET
//
#ifndef _MSC_VER
#define LOG_TEE(...) LOG_TEE_IMPL(__VA_ARGS__, "")
#else
#define LOG_TEE(str, ...) LOG_TEE_IMPL("%s" str, "", __VA_ARGS__, "")
#endif
// LOG macro variants with auto endline.
#ifndef _MSC_VER
#define LOGLN(...) LOG_IMPL(__VA_ARGS__, "\n")
#define LOG_TEELN(...) LOG_TEE_IMPL(__VA_ARGS__, "\n")
#else
#define LOGLN(str, ...) LOG_IMPL("%s" str, "", __VA_ARGS__, "\n")
#define LOG_TEELN(str, ...) LOG_TEE_IMPL("%s" str, "", __VA_ARGS__, "\n")
#endif
// INTERNAL, DO NOT USE
inline FILE *log_handler1_impl(bool change = false, LogTriState append = LogTriStateSame, LogTriState disable = LogTriStateSame, const std::string & filename = LOG_DEFAULT_FILE_NAME, FILE *target = nullptr)
{
static bool _initialized = false;
static bool _append = false;
static bool _disabled = filename.empty() && target == nullptr;
static std::string log_current_filename{filename};
static FILE *log_current_target{target};
static FILE *logfile = nullptr;
if (change)
{
if (append != LogTriStateSame)
{
_append = append == LogTriStateTrue;
return logfile;
}
if (disable == LogTriStateTrue)
{
// Disable primary target
_disabled = true;
}
// If previously disabled, only enable, and keep previous target
else if (disable == LogTriStateFalse)
{
_disabled = false;
}
// Otherwise, process the arguments
else if (log_current_filename != filename || log_current_target != target)
{
_initialized = false;
}
}
if (_disabled)
{
// Log is disabled
return nullptr;
}
if (_initialized)
{
// with fallback in case something went wrong
return logfile ? logfile : stderr;
}
// do the (re)initialization
if (target != nullptr)
{
if (logfile != nullptr && logfile != stdout && logfile != stderr)
{
fclose(logfile);
}
log_current_filename = LOG_DEFAULT_FILE_NAME;
log_current_target = target;
logfile = target;
}
else
{
if (log_current_filename != filename)
{
if (logfile != nullptr && logfile != stdout && logfile != stderr)
{
fclose(logfile);
}
}
logfile = fopen(filename.c_str(), _append ? "a" : "w");
}
if (!logfile)
{
// Verify whether the file was opened, otherwise fallback to stderr
logfile = stderr;
fprintf(stderr, "Failed to open logfile '%s' with error '%s'\n", filename.c_str(), std::strerror(errno));
fflush(stderr);
// At this point we let the init flag be to true below, and let the target fallback to stderr
// otherwise we would repeatedly fopen() which was already unsuccessful
}
_initialized = true;
return logfile ? logfile : stderr;
}
// INTERNAL, DO NOT USE
inline FILE *log_handler2_impl(bool change = false, LogTriState append = LogTriStateSame, LogTriState disable = LogTriStateSame, FILE *target = nullptr, const std::string & filename = LOG_DEFAULT_FILE_NAME)
{
return log_handler1_impl(change, append, disable, filename, target);
}
// Disables logs entirely at runtime.
// Makes LOG() and LOG_TEE() produce no output,
// until enabled back.
#define log_disable() log_disable_impl()
// INTERNAL, DO NOT USE
inline FILE *log_disable_impl()
{
return log_handler1_impl(true, LogTriStateSame, LogTriStateTrue);
}
// Enables logs at runtime.
#define log_enable() log_enable_impl()
// INTERNAL, DO NOT USE
inline FILE *log_enable_impl()
{
return log_handler1_impl(true, LogTriStateSame, LogTriStateFalse);
}
// Sets target fir logs, either by a file name or FILE* pointer (stdout, stderr, or any valid FILE*)
#define log_set_target(target) log_set_target_impl(target)
// INTERNAL, DO NOT USE
inline FILE *log_set_target_impl(const std::string & filename) { return log_handler1_impl(true, LogTriStateSame, LogTriStateSame, filename); }
inline FILE *log_set_target_impl(FILE *target) { return log_handler2_impl(true, LogTriStateSame, LogTriStateSame, target); }
// INTERNAL, DO NOT USE
inline FILE *log_handler() { return log_handler1_impl(); }
// Enable or disable creating separate log files for each run.
// can ONLY be invoked BEFORE first log use.
#define log_multilog(enable) log_filename_generator_impl((enable) ? LogTriStateTrue : LogTriStateFalse, "", "")
// Enable or disable append mode for log file.
// can ONLY be invoked BEFORE first log use.
#define log_append(enable) log_append_impl(enable)
// INTERNAL, DO NOT USE
inline FILE *log_append_impl(bool enable)
{
return log_handler1_impl(true, enable ? LogTriStateTrue : LogTriStateFalse, LogTriStateSame);
}
inline void log_test()
{
log_disable();
LOG("01 Hello World to nobody, because logs are disabled!\n");
log_enable();
LOG("02 Hello World to default output, which is \"%s\" ( Yaaay, arguments! )!\n", LOG_STRINGIZE(LOG_TARGET));
LOG_TEE("03 Hello World to **both** default output and " LOG_TEE_TARGET_STRING "!\n");
log_set_target(stderr);
LOG("04 Hello World to stderr!\n");
LOG_TEE("05 Hello World TEE with double printing to stderr prevented!\n");
log_set_target(LOG_DEFAULT_FILE_NAME);
LOG("06 Hello World to default log file!\n");
log_set_target(stdout);
LOG("07 Hello World to stdout!\n");
log_set_target(LOG_DEFAULT_FILE_NAME);
LOG("08 Hello World to default log file again!\n");
log_disable();
LOG("09 Hello World _1_ into the void!\n");
log_enable();
LOG("10 Hello World back from the void ( you should not see _1_ in the log or the output )!\n");
log_disable();
log_set_target("llama.anotherlog.log");
LOG("11 Hello World _2_ to nobody, new target was selected but logs are still disabled!\n");
log_enable();
LOG("12 Hello World this time in a new file ( you should not see _2_ in the log or the output )?\n");
log_set_target("llama.yetanotherlog.log");
LOG("13 Hello World this time in yet new file?\n");
log_set_target(log_filename_generator("llama_autonamed", "log"));
LOG("14 Hello World in log with generated filename!\n");
#ifdef _MSC_VER
LOG_TEE("15 Hello msvc TEE without arguments\n");
LOG_TEE("16 Hello msvc TEE with (%d)(%s) arguments\n", 1, "test");
LOG_TEELN("17 Hello msvc TEELN without arguments\n");
LOG_TEELN("18 Hello msvc TEELN with (%d)(%s) arguments\n", 1, "test");
LOG("19 Hello msvc LOG without arguments\n");
LOG("20 Hello msvc LOG with (%d)(%s) arguments\n", 1, "test");
LOGLN("21 Hello msvc LOGLN without arguments\n");
LOGLN("22 Hello msvc LOGLN with (%d)(%s) arguments\n", 1, "test");
#endif
}
inline bool log_param_single_parse(const std::string & param)
{
if ( param == "--log-test")
{
log_test();
return true;
}
if ( param == "--log-disable")
{
log_disable();
return true;
}
if ( param == "--log-enable")
{
log_enable();
return true;
}
if (param == "--log-new")
{
log_multilog(true);
return true;
}
if (param == "--log-append")
{
log_append(true);
return true;
}
return false;
}
inline bool log_param_pair_parse(bool check_but_dont_parse, const std::string & param, const std::string & next = std::string())
{
if ( param == "--log-file")
{
if (!check_but_dont_parse)
{
log_set_target(log_filename_generator(next.empty() ? "unnamed" : next, "log"));
}
return true;
}
return false;
}
inline void log_print_usage()
{
printf("log options:\n");
/* format
printf(" -h, --help show this help message and exit\n");*/
/* spacing
printf("__-param----------------Description\n");*/
printf(" --log-test Run simple logging test\n");
printf(" --log-disable Disable trace logs\n");
printf(" --log-enable Enable trace logs\n");
printf(" --log-file Specify a log filename (without extension)\n");
printf(" --log-new Create a separate new log file on start. "
"Each log file will have unique name: \"<name>.<ID>.log\"\n");
printf(" --log-append Don't truncate the old log file.\n");
}
#define log_dump_cmdline(argc, argv) log_dump_cmdline_impl(argc, argv)
// INTERNAL, DO NOT USE
inline void log_dump_cmdline_impl(int argc, char **argv)
{
std::stringstream buf;
for (int i = 0; i < argc; ++i)
{
if (std::string(argv[i]).find(' ') != std::string::npos)
{
buf << " \"" << argv[i] <<"\"";
}
else
{
buf << " " << argv[i];
}
}
LOGLN("Cmd:%s", buf.str().c_str());
}
#define log_tostr(var) log_var_to_string_impl(var).c_str()
inline std::string log_var_to_string_impl(bool var)
{
return var ? "true" : "false";
}
inline std::string log_var_to_string_impl(std::string var)
{
return var;
}
inline std::string log_var_to_string_impl(const std::vector<int> & var)
{
std::stringstream buf;
buf << "[ ";
bool first = true;
for (auto e : var)
{
if (first)
{
first = false;
}
else
{
buf << ", ";
}
buf << std::to_string(e);
}
buf << " ]";
return buf.str();
}
template <typename C, typename T>
inline std::string LOG_TOKENS_TOSTR_PRETTY(const C & ctx, const T & tokens)
{
std::stringstream buf;
buf << "[ ";
bool first = true;
for (const auto &token : tokens)
{
if (!first) {
buf << ", ";
} else {
first = false;
}
auto detokenized = llama_token_to_piece(ctx, token);
detokenized.erase(
std::remove_if(
detokenized.begin(),
detokenized.end(),
[](const unsigned char c) { return !std::isprint(c); }),
detokenized.end());
buf
<< "'" << detokenized << "'"
<< ":" << std::to_string(token);
}
buf << " ]";
return buf.str();
}
template <typename C, typename B>
inline std::string LOG_BATCH_TOSTR_PRETTY(const C & ctx, const B & batch)
{
std::stringstream buf;
buf << "[ ";
bool first = true;
for (int i = 0; i < batch.n_tokens; ++i)
{
if (!first) {
buf << ", ";
} else {
first = false;
}
auto detokenized = llama_token_to_piece(ctx, batch.token[i]);
detokenized.erase(
std::remove_if(
detokenized.begin(),
detokenized.end(),
[](const unsigned char c) { return !std::isprint(c); }),
detokenized.end());
buf
<< "\n" << std::to_string(i)
<< ":token '" << detokenized << "'"
<< ":pos " << std::to_string(batch.pos[i])
<< ":n_seq_id " << std::to_string(batch.n_seq_id[i])
<< ":seq_id " << std::to_string(batch.seq_id[i][0])
<< ":logits " << std::to_string(batch.logits[i]);
}
buf << " ]";
return buf.str();
}
#ifdef LOG_DISABLE_LOGS
#undef LOG
#define LOG(...) // dummy stub
#undef LOGLN
#define LOGLN(...) // dummy stub
#undef LOG_TEE
#define LOG_TEE(...) fprintf(stderr, __VA_ARGS__) // convert to normal fprintf
#undef LOG_TEELN
#define LOG_TEELN(...) fprintf(stderr, __VA_ARGS__) // convert to normal fprintf
#undef LOG_DISABLE
#define LOG_DISABLE() // dummy stub
#undef LOG_ENABLE
#define LOG_ENABLE() // dummy stub
#undef LOG_ENABLE
#define LOG_ENABLE() // dummy stub
#undef LOG_SET_TARGET
#define LOG_SET_TARGET(...) // dummy stub
#undef LOG_DUMP_CMDLINE
#define LOG_DUMP_CMDLINE(...) // dummy stub
#endif // LOG_DISABLE_LOGS

