Skip to content

Commit a6b641e

Browse files
CISCCISC
authored
Merge branch 'master' into cisc/unary-reglu-geglu-swiglu
2 parents 8bca3e2 + 27208bf commit a6b641e

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

88 files changed

+4461
-2106
lines changed

‎.github/workflows/build-cmake-pkg.yml‎

Lines changed: 51 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,51 @@
1+
name: Build relocatable cmake package
2+
on:
3+
workflow_dispatch:
4+
workflow_call:
5+
6+
jobs:
7+
linux:
8+
runs-on: ubuntu-24.04
9+
steps:
10+
- uses: actions/checkout@v4
11+
with:
12+
fetch-depth: 0
13+
14+
- name: Install dependencies
15+
run: |
16+
sudo apt update
17+
sudo apt install -y build-essential tcl
18+
19+
- name: Build
20+
run: |
21+
PREFIX="$(pwd)"/inst
22+
cmake -S . -B build -DCMAKE_PREFIX_PATH="$PREFIX" \
23+
-DLLAMA_CURL=OFF -DLLAMA_BUILD_TESTS=OFF -DLLAMA_BUILD_TOOLS=OFF \
24+
-DLLAMA_BUILD_EXAMPLES=OFF -DCMAKE_BUILD_TYPE=Release
25+
cmake --build build --config Release
26+
cmake --install build --prefix "$PREFIX" --config Release
27+
28+
export LLAMA_CONFIG="$PREFIX"/lib/cmake/llama/llama-config.cmake
29+
tclsh <<'EOF'
30+
set build(commit) [string trim [exec git rev-parse --short HEAD]]
31+
set build(number) [string trim [exec git rev-list --count HEAD]]
32+
set build(version) "0.0.$build(number)"
33+
34+
set llamaconfig [read [open "$env(LLAMA_CONFIG)" r]]
35+
set checks [list "set\\(LLAMA_VERSION \\s+$build(version)\\)" \
36+
"set\\(LLAMA_BUILD_COMMIT\\s+$build(commit)\\)" \
37+
"set\\(LLAMA_BUILD_NUMBER\\s+$build(number)\\)"]
38+
39+
puts -nonewline "Checking llama-config.cmake version... "
40+
foreach check $checks {
41+
if {![regexp -expanded -- $check $llamaconfig]} {
42+
puts "\"$check\" failed!"
43+
exit 1
44+
}
45+
}
46+
puts "success."
47+
EOF
48+
49+
cd examples/simple-cmake-pkg
50+
cmake -S . -B build -DCMAKE_PREFIX_PATH="$PREFIX"/lib/cmake
51+
cmake --build build

‎.github/workflows/build.yml‎

Lines changed: 53 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -5,10 +5,43 @@ on:
55
push:
66
branches:
77
- master
8-
paths: ['.github/workflows/build.yml', '.github/workflows/build-linux-cross.yml', '**/CMakeLists.txt', '**/.cmake', '**/*.h', '**/*.hpp', '**/*.c', '**/*.cpp', '**/*.cu', '**/*.cuh', '**/*.swift', '**/*.m', '**/*.metal', '**/*.comp']
8+
paths: [
9+
'.github/workflows/build.yml',
10+
'.github/workflows/build-linux-cross.yml',
11+
'.github/workflows/build-cmake-pkg.yml',
12+
'**/CMakeLists.txt',
13+
'**/.cmake',
14+
'**/*.h',
15+
'**/*.hpp',
16+
'**/*.c',
17+
'**/*.cpp',
18+
'**/*.cu',
19+
'**/*.cuh',
20+
'**/*.swift',
21+
'**/*.m',
22+
'**/*.metal',
23+
'**/*.comp'
24+
]
25+
926
pull_request:
1027
types: [opened, synchronize, reopened]
11-
paths: ['.github/workflows/build.yml', '.github/workflows/build-linux-cross.yml', '**/CMakeLists.txt', '**/.cmake', '**/*.h', '**/*.hpp', '**/*.c', '**/*.cpp', '**/*.cu', '**/*.cuh', '**/*.swift', '**/*.m', '**/*.metal', '**/*.comp']
28+
paths: [
29+
'.github/workflows/build.yml',
30+
'.github/workflows/build-linux-cross.yml',
31+
'.github/workflows/build-cmake-pkg.yml',
32+
'**/CMakeLists.txt',
33+
'**/.cmake',
34+
'**/*.h',
35+
'**/*.hpp',
36+
'**/*.c',
37+
'**/*.cpp',
38+
'**/*.cu',
39+
'**/*.cuh',
40+
'**/*.swift',
41+
'**/*.m',
42+
'**/*.metal',
43+
'**/*.comp'
44+
]
1245

