Skip to content

Commit bdc8f57

Browse files
Vect0rMVect0rM
committed
ci: add macOS ARM64 build workflow with notarization
Static build (BUILD_SHARED_LIBS=OFF, LLAMA_CURL=OFF, LLAMA_OPENSSL=OFF), Metal + BF16, code signing, notarization via notarytool, GitHub Release. Made-with: Cursor
1 parent 8ba9f12 commit bdc8f57

1 file changed

Lines changed: 218 additions & 0 deletions

File tree

‎.github/workflows/build-turboquant-macos.yml‎

Lines changed: 218 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,218 @@
1+
name: Build & Release TurboQuant (macOS ARM64)
2+
3+
on:
4+
workflow_dispatch:
5+
push:
6+
branches:
7+
- feature/turboquant-kv-cache
8+
paths:
9+
- '.github/workflows/build-turboquant-macos.yml'
10+
- '**/CMakeLists.txt'
11+
- '**/*.h'
12+
- '**/*.hpp'
13+
- '**/*.c'
14+
- '**/*.cpp'
15+
- '**/*.metal'
16+
17+
concurrency:
18+
group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
19+
cancel-in-progress: true
20+
21+
env:
22+
LLAMA_LOG_COLORS: 1
23+
LLAMA_LOG_PREFIX: 1
24+
LLAMA_LOG_TIMESTAMPS: 1
25+
26+
jobs:
27+
macOS-arm64-metal:
28+
runs-on: macos-latest
29+
permissions:
30+
contents: write
31+
32+
steps:
33+
- name: Clone
34+
uses: actions/checkout@v6
35+
36+
- name: Set version tag
37+
id: version
38+
run: |
39+
SHORT_SHA=$(git rev-parse --short HEAD)
40+
TAG="turboquant-macos-arm64-${SHORT_SHA}"
41+
echo "tag=${TAG}" >> "$GITHUB_OUTPUT"
42+
echo "short_sha=${SHORT_SHA}" >> "$GITHUB_OUTPUT"
43+
44+
- name: Import code signing certificate
45+
env:
46+
MACOS_CERTIFICATE_P12: ${{ secrets.MACOS_CERTIFICATE_P12 }}
47+
MACOS_CERTIFICATE_PASSWORD: ${{ secrets.MACOS_CERTIFICATE_PASSWORD }}
48+
KEYCHAIN_PASSWORD: ${{ secrets.KEYCHAIN_PASSWORD }}
49+
run: |
50+
CERTIFICATE_PATH=$RUNNER_TEMP/build_certificate.p12
51+
KEYCHAIN_PATH=$RUNNER_TEMP/app-signing.keychain-db
52+
53+
echo -n "$MACOS_CERTIFICATE_P12" | base64 --decode -o $CERTIFICATE_PATH
54+
55+
security create-keychain -p "$KEYCHAIN_PASSWORD" $KEYCHAIN_PATH
56+
security set-keychain-settings -lut 21600 $KEYCHAIN_PATH
57+
security unlock-keychain -p "$KEYCHAIN_PASSWORD" $KEYCHAIN_PATH
58+
59+
security import $CERTIFICATE_PATH -P "$MACOS_CERTIFICATE_PASSWORD" \
60+
-A -t cert -f pkcs12 -k $KEYCHAIN_PATH
61+
security set-key-partition-list -S apple-tool:,apple: \
62+
-k "$KEYCHAIN_PASSWORD" $KEYCHAIN_PATH
63+
security list-keychain -d user -s $KEYCHAIN_PATH
64+
65+
echo "KEYCHAIN_PATH=$KEYCHAIN_PATH" >> "$GITHUB_ENV"
66+
67+
IDENTITY=$(security find-identity -v -p codesigning $KEYCHAIN_PATH | head -1 | grep -o '".*"' | tr -d '"')
68+
echo "CODESIGN_IDENTITY=$IDENTITY" >> "$GITHUB_ENV"
69+
echo "Signing identity: $IDENTITY"
70+
71+
- name: Build
72+
id: cmake_build
73+
run: |
74+
sysctl -a
75+
cmake -B build \
76+
-DCMAKE_BUILD_TYPE=Release \
77+
-DBUILD_SHARED_LIBS=OFF \
78+
-DLLAMA_CURL=OFF \
79+
-DLLAMA_OPENSSL=OFF \
80+
-DGGML_METAL=ON \
81+
-DGGML_METAL_USE_BF16=ON \
82+
-DGGML_METAL_EMBED_LIBRARY=ON \
83+
-DLLAMA_BUILD_SERVER=ON \
84+
-DLLAMA_BUILD_TOOLS=ON \
85+
-DLLAMA_BUILD_TESTS=OFF \
86+
-DLLAMA_BUILD_EXAMPLES=OFF
87+
cmake --build build --config Release -j $(sysctl -n hw.ncpu)
88+
89+
- name: Verify turbo3 support
90+
run: |
91+
./build/bin/llama-server --help 2>&1 | grep -A2 "cache-type-k" || true
92+
echo "---"
93+
file ./build/bin/llama-server
94+
./build/bin/llama-server --version 2>&1 || true
95+
96+
- name: Verify static binary
97+
run: |
98+
echo "=== All dynamic dependencies (should be system-only) ==="
99+
otool -L build/bin/llama-server
100+
echo "---"
101+
echo "=== Checking for non-system dylibs (should be empty) ==="
102+
if otool -L build/bin/llama-server | grep -vE '/usr/lib|/System|llama-server' | grep '\.dylib'; then
103+
echo "ERROR: Found non-system dynamic dependency!"
104+
exit 1
105+
fi
106+
echo "OK: Only system dylibs — fully self-contained binary"
107+
echo "---"
108+
echo "=== Binary size ==="
109+
ls -lh build/bin/llama-server
110+
111+
- name: Sign binaries
112+