117
llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/sampling.h Normal file
View file

@ -0,0 +1,117 @@
#pragma once
#include "llama.h"
#include "grammar-parser.h"
#include <string>
#include <vector>
#include <unordered_map>
// sampling parameters
typedef struct llama_sampling_params {
int32_t n_prev = 64; // number of previous tokens to remember
int32_t n_probs = 0; // if greater than 0, output the probabilities of top n_probs tokens.
int32_t top_k = 40; // <= 0 to use vocab size
float top_p = 0.95f; // 1.0 = disabled
float min_p = 0.05f; // 0.0 = disabled
float tfs_z = 1.00f; // 1.0 = disabled
float typical_p = 1.00f; // 1.0 = disabled
float temp = 0.80f; // <= 0.0 to sample greedily, 0.0 to not output probabilities
int32_t penalty_last_n = 64; // last n tokens to penalize (0 = disable penalty, -1 = context size)
float penalty_repeat = 1.10f; // 1.0 = disabled
float penalty_freq = 0.00f; // 0.0 = disabled
float penalty_present = 0.00f; // 0.0 = disabled
int32_t mirostat = 0; // 0 = disabled, 1 = mirostat, 2 = mirostat 2.0
float mirostat_tau = 5.00f; // target entropy
float mirostat_eta = 0.10f; // learning rate
bool penalize_nl = true; // consider newlines as a repeatable token
std::string samplers_sequence = "kfypmt"; // top_k, tail_free, typical_p, top_p, min_p, temp
std::string grammar; // optional BNF-like grammar to constrain sampling
// Classifier-Free Guidance
// https://arxiv.org/abs/2306.17806
std::string cfg_negative_prompt; // string to help guidance
float cfg_scale = 1.f; // how strong is guidance
std::unordered_map<llama_token, float> logit_bias; // logit bias for specific tokens
std::vector<llama_token> penalty_prompt_tokens;
bool use_penalty_prompt_tokens = false;
} llama_sampling_params;
// general sampler context
// TODO: move to llama.h
struct llama_sampling_context {
// parameters that will be used for sampling
llama_sampling_params params;
// mirostat sampler state
float mirostat_mu;
llama_grammar * grammar;
// internal
grammar_parser::parse_state parsed_grammar;
// TODO: replace with ring-buffer
std::vector<llama_token> prev;
std::vector<llama_token_data> cur;
};
#include "common.h"
// Create a new sampling context instance.
struct llama_sampling_context * llama_sampling_init(const struct llama_sampling_params & params);
void llama_sampling_free(struct llama_sampling_context * ctx);
// Reset the sampler context
// - clear prev tokens
// - reset grammar
void llama_sampling_reset(llama_sampling_context * ctx);
// Copy the sampler context
void llama_sampling_cp(llama_sampling_context * src, llama_sampling_context * dst);
// Get the last sampled token
llama_token llama_sampling_last(llama_sampling_context * ctx);
// Get a string representation of the last sampled tokens
std::string llama_sampling_prev_str(llama_sampling_context * ctx_sampling, llama_context * ctx_main, int n);
// Print sampling parameters into a string
std::string llama_sampling_print(const llama_sampling_params & params);
// Print sampling order into a string
std::string llama_sampling_order_print(const llama_sampling_params & params);
// this is a common sampling function used across the examples for convenience
// it can serve as a starting point for implementing your own sampling function
// Note: When using multiple sequences, it is the caller's responsibility to call
// llama_sampling_reset when a sequence ends
//
// required:
// - ctx_main: context to use for sampling
// - ctx_sampling: sampling-specific context
//
// optional:
// - ctx_cfg: context to use for classifier-free guidance
// - idx: sample from llama_get_logits_ith(ctx, idx)
//
// returns:
// - token: sampled token
// - candidates: vector of candidate tokens
//
llama_token llama_sampling_sample(
struct llama_sampling_context * ctx_sampling,
struct llama_context * ctx_main,
struct llama_context * ctx_cfg,
int idx = 0);
void llama_sampling_accept(
struct llama_sampling_context * ctx_sampling,
struct llama_context * ctx_main,
llama_token id,
bool apply_grammar);

31
llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/tokenize.h Normal file
View file

@ -0,0 +1,31 @@
//
// This source file is part of the Stanford Spezi open source project
//
// SPDX-FileCopyrightText: 2022 Stanford University and the project authors (see CONTRIBUTORS.md)
//
// SPDX-License-Identifier: MIT
//
#ifndef tokenize_hpp
#define tokenize_hpp
#include <vector>
#include "common.h"
/// Tokenize a `String` via a given `llama_context`.
std::vector<llama_token> llama_tokenize_with_context(
const struct llama_context * ctx,
const std::string & text,
bool add_bos,
bool special = false);
/// Tokenize a `String` via a given `llama_model`.
std::vector<llama_token> llama_tokenize_with_model(
const struct llama_model * model,
const std::string & text,
bool add_bos,
bool special = false);
#endif

20
llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/vector.h Normal file
View file

@ -0,0 +1,20 @@
//
// This source file is part of the Stanford Spezi open source project
//
// SPDX-FileCopyrightText: 2022 Stanford University and the project authors (see CONTRIBUTORS.md)
//
// SPDX-License-Identifier: MIT
//
#ifndef vector_hpp
#define vector_hpp
#include <vector>
#include "common.h"
/// Create an empty `vector` of `llama_seq_id`s that serve as a buffer for batch processing.
const std::vector<llama_seq_id> getLlamaSeqIdVector();
#endif

BIN
llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Info.plist Normal file
View file

Binary file not shown.