1346
concurrency:
1447
group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
@@ -478,6 +511,9 @@ jobs:
478511
build-linux-cross:
479512
uses: ./.github/workflows/build-linux-cross.yml
480513

514+
build-cmake-pkg:
515+
uses: ./.github/workflows/build-cmake-pkg.yml
516+
481517
macOS-latest-cmake-ios:
482518
runs-on: macos-latest
483519

@@ -628,7 +664,7 @@ jobs:
628664
./build-xcframework.sh
629665
630666
windows-msys2:
631-
runs-on: windows-latest
667+
runs-on: windows-2025
632668

633669
strategy:
634670
fail-fast: false
@@ -678,27 +714,33 @@ jobs:
678714
cmake --build build --config ${{ matrix.build }} -j $(nproc)
679715
680716
windows-latest-cmake:
681-
runs-on: windows-latest
717+
runs-on: windows-2025
682718

683719
env:
684720
OPENBLAS_VERSION: 0.3.23
685721
SDE_VERSION: 9.33.0-2024-01-07
686-
VULKAN_VERSION: 1.4.309.0
722+
VULKAN_VERSION: 1.4.313.2
687723

688724
strategy:
689725
matrix:
690726
include:
691727
- build: 'cpu-x64 (static)'
728+
arch: 'x64'
692729
defines: '-G "Ninja Multi-Config" -D CMAKE_TOOLCHAIN_FILE=cmake/x64-windows-llvm.cmake -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_RPC=ON -DBUILD_SHARED_LIBS=OFF'
693730
- build: 'openblas-x64'
731+
arch: 'x64'
694732
defines: '-G "Ninja Multi-Config" -D CMAKE_TOOLCHAIN_FILE=cmake/x64-windows-llvm.cmake -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_RPC=ON -DGGML_BACKEND_DL=ON -DGGML_CPU_ALL_VARIANTS=ON -DGGML_OPENMP=OFF -DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS -DBLAS_INCLUDE_DIRS="$env:RUNNER_TEMP/openblas/include" -DBLAS_LIBRARIES="$env:RUNNER_TEMP/openblas/lib/openblas.lib"'
695733
- build: 'vulkan-x64'
734+
arch: 'x64'
696735
defines: '-DCMAKE_BUILD_TYPE=Release -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_RPC=ON -DGGML_BACKEND_DL=ON -DGGML_CPU_ALL_VARIANTS=ON -DGGML_VULKAN=ON'
697736
- build: 'llvm-arm64'
737+
arch: 'arm64'
698738
defines: '-G "Ninja Multi-Config" -D CMAKE_TOOLCHAIN_FILE=cmake/arm64-windows-llvm.cmake -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON'
699739
- build: 'llvm-arm64-opencl-adreno'
740+
arch: 'arm64'
700741
defines: '-G "Ninja Multi-Config" -D CMAKE_TOOLCHAIN_FILE=cmake/arm64-windows-llvm.cmake -DCMAKE_PREFIX_PATH="$env:RUNNER_TEMP/opencl-arm64-release" -DGGML_OPENCL=ON -DGGML_OPENCL_USE_ADRENO_KERNELS=ON'
701742
# - build: 'kompute-x64'
743+
# arch: 'x64'
702744
# defines: '-G "Ninja Multi-Config" -D CMAKE_TOOLCHAIN_FILE=cmake/x64-windows-llvm.cmake -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_RPC=ON -DGGML_BACKEND_DL=ON -DGGML_CPU_ALL_VARIANTS=ON -DGGML_OPENMP=OFF -DGGML_KOMPUTE=ON -DKOMPUTE_OPT_DISABLE_VULKAN_VERSION_CHECK=ON'
703745