run: |
113+
for bin in build/bin/llama-server build/bin/llama-cli build/bin/llama-bench build/bin/llama-perplexity; do
114+
if [ -f "$bin" ]; then
115+
echo "Signing $bin ..."
116+
codesign --force --options runtime --timestamp \
117+
--entitlements .github/entitlements.plist \
118+
--sign "$CODESIGN_IDENTITY" "$bin"
119+
codesign --verify --verbose "$bin"
120+
fi
121+
done
122+
for lib in $(find build -name "*.dylib" 2>/dev/null); do
123+
echo "Signing dylib $lib ..."
124+
codesign --force --options runtime --timestamp \
125+
--sign "$CODESIGN_IDENTITY" "$lib"
126+
done
127+
128+
- name: Prepare release archive
129+
run: |
130+
mkdir -p release/build/bin
131+
cp build/bin/llama-server release/build/bin/
132+
cp build/bin/llama-cli release/build/bin/ 2>/dev/null || true
133+
cp build/bin/llama-bench release/build/bin/ 2>/dev/null || true
134+
cp build/bin/llama-perplexity release/build/bin/ 2>/dev/null || true
135+
find build -name "*.dylib" -exec cp {} release/build/bin/ \; 2>/dev/null || true
136+
find build -name "*.metal" -path "*/bin/*" -exec cp {} release/build/bin/ \; 2>/dev/null || true
137+
cd release
138+
zip -r ../llama-turboquant-macos-arm64.zip .
139+
tar -czf ../llama-turboquant-macos-arm64.tar.gz .
140+
cd ..
141+
ls -lh llama-turboquant-macos-arm64.zip llama-turboquant-macos-arm64.tar.gz
142+
143+
- name: Notarize release archive
144+
env:
145+
APPLE_ID: ${{ secrets.APPLE_ID }}
146+
APPLE_ID_PASSWORD: ${{ secrets.APPLE_ID_PASSWORD }}
147+
APPLE_TEAM_ID: ${{ secrets.APPLE_TEAM_ID }}
148+
run: |
149+
echo "=== Notarizing zip archive ==="
150+
xcrun notarytool submit llama-turboquant-macos-arm64.zip \
151+
--apple-id "$APPLE_ID" \
152+
--password "$APPLE_ID_PASSWORD" \
153+
--team-id "$APPLE_TEAM_ID" \
154+
--wait --timeout 10m
155+
echo "=== Notarizing individual binaries (for tar.gz users) ==="
156+
for bin in build/bin/llama-server build/bin/llama-cli build/bin/llama-bench build/bin/llama-perplexity; do
157+
if [ -f "$bin" ]; then
158+
name=$(basename "$bin")
159+
echo "--- Notarizing $name ---"
160+
zip -j "${name}.zip" "$bin"
161+
xcrun notarytool submit "${name}.zip" \
162+
--apple-id "$APPLE_ID" \
163+
--password "$APPLE_ID_PASSWORD" \
164+
--team-id "$APPLE_TEAM_ID" \
165+
--wait --timeout 10m
166+
rm "${name}.zip"
167+
fi
168+
done
169+
echo "=== All notarization complete ==="
170+
171+
- name: Clean up keychain
172+
if: always()
173+
run: |
174+
security delete-keychain $KEYCHAIN_PATH 2>/dev/null || true
175+
176+
- name: Upload artifact
177+
uses: actions/upload-artifact@v4
178+
with:
179+
name: llama-turboquant-macos-arm64
180+
path: release/
181+
retention-days: 30
182+
183+
- name: Create GitHub Release
184+
uses: softprops/action-gh-release@v2
185+
with:
186+
tag_name: ${{ steps.version.outputs.tag }}
187+
target_commitish: ${{ github.sha }}
188+
name: "TurboQuant macOS ARM64 (${{ steps.version.outputs.short_sha }})"
189+
body: |
190+
## TurboQuant KV Cache — macOS ARM64 (Metal)
191+
192+
Built from `feature/turboquant-kv-cache` branch at commit `${{ steps.version.outputs.short_sha }}`.
193+
194+
### What's included
195+
- `llama-server` with `--cache-type-k turbo3` / `turbo4` support
196+
- `llama-cli`, `llama-bench`, `llama-perplexity`
197+
- Metal backend with BF16 + embedded shader library
198+
199+
### Usage
200+
```bash
201+
# Option 1: zip (notarized + stapled)
202+
unzip llama-turboquant-macos-arm64.zip
203+
# Option 2: tar.gz
204+
tar -xzf llama-turboquant-macos-arm64.tar.gz
205+
206+
./build/bin/llama-server -m model.gguf --cache-type-k turbo3 --cache-type-v turbo3
207+
```
208+
209+
### For Atomic Chat integration
210+
Replace the binary at:
211+
```
212+
~/Library/Application Support/Atomic Chat/data/llamacpp/backends/<version>/macos-arm64/build/bin/llama-server
213+
```
214+
files: |
215+
llama-turboquant-macos-arm64.zip
216+
llama-turboquant-macos-arm64.tar.gz
217+
draft: false
218+
prerelease: true

0 commit comments

Comments
 (0)