21
llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Modules/module.modulemap Normal file
View file

@ -0,0 +1,21 @@
//
// This source file is part of the Stanford Biodesign Digital Health Group open-source organization
//
// SPDX-FileCopyrightText: 2022 Stanford University and the project authors (see CONTRIBUTORS.md)
//
// SPDX-License-Identifier: MIT
//
framework module llama {
umbrella header "llama.h"
header "common.h"
header "sampling.h"
header "log.h"
header "grammar-parser.h"
header "tokenize.h"
header "vector.h"
export *
module * { export * }
}

245
llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/_CodeSignature/CodeResources Normal file
View file

@ -0,0 +1,245 @@
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
<key>files</key>
<dict>
<key>Headers/common.h</key>
<data>
ccJTLLUsgSf/AV/abgh+IY0T6ow=
</data>
<key>Headers/ggml-alloc.h</key>
<data>
dg6+B6JZ7fq/Of8Cv4kpDJVKTpw=
</data>
<key>Headers/ggml-backend.h</key>
<data>
JD2Mm6dTAVM9F617pSTY5QJMfbo=
</data>
<key>Headers/ggml.h</key>
<data>
oodXwbw09pAPSPlBHU7skZi7Ik4=
</data>
<key>Headers/grammar-parser.h</key>
<data>
RGyClQ/Ka2DSO36xeqWdIIVcbp8=
</data>
<key>Headers/llama.h</key>
<data>
CjUx0Cp75fFrO9DHvEiNff5LkZY=
</data>
<key>Headers/log.h</key>
<data>
AUSS72p+klCokW4hrNI3ovFUWkA=
</data>
<key>Headers/sampling.h</key>
<data>
YpY5vLgRVtf44wKALIx7ni6bMFI=
</data>
<key>Headers/tokenize.h</key>
<data>
4+UZgLFkmr9R4LmEeO4aUKrBiQQ=
</data>
<key>Headers/vector.h</key>
<data>
hNjPw11RWceCFywmNFIm3NY9uj4=
</data>
<key>Info.plist</key>
<data>
HeJpCFpJksp6N94iNViJ3yQRPFY=
</data>
<key>Modules/module.modulemap</key>
<data>
cRLcYZrRPoNY51D5f4mvxPV5xOw=
</data>
<key>default.metallib</key>
<data>
4AX0hqyWtJbXyzKMPiYeRIwv3ZE=
</data>
<key>ggml-metal.dat</key>
<data>
vQ2/0Cc6DnvPmJp5DTVyrTCTiCQ=
</data>
</dict>
<key>files2</key>
<dict>
<key>Headers/common.h</key>
<dict>
<key>hash2</key>
<data>
WgpqkYY9/IBE5i4FtOF9M+t8Kp8koyiHnlAbM7AThbs=
</data>
</dict>
<key>Headers/ggml-alloc.h</key>
<dict>
<key>hash2</key>
<data>
fmfygbjuVqFsG+sps1PEt7kk14bxpFOJa2EWFlpdO6U=
</data>
</dict>
<key>Headers/ggml-backend.h</key>
<dict>
<key>hash2</key>
<data>
IwSTKOFUc8Gq2nUZPrKsuP6iJNKO3YMSEd6imOM1hmY=
</data>
</dict>
<key>Headers/ggml.h</key>
<dict>
<key>hash2</key>
<data>
jDel1kg6P5ScDxsA0M+aDoGNP8crjrfbCjK/Lx4n+oc=
</data>
</dict>
<key>Headers/grammar-parser.h</key>
<dict>
<key>hash2</key>
<data>
ZGQBEnYU3HrvwLQZYT38vF43D9t3j1hAB6WQkKyTZDo=
</data>
</dict>
<key>Headers/llama.h</key>
<dict>
<key>hash2</key>
<data>
JLNNCCJ7zkAUs011XUHoZhxxEWnyhQ+rsXdwc6tleU8=
</data>
</dict>
<key>Headers/log.h</key>
<dict>
<key>hash2</key>
<data>
Or2LBsdTY1IPMzn7QJyTJoKgbfPh8zQ5uCv8gASU69Y=
</data>
</dict>
<key>Headers/sampling.h</key>
<dict>
<key>hash2</key>
<data>
OU5e/zSQWKNZ3mI0lx9jUJvj1FXADWe0oYL6eoCsvk8=
</data>
</dict>
<key>Headers/tokenize.h</key>
<dict>
<key>hash2</key>
<data>
V/dNu95+bR1zdfXev3yXYpQezyFccynCYmDR8RyBXzc=
</data>
</dict>
<key>Headers/vector.h</key>
<dict>
<key>hash2</key>
<data>
cZ9Iqs2ysi/8ZrobbqbQ3+156n4gI2U3ocaorK3C4AU=
</data>
</dict>
<key>Modules/module.modulemap</key>
<dict>
<key>hash2</key>
<data>
CqJJfiWIk28AhsmMlRerM9MZOVAcflGpnXIJ9vR595U=
</data>
</dict>
<key>default.metallib</key>
<dict>
<key>hash2</key>
<data>
Wf7LxVzu4pTWWEmH8brqQzAK6nQO2PMRkWjXvjoY9KI=
</data>
</dict>
<key>ggml-metal.dat</key>
<dict>
<key>hash2</key>
<data>
PcypCe70Xmdo/PObEO69fPkl+nYFpUgPatshgAPLD/4=
</data>
</dict>
</dict>
<key>rules</key>
<dict>
<key>^.*</key>
<true/>
<key>^.*\.lproj/</key>
<dict>
<key>optional</key>
<true/>
<key>weight</key>
<real>1000</real>
</dict>
<key>^.*\.lproj/locversion.plist$</key>
<dict>
<key>omit</key>
<true/>
<key>weight</key>
<real>1100</real>
</dict>
<key>^Base\.lproj/</key>
<dict>
<key>weight</key>
<real>1010</real>
</dict>
<key>^version.plist$</key>
<true/>
</dict>
<key>rules2</key>
<dict>
<key>.*\.dSYM($|/)</key>
<dict>
<key>weight</key>
<real>11</real>
</dict>
<key>^(.*/)?\.DS_Store$</key>
<dict>
<key>omit</key>
<true/>
<key>weight</key>
<real>2000</real>
</dict>
<key>^.*</key>
<true/>
<key>^.*\.lproj/</key>
<dict>
<key>optional</key>
<true/>
<key>weight</key>
<real>1000</real>
</dict>
<key>^.*\.lproj/locversion.plist$</key>
<dict>
<key>omit</key>
<true/>
<key>weight</key>
<real>1100</real>
</dict>
<key>^Base\.lproj/</key>
<dict>
<key>weight</key>
<real>1010</real>
</dict>
<key>^Info\.plist$</key>
<dict>
<key>omit</key>
<true/>
<key>weight</key>
<real>20</real>
</dict>
<key>^PkgInfo$</key>
<dict>
<key>omit</key>
<true/>
<key>weight</key>
<real>20</real>
</dict>
<key>^embedded\.provisionprofile$</key>
<dict>
<key>weight</key>
<real>20</real>
</dict>
<key>^version\.plist$</key>
<dict>
<key>weight</key>
<real>20</real>
</dict>
</dict>
</dict>
</plist>

BIN
llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/default.metallib Normal file
View file

Binary file not shown.