704746
steps:
@@ -736,7 +778,7 @@ jobs:
736778
id: get_vulkan
737779
if: ${{ matrix.build == 'kompute-x64' || matrix.build == 'vulkan-x64' }}
738780
run: |
739-
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"
781+
curl.exe -o $env:RUNNER_TEMP/VulkanSDK-Installer.exe -L "https://sdk.lunarg.com/sdk/download/${env:VULKAN_VERSION}/windows/vulkansdk-windows-X64-${env:VULKAN_VERSION}.exe"
740782
& "$env:RUNNER_TEMP\VulkanSDK-Installer.exe" --accept-licenses --default-answer --confirm-command install
741783
Add-Content $env:GITHUB_ENV "VULKAN_SDK=C:\VulkanSDK\${env:VULKAN_VERSION}"
742784
Add-Content $env:GITHUB_PATH "C:\VulkanSDK\${env:VULKAN_VERSION}\bin"
@@ -769,6 +811,8 @@ jobs:
769811
- name: libCURL
770812
id: get_libcurl
771813
uses: ./.github/actions/windows-setup-curl
814+
with:
815+
architecture: ${{ matrix.arch == 'x64' && 'win64' || 'win64a' }}
772816

773817
- name: Build
774818
id: cmake_build
@@ -789,7 +833,7 @@ jobs:
789833
790834
- name: Test
791835
id: cmake_test
792-
if: ${{ matrix.build != 'llvm-arm64' && matrix.build != 'llvm-arm64-opencl-adreno' }}
836+
if: ${{ matrix.arch == 'x64' }}
793837
run: |
794838
cd build
795839
ctest -L main -C Release --verbose --timeout 900
@@ -894,7 +938,7 @@ jobs:
894938
cmake --build build --config Release
895939
896940
windows-latest-cmake-sycl:
897-
runs-on: windows-latest
941+
runs-on: windows-2022
898942

899943
defaults:
900944
run:
@@ -928,7 +972,7 @@ jobs:
928972

929973
windows-latest-cmake-hip:
930974
if: ${{ github.event.inputs.create_release != 'true' }}
931-
runs-on: windows-latest
975+
runs-on: windows-2022
932976

933977
steps:
934978
- name: Clone

‎.github/workflows/release.yml‎

Lines changed: 8 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -235,7 +235,7 @@ jobs:
235235
name: llama-bin-ubuntu-vulkan-x64.zip
236236

237237
windows-cpu:
238-
runs-on: windows-latest
238+
runs-on: windows-2025
239239

240240
strategy:
241241
matrix:
@@ -271,7 +271,7 @@ jobs:
271271
env:
272272
CURL_PATH: ${{ steps.get_libcurl.outputs.curl_path }}
273273
run: |
274-
call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvarsall.bat" ${{ matrix.arch }}
274+
call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvarsall.bat" ${{ matrix.arch == 'x64' && 'x64' || 'amd64_arm64' }}
275275
cmake -S . -B build -G "Ninja Multi-Config" ^
276276
-D CMAKE_TOOLCHAIN_FILE=cmake/${{ matrix.arch }}-windows-llvm.cmake ^
277277
-DGGML_NATIVE=OFF ^
@@ -288,7 +288,7 @@ jobs:
288288
CURL_PATH: ${{ steps.get_libcurl.outputs.curl_path }}
289289
run: |
290290
Copy-Item $env:CURL_PATH\bin\libcurl-${{ matrix.arch }}.dll .\build\bin\Release\
291-
Copy-Item "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Redist\MSVC\14.42.34433\debug_nonredist\${{ matrix.arch }}\Microsoft.VC143.OpenMP.LLVM\libomp140.${{ matrix.arch == 'x64' && 'x86_64' || 'aarch64' }}.dll" .\build\bin\Release\
291+
Copy-Item "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Redist\MSVC\14.44.35112\debug_nonredist\${{ matrix.arch }}\Microsoft.VC143.OpenMP.LLVM\libomp140.${{ matrix.arch == 'x64' && 'x86_64' || 'aarch64' }}.dll" .\build\bin\Release\
292292
7z a llama-bin-win-cpu-${{ matrix.arch }}.zip .\build\bin\Release\*
293293
294294
- name: Upload artifacts
@@ -298,11 +298,11 @@ jobs:
298298
name: llama-bin-win-cpu-${{ matrix.arch }}.zip
299299