3
llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/ggml-metal.dat Normal file
View file

@ -0,0 +1,3 @@
dependencies: \
/Users/githubaction/runner/_work/llama.cpp/llama.cpp/ggml-metal.metal \
/Applications/Xcode-15.2.0.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/metal/ios/lib/clang/32023.98/include/metal/module.modulemap

BIN
llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/llama Normal file
View file

Binary file not shown.

541
llama.xcodeproj/project.pbxproj Normal file
View file

@ -0,0 +1,541 @@
// !$*UTF8*$!
{
archiveVersion = 1;
classes = {
};
objectVersion = 56;
objects = {
/* Begin PBXBuildFile section */
972079C42AFAACEF008A9E53 /* Accelerate.framework in Frameworks */ = {isa = PBXBuildFile; fileRef = 972079C32AFAACEF008A9E53 /* Accelerate.framework */; };
975AAE542B5F600F0079F713 /* ggml-backend.h in Headers */ = {isa = PBXBuildFile; fileRef = 975AAE532B5F600F0079F713 /* ggml-backend.h */; settings = {ATTRIBUTES = (Public, ); }; };
9789E9FC2B0D49BA00EB6AD2 /* sampling.h in Sources */ = {isa = PBXBuildFile; fileRef = 97FFE3072B056DD100B12DEC /* sampling.h */; };
9789E9FD2B0D49BD00EB6AD2 /* grammar-parser.h in Sources */ = {isa = PBXBuildFile; fileRef = 97FFE3062B056DD100B12DEC /* grammar-parser.h */; };
9789E9FE2B0D49BF00EB6AD2 /* log.h in Sources */ = {isa = PBXBuildFile; fileRef = 97FFE3052B056DD100B12DEC /* log.h */; };
979E9D0B2B60388300B5F255 /* ggml-alloc.h in Headers */ = {isa = PBXBuildFile; fileRef = 979E9D0A2B60388300B5F255 /* ggml-alloc.h */; settings = {ATTRIBUTES = (Public, ); }; };
97B76BF72B199E890019D706 /* tokenize.h in Headers */ = {isa = PBXBuildFile; fileRef = 97B76BF32B199E890019D706 /* tokenize.h */; settings = {ATTRIBUTES = (Public, ); }; };
97B76BF82B199E890019D706 /* vector.h in Headers */ = {isa = PBXBuildFile; fileRef = 97B76BF42B199E890019D706 /* vector.h */; settings = {ATTRIBUTES = (Public, ); }; };
97B76BF92B199E890019D706 /* vector.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 97B76BF52B199E890019D706 /* vector.cpp */; };
97B76BFB2B199EC60019D706 /* tokenize.h in Sources */ = {isa = PBXBuildFile; fileRef = 97B76BF32B199E890019D706 /* tokenize.h */; };
97B76BFC2B199EC60019D706 /* vector.h in Sources */ = {isa = PBXBuildFile; fileRef = 97B76BF42B199E890019D706 /* vector.h */; };
97B76BFD2B199EC60019D706 /* tokenize.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 97B76BF62B199E890019D706 /* tokenize.cpp */; };
97F217F92B044111008B81FE /* common.h in Headers */ = {isa = PBXBuildFile; fileRef = 97F217F82B044111008B81FE /* common.h */; settings = {ATTRIBUTES = (Public, ); }; };
97F217FB2B044120008B81FE /* common.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 97F217FA2B044120008B81FE /* common.cpp */; };
97F217FC2B04413F008B81FE /* common.h in Sources */ = {isa = PBXBuildFile; fileRef = 97F217F82B044111008B81FE /* common.h */; };
97F996D42AFB29C700194A6F /* ggml-metal.metal in Resources */ = {isa = PBXBuildFile; fileRef = 97F996CD2AFB29C600194A6F /* ggml-metal.metal */; };
97F996D52AFB29C700194A6F /* ggml-quants.c in Sources */ = {isa = PBXBuildFile; fileRef = 97F996CE2AFB29C600194A6F /* ggml-quants.c */; };
97F996D62AFB29C700194A6F /* ggml-metal.m in Sources */ = {isa = PBXBuildFile; fileRef = 97F996CF2AFB29C700194A6F /* ggml-metal.m */; settings = {COMPILER_FLAGS = "-fno-objc-arc"; }; };
97F996D72AFB29C700194A6F /* ggml-backend.c in Sources */ = {isa = PBXBuildFile; fileRef = 97F996D02AFB29C700194A6F /* ggml-backend.c */; };
97F996D82AFB29C700194A6F /* ggml.h in Headers */ = {isa = PBXBuildFile; fileRef = 97F996D12AFB29C700194A6F /* ggml.h */; settings = {ATTRIBUTES = (Public, ); }; };
97F996D92AFB29C700194A6F /* ggml-alloc.c in Sources */ = {isa = PBXBuildFile; fileRef = 97F996D22AFB29C700194A6F /* ggml-alloc.c */; };
97F996DA2AFB29C700194A6F /* llama.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 97F996D32AFB29C700194A6F /* llama.cpp */; };
97F996DC2AFB29F900194A6F /* llama.h in Headers */ = {isa = PBXBuildFile; fileRef = 97F996DB2AFB29F900194A6F /* llama.h */; settings = {ATTRIBUTES = (Public, ); }; };
97F996DE2AFB2A3F00194A6F /* ggml.c in Sources */ = {isa = PBXBuildFile; fileRef = 97F996DD2AFB2A3F00194A6F /* ggml.c */; };
97F996DF2AFB2AE100194A6F /* ggml.h in Sources */ = {isa = PBXBuildFile; fileRef = 97F996D12AFB29C700194A6F /* ggml.h */; };
97F996E02AFB2AE100194A6F /* llama.h in Sources */ = {isa = PBXBuildFile; fileRef = 97F996DB2AFB29F900194A6F /* llama.h */; };
97FAF1B32B0D4DA900E23209 /* grammar-parser.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 97FAF1B12B0D4DA900E23209 /* grammar-parser.cpp */; };
97FAF1B42B0D4DA900E23209 /* sampling.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 97FAF1B22B0D4DA900E23209 /* sampling.cpp */; };
97FFE3082B056DD100B12DEC /* log.h in Headers */ = {isa = PBXBuildFile; fileRef = 97FFE3052B056DD100B12DEC /* log.h */; settings = {ATTRIBUTES = (Public, ); }; };
97FFE3092B056DD100B12DEC /* grammar-parser.h in Headers */ = {isa = PBXBuildFile; fileRef = 97FFE3062B056DD100B12DEC /* grammar-parser.h */; settings = {ATTRIBUTES = (Public, ); }; };
97FFE30A2B056DD100B12DEC /* sampling.h in Headers */ = {isa = PBXBuildFile; fileRef = 97FFE3072B056DD100B12DEC /* sampling.h */; settings = {ATTRIBUTES = (Public, ); }; };
/* End PBXBuildFile section */
/* Begin PBXFileReference section */
972079A62AFAA818008A9E53 /* llama.framework */ = {isa = PBXFileReference; explicitFileType = wrapper.framework; includeInIndex = 0; path = llama.framework; sourceTree = BUILT_PRODUCTS_DIR; };
972079C32AFAACEF008A9E53 /* Accelerate.framework */ = {isa = PBXFileReference; lastKnownFileType = wrapper.framework; name = Accelerate.framework; path = Platforms/MacOSX.platform/Developer/SDKs/MacOSX14.0.sdk/System/Library/Frameworks/Accelerate.framework; sourceTree = DEVELOPER_DIR; };
975AAE532B5F600F0079F713 /* ggml-backend.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = "ggml-backend.h"; sourceTree = "<group>"; };
979D479D2B081F3E00EF8149 /* llama-module.modulemap */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = "sourcecode.module-map"; path = "llama-module.modulemap"; sourceTree = "<group>"; };
979E9D0A2B60388300B5F255 /* ggml-alloc.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = "ggml-alloc.h"; sourceTree = "<group>"; };
97B76BF32B199E890019D706 /* tokenize.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = tokenize.h; sourceTree = "<group>"; };
97B76BF42B199E890019D706 /* vector.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = vector.h; sourceTree = "<group>"; };
97B76BF52B199E890019D706 /* vector.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = vector.cpp; sourceTree = "<group>"; };
97B76BF62B199E890019D706 /* tokenize.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = tokenize.cpp; sourceTree = "<group>"; };
97F217F82B044111008B81FE /* common.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; name = common.h; path = common/common.h; sourceTree = "<group>"; };
97F217FA2B044120008B81FE /* common.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = common.cpp; path = common/common.cpp; sourceTree = "<group>"; };
97F996CD2AFB29C600194A6F /* ggml-metal.metal */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.metal; path = "ggml-metal.metal"; sourceTree = "<group>"; };
97F996CE2AFB29C600194A6F /* ggml-quants.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = "ggml-quants.c"; sourceTree = "<group>"; };
97F996CF2AFB29C700194A6F /* ggml-metal.m */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.objc; path = "ggml-metal.m"; sourceTree = "<group>"; };
97F996D02AFB29C700194A6F /* ggml-backend.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = "ggml-backend.c"; sourceTree = "<group>"; };
97F996D12AFB29C700194A6F /* ggml.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = ggml.h; sourceTree = "<group>"; };
97F996D22AFB29C700194A6F /* ggml-alloc.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = "ggml-alloc.c"; sourceTree = "<group>"; };
97F996D32AFB29C700194A6F /* llama.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = llama.cpp; sourceTree = "<group>"; };
97F996DB2AFB29F900194A6F /* llama.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = llama.h; sourceTree = "<group>"; };
97F996DD2AFB2A3F00194A6F /* ggml.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = ggml.c; sourceTree = "<group>"; };
97FAF1B12B0D4DA900E23209 /* grammar-parser.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = "grammar-parser.cpp"; path = "common/grammar-parser.cpp"; sourceTree = "<group>"; };
97FAF1B22B0D4DA900E23209 /* sampling.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = sampling.cpp; path = common/sampling.cpp; sourceTree = "<group>"; };
97FFE3052B056DD100B12DEC /* log.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; name = log.h; path = common/log.h; sourceTree = "<group>"; };
97FFE3062B056DD100B12DEC /* grammar-parser.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; name = "grammar-parser.h"; path = "common/grammar-parser.h"; sourceTree = "<group>"; };
97FFE3072B056DD100B12DEC /* sampling.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; name = sampling.h; path = common/sampling.h; sourceTree = "<group>"; };
/* End PBXFileReference section */
/* Begin PBXFrameworksBuildPhase section */
972079A32AFAA818008A9E53 /* Frameworks */ = {
isa = PBXFrameworksBuildPhase;
buildActionMask = 2147483647;
files = (
972079C42AFAACEF008A9E53 /* Accelerate.framework in Frameworks */,
);
runOnlyForDeploymentPostprocessing = 0;
};
/* End PBXFrameworksBuildPhase section */
/* Begin PBXGroup section */
9720799C2AFAA818008A9E53 = {
isa = PBXGroup;
children = (
97B76BF12B199E040019D706 /* swift */,
979D479D2B081F3E00EF8149 /* llama-module.modulemap */,
97FFE3062B056DD100B12DEC /* grammar-parser.h */,
97FAF1B12B0D4DA900E23209 /* grammar-parser.cpp */,
97FFE3052B056DD100B12DEC /* log.h */,
97FFE3072B056DD100B12DEC /* sampling.h */,
97FAF1B22B0D4DA900E23209 /* sampling.cpp */,
97F217F82B044111008B81FE /* common.h */,
97F217FA2B044120008B81FE /* common.cpp */,
97F996D12AFB29C700194A6F /* ggml.h */,
97F996DD2AFB2A3F00194A6F /* ggml.c */,
979E9D0A2B60388300B5F255 /* ggml-alloc.h */,
97F996D22AFB29C700194A6F /* ggml-alloc.c */,
975AAE532B5F600F0079F713 /* ggml-backend.h */,
97F996D02AFB29C700194A6F /* ggml-backend.c */,
97F996CF2AFB29C700194A6F /* ggml-metal.m */,
97F996CD2AFB29C600194A6F /* ggml-metal.metal */,
97F996CE2AFB29C600194A6F /* ggml-quants.c */,
97F996DB2AFB29F900194A6F /* llama.h */,
97F996D32AFB29C700194A6F /* llama.cpp */,
972079A72AFAA818008A9E53 /* Products */,
972079C22AFAACEF008A9E53 /* Frameworks */,
);
sourceTree = "<group>";
};
972079A72AFAA818008A9E53 /* Products */ = {
isa = PBXGroup;
children = (
972079A62AFAA818008A9E53 /* llama.framework */,
);
name = Products;
sourceTree = "<group>";
};
972079C22AFAACEF008A9E53 /* Frameworks */ = {
isa = PBXGroup;
children = (
972079C32AFAACEF008A9E53 /* Accelerate.framework */,
);
name = Frameworks;
sourceTree = "<group>";
};
97B76BF12B199E040019D706 /* swift */ = {
isa = PBXGroup;
children = (
97B76BF22B199E890019D706 /* include */,
97B76BF62B199E890019D706 /* tokenize.cpp */,
97B76BF52B199E890019D706 /* vector.cpp */,
);
path = swift;
sourceTree = "<group>";
};
97B76BF22B199E890019D706 /* include */ = {
isa = PBXGroup;
children = (
97B76BF32B199E890019D706 /* tokenize.h */,
97B76BF42B199E890019D706 /* vector.h */,
);
path = include;
sourceTree = "<group>";
};
/* End PBXGroup section */
/* Begin PBXHeadersBuildPhase section */
972079A12AFAA818008A9E53 /* Headers */ = {
isa = PBXHeadersBuildPhase;
buildActionMask = 2147483647;
files = (
97F996DC2AFB29F900194A6F /* llama.h in Headers */,
97F996D82AFB29C700194A6F /* ggml.h in Headers */,
979E9D0B2B60388300B5F255 /* ggml-alloc.h in Headers */,
975AAE542B5F600F0079F713 /* ggml-backend.h in Headers */,
97F217F92B044111008B81FE /* common.h in Headers */,
97FFE30A2B056DD100B12DEC /* sampling.h in Headers */,
97FFE3092B056DD100B12DEC /* grammar-parser.h in Headers */,
97FFE3082B056DD100B12DEC /* log.h in Headers */,
97B76BF82B199E890019D706 /* vector.h in Headers */,
97B76BF72B199E890019D706 /* tokenize.h in Headers */,
);
runOnlyForDeploymentPostprocessing = 0;