300300
windows:
301-
runs-on: windows-latest
301+
runs-on: windows-2025
302302

303303
env:
304304
OPENBLAS_VERSION: 0.3.23
305-
VULKAN_VERSION: 1.4.309.0
305+
VULKAN_VERSION: 1.4.313.2
306306

307307
strategy:
308308
matrix:
@@ -332,7 +332,7 @@ jobs:
332332
id: get_vulkan
333333
if: ${{ matrix.backend == 'vulkan' }}
334334
run: |
335-
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"
335+
curl.exe -o $env:RUNNER_TEMP/VulkanSDK-Installer.exe -L "https://sdk.lunarg.com/sdk/download/${env:VULKAN_VERSION}/windows/vulkansdk-windows-X64-${env:VULKAN_VERSION}.exe"
336336
& "$env:RUNNER_TEMP\VulkanSDK-Installer.exe" --accept-licenses --default-answer --confirm-command install
337337
Add-Content $env:GITHUB_ENV "VULKAN_SDK=C:\VulkanSDK\${env:VULKAN_VERSION}"
338338
Add-Content $env:GITHUB_PATH "C:\VulkanSDK\${env:VULKAN_VERSION}\bin"
@@ -448,7 +448,7 @@ jobs:
448448
name: cudart-llama-bin-win-cuda-${{ matrix.cuda }}-x64.zip
449449

450450
windows-sycl:
451-
runs-on: windows-latest
451+
runs-on: windows-2022
452452

453453
defaults:
454454
run:
@@ -520,7 +520,7 @@ jobs:
520520
name: llama-bin-win-sycl-x64.zip
521521

522522
windows-hip:
523-
runs-on: windows-latest
523+
runs-on: windows-2022
524524

525525
strategy:
526526
matrix:

‎CMakeLists.txt‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -95,7 +95,7 @@ endif()
9595
if (NOT DEFINED LLAMA_BUILD_COMMIT)
9696
set(LLAMA_BUILD_COMMIT ${BUILD_COMMIT})
9797
endif()
98-
set(LLAMA_INSTALL_VERSION 0.0.${BUILD_NUMBER})
98+
set(LLAMA_INSTALL_VERSION 0.0.${LLAMA_BUILD_NUMBER})
9999

100100
# override ggml options
101101
set(GGML_ALL_WARNINGS ${LLAMA_ALL_WARNINGS})

‎common/json-schema-to-grammar.cpp‎

Lines changed: 3 additions & 46 deletions
Original file line numberDiff line numberDiff line change
@@ -41,49 +41,6 @@ static std::string build_repetition(const std::string & item_rule, int min_items
4141
return result;
4242
}
4343

44-
/* Minimalistic replacement for std::string_view, which is only available from C++17 onwards */
45-
class string_view {
46-
const std::string & _str;
47-
const size_t _start;
48-
const size_t _end;
49-
public:
50-
string_view(const std::string & str, size_t start = 0, size_t end = std::string::npos) : _str(str), _start(start), _end(end == std::string::npos ? str.length() : end) {}
51-
52-
size_t size() const {
53-
return _end - _start;
54-
}
55-
56-
size_t length() const {
57-
return size();
58-
}
59-
60-
operator std::string() const {
61-
return str();
62-
}
63-
64-
std::string str() const {
65-
return _str.substr(_start, _end - _start);
66-
}
67-
68-
string_view substr(size_t pos, size_t len = std::string::npos) const {
69-
return string_view(_str, _start + pos, len == std::string::npos ? _end : _start + pos + len);
70-
}
71-
72-
char operator[](size_t pos) const {
73-
auto index = _start + pos;
74-
if (index >= _end) {
75-
throw std::out_of_range("string_view index out of range");
76-
}
77-
return _str[_start + pos];
78-
}
79-
80-
bool operator==(const string_view & other) const {
81-
std::string this_str = *this;
82-
std::string other_str = other;
83-
return this_str == other_str;
84-
}
85-
};
86-
8744
static void _build_min_max_int(int min_value, int max_value, std::stringstream & out, int decimals_left = 16, bool top_level = true) {
8845
auto has_min = min_value != std::numeric_limits<int>::min();
8946
auto has_max = max_value != std::numeric_limits<int>::max();
@@ -112,14 +69,14 @@ static void _build_min_max_int(int min_value, int max_value, std::stringstream &
11269
}
11370
out << "}";
11471
};
115-
std::function<void(const string_view &, const string_view &)> uniform_range =
116-
[&](const string_view & from, const string_view & to) {
72+
std::function<void(const std::string_view &, const std::string_view &)> uniform_range =
73+
[&](const std::string_view & from, const std::string_view & to) {
11774
size_t i = 0;
11875
while (i < from.length() && i < to.length() && from[i] == to[i]) {
11976
i++;
12077
}
12178
if (i > 0) {
122-
out << "\"" << from.substr(0, i).str() << "\"";
79+
out << "\"" << from.substr(0, i) << "\"";
12380
}
12481
if (i < from.length() && i < to.length()) {
12582
if (i > 0) {

0 commit comments

Comments
 (0)