};
/* End PBXHeadersBuildPhase section */
/* Begin PBXNativeTarget section */
972079A52AFAA818008A9E53 /* llama */ = {
isa = PBXNativeTarget;
buildConfigurationList = 972079AD2AFAA818008A9E53 /* Build configuration list for PBXNativeTarget "llama" */;
buildPhases = (
972079A12AFAA818008A9E53 /* Headers */,
972079A22AFAA818008A9E53 /* Sources */,
972079A32AFAA818008A9E53 /* Frameworks */,
972079A42AFAA818008A9E53 /* Resources */,
);
buildRules = (
);
dependencies = (
);
name = llama;
productName = llama;
productReference = 972079A62AFAA818008A9E53 /* llama.framework */;
productType = "com.apple.product-type.framework";
};
/* End PBXNativeTarget section */
/* Begin PBXProject section */
9720799D2AFAA818008A9E53 /* Project object */ = {
isa = PBXProject;
attributes = {
BuildIndependentTargetsInParallel = 1;
LastUpgradeCheck = 1520;
TargetAttributes = {
972079A52AFAA818008A9E53 = {
CreatedOnToolsVersion = 15.0;
};
};
};
buildConfigurationList = 972079A02AFAA818008A9E53 /* Build configuration list for PBXProject "llama" */;
compatibilityVersion = "Xcode 14.0";
developmentRegion = en;
hasScannedForEncodings = 0;
knownRegions = (
en,
Base,
);
mainGroup = 9720799C2AFAA818008A9E53;
productRefGroup = 972079A72AFAA818008A9E53 /* Products */;
projectDirPath = "";
projectRoot = "";
targets = (
972079A52AFAA818008A9E53 /* llama */,
);
};
/* End PBXProject section */
/* Begin PBXResourcesBuildPhase section */
972079A42AFAA818008A9E53 /* Resources */ = {
isa = PBXResourcesBuildPhase;
buildActionMask = 2147483647;
files = (
97F996D42AFB29C700194A6F /* ggml-metal.metal in Resources */,
);
runOnlyForDeploymentPostprocessing = 0;
};
/* End PBXResourcesBuildPhase section */
/* Begin PBXSourcesBuildPhase section */
972079A22AFAA818008A9E53 /* Sources */ = {
isa = PBXSourcesBuildPhase;
buildActionMask = 2147483647;
files = (
97F996E02AFB2AE100194A6F /* llama.h in Sources */,
97F996DA2AFB29C700194A6F /* llama.cpp in Sources */,
97F996DF2AFB2AE100194A6F /* ggml.h in Sources */,
97F996DE2AFB2A3F00194A6F /* ggml.c in Sources */,
97F996D92AFB29C700194A6F /* ggml-alloc.c in Sources */,
97F996D72AFB29C700194A6F /* ggml-backend.c in Sources */,
97F996D52AFB29C700194A6F /* ggml-quants.c in Sources */,
97F996D62AFB29C700194A6F /* ggml-metal.m in Sources */,
97F217FC2B04413F008B81FE /* common.h in Sources */,
97F217FB2B044120008B81FE /* common.cpp in Sources */,
9789E9FC2B0D49BA00EB6AD2 /* sampling.h in Sources */,
97FAF1B42B0D4DA900E23209 /* sampling.cpp in Sources */,
9789E9FD2B0D49BD00EB6AD2 /* grammar-parser.h in Sources */,
97FAF1B32B0D4DA900E23209 /* grammar-parser.cpp in Sources */,
9789E9FE2B0D49BF00EB6AD2 /* log.h in Sources */,
97B76BFB2B199EC60019D706 /* tokenize.h in Sources */,
97B76BFD2B199EC60019D706 /* tokenize.cpp in Sources */,
97B76BFC2B199EC60019D706 /* vector.h in Sources */,
97B76BF92B199E890019D706 /* vector.cpp in Sources */,
);
runOnlyForDeploymentPostprocessing = 0;
};
/* End PBXSourcesBuildPhase section */
/* Begin XCBuildConfiguration section */
972079AB2AFAA818008A9E53 /* Debug */ = {
isa = XCBuildConfiguration;
buildSettings = {
ALWAYS_SEARCH_USER_PATHS = NO;
ASSETCATALOG_COMPILER_GENERATE_SWIFT_ASSET_SYMBOL_EXTENSIONS = YES;
CLANG_ANALYZER_NONNULL = YES;
CLANG_ANALYZER_NUMBER_OBJECT_CONVERSION = YES_AGGRESSIVE;
CLANG_CXX_LANGUAGE_STANDARD = "gnu++20";
CLANG_ENABLE_MODULES = YES;
CLANG_ENABLE_OBJC_ARC = YES;
CLANG_ENABLE_OBJC_WEAK = YES;
CLANG_WARN_BLOCK_CAPTURE_AUTORELEASING = YES;
CLANG_WARN_BOOL_CONVERSION = YES;
CLANG_WARN_COMMA = YES;
CLANG_WARN_CONSTANT_CONVERSION = YES;
CLANG_WARN_DEPRECATED_OBJC_IMPLEMENTATIONS = YES;
CLANG_WARN_DIRECT_OBJC_ISA_USAGE = YES_ERROR;
CLANG_WARN_DOCUMENTATION_COMMENTS = YES;
CLANG_WARN_EMPTY_BODY = YES;
CLANG_WARN_ENUM_CONVERSION = YES;
CLANG_WARN_INFINITE_RECURSION = YES;
CLANG_WARN_INT_CONVERSION = YES;
CLANG_WARN_NON_LITERAL_NULL_CONVERSION = YES;
CLANG_WARN_OBJC_IMPLICIT_RETAIN_SELF = YES;
CLANG_WARN_OBJC_LITERAL_CONVERSION = YES;
CLANG_WARN_OBJC_ROOT_CLASS = YES_ERROR;
CLANG_WARN_QUOTED_INCLUDE_IN_FRAMEWORK_HEADER = YES;
CLANG_WARN_RANGE_LOOP_ANALYSIS = YES;
CLANG_WARN_STRICT_PROTOTYPES = YES;
CLANG_WARN_SUSPICIOUS_MOVE = YES;
CLANG_WARN_UNGUARDED_AVAILABILITY = YES_AGGRESSIVE;
CLANG_WARN_UNREACHABLE_CODE = YES;
CLANG_WARN__DUPLICATE_METHOD_MATCH = YES;
COPY_PHASE_STRIP = NO;
CURRENT_PROJECT_VERSION = 1;
DEBUG_INFORMATION_FORMAT = dwarf;
ENABLE_STRICT_OBJC_MSGSEND = YES;
ENABLE_TESTABILITY = YES;
ENABLE_USER_SCRIPT_SANDBOXING = YES;
GCC_C_LANGUAGE_STANDARD = gnu17;
GCC_DYNAMIC_NO_PIC = NO;
GCC_NO_COMMON_BLOCKS = YES;
GCC_OPTIMIZATION_LEVEL = 0;
GCC_PREPROCESSOR_DEFINITIONS = (
"DEBUG=1",
"$(inherited)",
);
GCC_WARN_64_TO_32_BIT_CONVERSION = YES;
GCC_WARN_ABOUT_RETURN_TYPE = YES_ERROR;
GCC_WARN_UNDECLARED_SELECTOR = YES;
GCC_WARN_UNINITIALIZED_AUTOS = YES_AGGRESSIVE;
GCC_WARN_UNUSED_FUNCTION = YES;
GCC_WARN_UNUSED_VARIABLE = YES;
IPHONEOS_DEPLOYMENT_TARGET = 14.0;
LOCALIZATION_PREFERS_STRING_CATALOGS = YES;
MACOSX_DEPLOYMENT_TARGET = 12.0;
MTL_ENABLE_DEBUG_INFO = INCLUDE_SOURCE;
MTL_FAST_MATH = YES;
ONLY_ACTIVE_ARCH = YES;
SDKROOT = iphoneos;
SWIFT_ACTIVE_COMPILATION_CONDITIONS = "DEBUG $(inherited)";
SWIFT_OPTIMIZATION_LEVEL = "-Onone";
TVOS_DEPLOYMENT_TARGET = 14.0;
VERSIONING_SYSTEM = "apple-generic";
VERSION_INFO_PREFIX = "";
WATCHOS_DEPLOYMENT_TARGET = 10.0;
XROS_DEPLOYMENT_TARGET = 1.0;
};
name = Debug;
};
972079AC2AFAA818008A9E53 /* Release */ = {
isa = XCBuildConfiguration;
buildSettings = {
ALWAYS_SEARCH_USER_PATHS = NO;
ASSETCATALOG_COMPILER_GENERATE_SWIFT_ASSET_SYMBOL_EXTENSIONS = YES;
CLANG_ANALYZER_NONNULL = YES;
CLANG_ANALYZER_NUMBER_OBJECT_CONVERSION = YES_AGGRESSIVE;
CLANG_CXX_LANGUAGE_STANDARD = "gnu++20";
CLANG_ENABLE_MODULES = YES;
CLANG_ENABLE_OBJC_ARC = YES;
CLANG_ENABLE_OBJC_WEAK = YES;
CLANG_WARN_BLOCK_CAPTURE_AUTORELEASING = YES;
CLANG_WARN_BOOL_CONVERSION = YES;
CLANG_WARN_COMMA = YES;
CLANG_WARN_CONSTANT_CONVERSION = YES;
CLANG_WARN_DEPRECATED_OBJC_IMPLEMENTATIONS = YES;
CLANG_WARN_DIRECT_OBJC_ISA_USAGE = YES_ERROR;
CLANG_WARN_DOCUMENTATION_COMMENTS = YES;
CLANG_WARN_EMPTY_BODY = YES;
CLANG_WARN_ENUM_CONVERSION = YES;
CLANG_WARN_INFINITE_RECURSION = YES;
CLANG_WARN_INT_CONVERSION = YES;
CLANG_WARN_NON_LITERAL_NULL_CONVERSION = YES;
CLANG_WARN_OBJC_IMPLICIT_RETAIN_SELF = YES;
CLANG_WARN_OBJC_LITERAL_CONVERSION = YES;
CLANG_WARN_OBJC_ROOT_CLASS = YES_ERROR;
CLANG_WARN_QUOTED_INCLUDE_IN_FRAMEWORK_HEADER = YES;
CLANG_WARN_RANGE_LOOP_ANALYSIS = YES;
CLANG_WARN_STRICT_PROTOTYPES = YES;
CLANG_WARN_SUSPICIOUS_MOVE = YES;
CLANG_WARN_UNGUARDED_AVAILABILITY = YES_AGGRESSIVE;
CLANG_WARN_UNREACHABLE_CODE = YES;
CLANG_WARN__DUPLICATE_METHOD_MATCH = YES;
COPY_PHASE_STRIP = NO;
CURRENT_PROJECT_VERSION = 1;
DEBUG_INFORMATION_FORMAT = "dwarf-with-dsym";
ENABLE_NS_ASSERTIONS = NO;
ENABLE_STRICT_OBJC_MSGSEND = YES;
ENABLE_USER_SCRIPT_SANDBOXING = YES;
GCC_C_LANGUAGE_STANDARD = gnu17;
GCC_NO_COMMON_BLOCKS = YES;
GCC_WARN_64_TO_32_BIT_CONVERSION = YES;
GCC_WARN_ABOUT_RETURN_TYPE = YES_ERROR;
GCC_WARN_UNDECLARED_SELECTOR = YES;
GCC_WARN_UNINITIALIZED_AUTOS = YES_AGGRESSIVE;
GCC_WARN_UNUSED_FUNCTION = YES;
GCC_WARN_UNUSED_VARIABLE = YES;
IPHONEOS_DEPLOYMENT_TARGET = 14.0;
LOCALIZATION_PREFERS_STRING_CATALOGS = YES;
MACOSX_DEPLOYMENT_TARGET = 12.0;
MTL_ENABLE_DEBUG_INFO = NO;
MTL_FAST_MATH = YES;
SDKROOT = iphoneos;
SWIFT_COMPILATION_MODE = wholemodule;
TVOS_DEPLOYMENT_TARGET = 14.0;
VALIDATE_PRODUCT = YES;
VERSIONING_SYSTEM = "apple-generic";
VERSION_INFO_PREFIX = "";
WATCHOS_DEPLOYMENT_TARGET = 10.0;
XROS_DEPLOYMENT_TARGET = 1.0;
};
name = Release;
};
972079AE2AFAA818008A9E53 /* Debug */ = {
isa = XCBuildConfiguration;
buildSettings = {
ALWAYS_SEARCH_USER_PATHS = NO;
BUILD_LIBRARY_FOR_DISTRIBUTION = YES;
CLANG_ALLOW_NON_MODULAR_INCLUDES_IN_FRAMEWORK_MODULES = NO;
CLANG_CXX_LANGUAGE_STANDARD = "c++0x";
CLANG_WARN_QUOTED_INCLUDE_IN_FRAMEWORK_HEADER = YES;
CODE_SIGN_IDENTITY = "";
CODE_SIGN_STYLE = Automatic;
CURRENT_PROJECT_VERSION = 1;
DEFINES_MODULE = YES;
DYLIB_COMPATIBILITY_VERSION = 1;
DYLIB_CURRENT_VERSION = 1;
DYLIB_INSTALL_NAME_BASE = "@rpath";
ENABLE_MODULE_VERIFIER = NO;
GCC_OPTIMIZATION_LEVEL = 3;
GCC_PREPROCESSOR_DEFINITIONS = (
"$(inherited)",
"LLAMA_BUILD_NUMBER=1",
"LLAMA_COMMIT=\\\"development\\\"",
"LLAMA_COMPILER=\\\"swift-clang\\\"",
"LLAMA_BUILD_TARGET=\\\"iOS\\\"",
);
"GCC_PREPROCESSOR_DEFINITIONS[arch=arm64]" = (
"$(inherited)",
"GGML_USE_ACCELERATE=1",
"GGML_USE_METAL=1",
);
"GCC_PREPROCESSOR_DEFINITIONS[arch=x86_64]" = (
"$(inherited)",
"GGML_USE_ACCELERATE=1",
);
GENERATE_INFOPLIST_FILE = YES;
INFOPLIST_KEY_NSHumanReadableCopyright = "";
INSTALL_PATH = "$(LOCAL_LIBRARY_DIR)/Frameworks";
IPHONEOS_DEPLOYMENT_TARGET = 14.0;
LD_RUNPATH_SEARCH_PATHS = (
"$(inherited)",
"@executable_path/Frameworks",
"@loader_path/Frameworks",
);
MARKETING_VERSION = 1.0;
MODULEMAP_FILE = "llama-module.modulemap";
MODULE_VERIFIER_SUPPORTED_LANGUAGES = "objective-c objective-c++";
MODULE_VERIFIER_SUPPORTED_LANGUAGE_STANDARDS = "gnu17 gnu++20";
OTHER_CFLAGS = "-DNDEBUG";
PRODUCT_BUNDLE_IDENTIFIER = edu.stanford.llama;
PRODUCT_NAME = "$(TARGET_NAME:c99extidentifier)";
SKIP_INSTALL = YES;
SUPPORTED_PLATFORMS = "iphoneos iphonesimulator macosx xros xrsimulator";
SUPPORTS_MACCATALYST = YES;
SUPPORTS_MAC_DESIGNED_FOR_IPHONE_IPAD = YES;
SWIFT_EMIT_LOC_STRINGS = YES;
SWIFT_VERSION = 5.0;
TARGETED_DEVICE_FAMILY = "1,2,6";
TVOS_DEPLOYMENT_TARGET = 14.0;
USER_HEADER_SEARCH_PATHS = "";
WARNING_CFLAGS = "-Wno-shorten-64-to-32";
WATCHOS_DEPLOYMENT_TARGET = 10.0;
};
name = Debug;
};
972079AF2AFAA818008A9E53 /* Release */ = {
isa = XCBuildConfiguration;
buildSettings = {
ALWAYS_SEARCH_USER_PATHS = NO;
BUILD_LIBRARY_FOR_DISTRIBUTION = YES;
CLANG_ALLOW_NON_MODULAR_INCLUDES_IN_FRAMEWORK_MODULES = NO;
CLANG_CXX_LANGUAGE_STANDARD = "c++0x";
CLANG_WARN_QUOTED_INCLUDE_IN_FRAMEWORK_HEADER = YES;
CODE_SIGN_IDENTITY = "";
CURRENT_PROJECT_VERSION = 1;
DEFINES_MODULE = YES;
DYLIB_COMPATIBILITY_VERSION = 1;
DYLIB_CURRENT_VERSION = 1;
DYLIB_INSTALL_NAME_BASE = "@rpath";
ENABLE_MODULE_VERIFIER = NO;
GCC_OPTIMIZATION_LEVEL = 3;
GCC_PREPROCESSOR_DEFINITIONS = (
"LLAMA_BUILD_NUMBER=1",
"LLAMA_COMMIT=\\\"development\\\"",
"LLAMA_COMPILER=\\\"swift-clang\\\"",
"LLAMA_BUILD_TARGET=\\\"iOS\\\"",
);
"GCC_PREPROCESSOR_DEFINITIONS[arch=arm64]" = (
"$(inherited)",
"GGML_USE_ACCELERATE=1",
"GGML_USE_METAL=1",
);
"GCC_PREPROCESSOR_DEFINITIONS[arch=x86_64]" = (
"$(inherited)",
"GGML_USE_ACCELERATE=1",
);
GENERATE_INFOPLIST_FILE = YES;
INFOPLIST_KEY_NSHumanReadableCopyright = "";
INSTALL_PATH = "$(LOCAL_LIBRARY_DIR)/Frameworks";
IPHONEOS_DEPLOYMENT_TARGET = 14.0;
LD_RUNPATH_SEARCH_PATHS = (
"$(inherited)",
"@executable_path/Frameworks",
"@loader_path/Frameworks",
);
MARKETING_VERSION = 1.0;
MODULEMAP_FILE = "llama-module.modulemap";
MODULE_VERIFIER_SUPPORTED_LANGUAGES = "objective-c objective-c++";
MODULE_VERIFIER_SUPPORTED_LANGUAGE_STANDARDS = "gnu17 gnu++20";
OTHER_CFLAGS = "-DNDEBUG";
PRODUCT_BUNDLE_IDENTIFIER = edu.stanford.llama;
PRODUCT_NAME = "$(TARGET_NAME:c99extidentifier)";
SKIP_INSTALL = YES;
SUPPORTED_PLATFORMS = "iphoneos iphonesimulator macosx xros xrsimulator";
SUPPORTS_MACCATALYST = YES;
SUPPORTS_MAC_DESIGNED_FOR_IPHONE_IPAD = YES;
SWIFT_EMIT_LOC_STRINGS = YES;
SWIFT_VERSION = 5.0;
TARGETED_DEVICE_FAMILY = "1,2,6";
TVOS_DEPLOYMENT_TARGET = 14.0;
USER_HEADER_SEARCH_PATHS = "";
WARNING_CFLAGS = "-Wno-shorten-64-to-32";
WATCHOS_DEPLOYMENT_TARGET = 10.0;
};
name = Release;
};
/* End XCBuildConfiguration section */
/* Begin XCConfigurationList section */
972079A02AFAA818008A9E53 /* Build configuration list for PBXProject "llama" */ = {
isa = XCConfigurationList;
buildConfigurations = (
972079AB2AFAA818008A9E53 /* Debug */,
972079AC2AFAA818008A9E53 /* Release */,
);
defaultConfigurationIsVisible = 0;
defaultConfigurationName = Release;
};
972079AD2AFAA818008A9E53 /* Build configuration list for PBXNativeTarget "llama" */ = {
isa = XCConfigurationList;
buildConfigurations = (
972079AE2AFAA818008A9E53 /* Debug */,
972079AF2AFAA818008A9E53 /* Release */,
);
defaultConfigurationIsVisible = 0;
defaultConfigurationName = Release;
};
/* End XCConfigurationList section */
};
rootObject = 9720799D2AFAA818008A9E53 /* Project object */;
}

8
llama.xcodeproj/project.xcworkspace/xcshareddata/IDEWorkspaceChecks.plist Normal file
View file

@ -0,0 +1,8 @@
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
<key>IDEDidComputeMac32BitWarning</key>
<true/>
</dict>
</plist>

66
llama.xcodeproj/xcshareddata/xcschemes/llama.xcscheme Normal file
View file

@ -0,0 +1,66 @@
<?xml version="1.0" encoding="UTF-8"?>
<Scheme
LastUpgradeVersion = "1520"
version = "1.7">
<BuildAction
parallelizeBuildables = "YES"
buildImplicitDependencies = "YES">
<BuildActionEntries>
<BuildActionEntry
buildForTesting = "YES"
buildForRunning = "YES"
buildForProfiling = "YES"
buildForArchiving = "YES"
buildForAnalyzing = "YES">
<BuildableReference
BuildableIdentifier = "primary"
BlueprintIdentifier = "972079A52AFAA818008A9E53"
BuildableName = "llama.framework"
BlueprintName = "llama"
ReferencedContainer = "container:llama.xcodeproj">
</BuildableReference>
</BuildActionEntry>
</BuildActionEntries>
</BuildAction>
<TestAction
buildConfiguration = "Debug"
selectedDebuggerIdentifier = "Xcode.DebuggerFoundation.Debugger.LLDB"
selectedLauncherIdentifier = "Xcode.DebuggerFoundation.Launcher.LLDB"
shouldUseLaunchSchemeArgsEnv = "YES"
shouldAutocreateTestPlan = "YES">
</TestAction>
<LaunchAction
buildConfiguration = "Debug"
selectedDebuggerIdentifier = "Xcode.DebuggerFoundation.Debugger.LLDB"
selectedLauncherIdentifier = "Xcode.DebuggerFoundation.Launcher.LLDB"
launchStyle = "0"
useCustomWorkingDirectory = "NO"
ignoresPersistentStateOnLaunch = "NO"
debugDocumentVersioning = "YES"
debugServiceExtension = "internal"
allowLocationSimulation = "YES">
</LaunchAction>
<ProfileAction
buildConfiguration = "Release"
shouldUseLaunchSchemeArgsEnv = "YES"
savedToolIdentifier = ""
useCustomWorkingDirectory = "NO"
debugDocumentVersioning = "YES">
<MacroExpansion>
<BuildableReference
BuildableIdentifier = "primary"
BlueprintIdentifier = "972079A52AFAA818008A9E53"
BuildableName = "llama.framework"
BlueprintName = "llama"
ReferencedContainer = "container:llama.xcodeproj">
</BuildableReference>
</MacroExpansion>
</ProfileAction>
<AnalyzeAction
buildConfiguration = "Debug">
</AnalyzeAction>
<ArchiveAction
buildConfiguration = "Release"
revealArchiveInOrganizer = "YES">
</ArchiveAction>
</Scheme>

7
llama.xcworkspace/contents.xcworkspacedata generated Normal file
View file

@ -0,0 +1,7 @@
<?xml version="1.0" encoding="UTF-8"?>
<Workspace
version = "1.0">
<FileRef
location = "group:llama.xcodeproj">
</FileRef>
</Workspace>

8
llama.xcworkspace/xcshareddata/IDEWorkspaceChecks.plist Normal file
View file

@ -0,0 +1,8 @@
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
<key>IDEDidComputeMac32BitWarning</key>
<true/>
</dict>
</plist>

31
swift/include/tokenize.h Normal file
View file

@ -0,0 +1,31 @@
//
// This source file is part of the Stanford Spezi open source project
//
// SPDX-FileCopyrightText: 2022 Stanford University and the project authors (see CONTRIBUTORS.md)
//
// SPDX-License-Identifier: MIT
//
#ifndef tokenize_hpp
#define tokenize_hpp
#include <vector>
#include "common.h"
/// Tokenize a `String` via a given `llama_context`.
std::vector<llama_token> llama_tokenize_with_context(
const struct llama_context * ctx,
const std::string & text,
bool add_bos,
bool special = false);
/// Tokenize a `String` via a given `llama_model`.
std::vector<llama_token> llama_tokenize_with_model(
const struct llama_model * model,
const std::string & text,
bool add_bos,
bool special = false);
#endif

20
swift/include/vector.h Normal file
View file

@ -0,0 +1,20 @@
//
// This source file is part of the Stanford Spezi open source project
//
// SPDX-FileCopyrightText: 2022 Stanford University and the project authors (see CONTRIBUTORS.md)
//
// SPDX-License-Identifier: MIT
//
#ifndef vector_hpp
#define vector_hpp
#include <vector>
#include "common.h"
/// Create an empty `vector` of `llama_seq_id`s that serve as a buffer for batch processing.
const std::vector<llama_seq_id> getLlamaSeqIdVector();
#endif

28
swift/tokenize.cpp Normal file
View file

@ -0,0 +1,28 @@
//
// This source file is part of the Stanford Spezi open source project
//
// SPDX-FileCopyrightText: 2022 Stanford University and the project authors (see CONTRIBUTORS.md)
//
// SPDX-License-Identifier: MIT
//
#include "tokenize.h"
/// Tokenize a `String` via a given `llama_context`.
std::vector<llama_token> llama_tokenize_with_context(
const struct llama_context * ctx,
const std::string & text,
bool add_bos,
bool special) {
return llama_tokenize(ctx, text, add_bos, special);
}
/// Tokenize a `String` via a given `llama_model`.
std::vector<llama_token> llama_tokenize_with_model(
const struct llama_model * model,
const std::string & text,
bool add_bos,
bool special) {
return llama_tokenize(model, text, add_bos, special);
}

16
swift/vector.cpp Normal file
View file

@ -0,0 +1,16 @@
//
// This source file is part of the Stanford Spezi open source project
//
// SPDX-FileCopyrightText: 2022 Stanford University and the project authors (see CONTRIBUTORS.md)
//
// SPDX-License-Identifier: MIT
//
#include "vector.h"
/// Create an empty `vector` of `llama_seq_id`s that serve as a buffer for batch processing.
const std::vector<llama_seq_id> getLlamaSeqIdVector() {
const std::vector<llama_seq_id> vec = { 0 };
return vec;
}