Skip to content

Commit 5ce810e

Browse files
SamMalayekSamMalayek
committed
Update test and workflow to match new RFC
1 parent 2de1e68 commit 5ce810e

File tree

2 files changed

+329
-111
lines changed

2 files changed

+329
-111
lines changed

‎.github/workflows/embedding.yml‎

Lines changed: 174 additions & 19 deletions
Original file line numberDiff line numberDiff line change
@@ -25,41 +25,196 @@ on:
2525
- 'tests/e2e/embedding/**'
2626

2727
jobs:
28-
embedding-cli-tests:
28+
embedding-cli-tests-linux:
2929
runs-on: ubuntu-latest
30+
env:
31+
LLAMA_CACHE: tmp # stable path for cache
32+
EMBD_TEST_DEBUG: "1"
3033

3134
steps:
35+
- uses: actions/checkout@v4
36+
with: { fetch-depth: 0 }
37+
38+
- name: Restore model cache
39+
uses: actions/cache@v4
40+
with:
41+
path: |
42+
~/.cache/llama.cpp
43+
tmp
44+
key: hf-${{ runner.os }}-embeddinggemma-300M-q4_0-v1
45+
restore-keys: |
46+
hf-${{ runner.os }}-
47+
hf-
48+
3249
- name: Install system deps
3350
run: |
3451
sudo apt-get update
3552
sudo apt-get -y install \
36-
build-essential \
37-
cmake \
38-
curl \
39-
libcurl4-openssl-dev \
40-
python3-pip
41-
42-
- name: Checkout repository
43-
uses: actions/checkout@v4
44-
with:
45-
fetch-depth: 0
53+
build-essential cmake curl libcurl4-openssl-dev python3-pip
4654
4755
- name: Set up Python
4856
uses: actions/setup-python@v5
49-
with:
50-
python-version: '3.11'
57+
with: { python-version: '3.11' }
5158

5259
- name: Install Python deps
5360
run: |
54-
pip install -r requirements.txt || echo "No extra requirements found"
55-
pip install pytest
61+
python -m pip install -r requirements.txt || echo "No extra requirements found"
62+
python -m pip install pytest numpy pytest-timeout
5663
5764
- name: Build llama-embedding
5865
run: |
59-
cmake -B build \
60-
-DCMAKE_BUILD_TYPE=Release
66+
cmake -B build -DCMAKE_BUILD_TYPE=Release
6167
cmake --build build --target llama-embedding -j $(nproc)
6268
63-
- name: Run embedding tests
69+
- name: Pre-download tiny model (retry x3 on network)
70+
run: |
71+
set -e
72+
tries=0
73+
until ./build/bin/llama-embedding \
74+
-hfr ggml-org/embeddinggemma-300M-qat-q4_0-GGUF \
75+
-hff embeddinggemma-300M-qat-Q4_0.gguf \
76+
--ctx-size 16 --embd-output-format json --no-warmup --threads 1 --seed 42 <<< "ok"; do
77+
tries=$((tries+1))
78+
if [ $tries -ge 3 ]; then
79+
echo "Pre-download failed after $tries attempts"
80+
exit 1
81+
fi
82+
echo "Retrying download ($tries/3)..."
83+
sleep 3
84+
done
85+
86+
- name: Run embedding tests (30s per-test cap)
87+
shell: bash
88+
run: |
89+
set -o pipefail
90+
pytest -v tests/e2e/embedding \
91+
--timeout=30 \
92+
--durations=10 \
93+
--junitxml=pytest-report.xml | tee pytest-output.txt
94+
95+
- name: Upload test artifacts
96+
if: always()
97+
uses: actions/upload-artifact@v4
98+
with:
99+
name: linux-embedding-tests
100+
path: |
101+
pytest-output.txt
102+
pytest-report.xml
103+
104+
- name: Save model cache
105+
if: always()
106+
uses: actions/cache@v4
107+
with:
108+
path: |
109+
~/.cache/llama.cpp
110+
tmp
111+
key: hf-${{ runner.os }}-embeddinggemma-300M-q4_0-v1
112+
113+
embedding-cli-tests-windows:
114+
runs-on: windows-latest
115+
continue-on-error: true
116+
env:
117+
LLAMA_CACHE: tmp
118+
EMBD_TEST_DEBUG: "1"
119+
120+
steps:
121+
- uses: actions/checkout@v4
122+
- uses: actions/setup-python@v5
123+
with: { python-version: '3.11' }
124+
125+
# --- vcpkg plain bootstrap (no actions, no submodules) ---
126+
- name: Bootstrap vcpkg
127+
shell: pwsh
128+
run: |
129+
$env:VCPKG_ROOT = "$env:RUNNER_TEMP\vcpkg"
130+
git clone https://github.com/microsoft/vcpkg $env:VCPKG_ROOT
131+
& "$env:VCPKG_ROOT\bootstrap-vcpkg.bat" -disableMetrics
132+
echo "VCPKG_ROOT=$env:VCPKG_ROOT" | Out-File -FilePath $env:GITHUB_ENV -Append
133+
134+
- name: Install curl with OpenSSL via vcpkg
135+
shell: pwsh
136+
run: |
137+
& "$env:VCPKG_ROOT\vcpkg.exe" install curl[openssl]:x64-windows
138+
139+
- name: Restore model cache
140+
uses: actions/cache@v4
141+
with:
142+
path: |
143+
$HOME/.cache/llama.cpp
144+
tmp
145+
key: hf-${{ runner.os }}-embeddinggemma-300M-q4_0-v1
146+
restore-keys: |
147+
hf-${{ runner.os }}-
148+
hf-
149+
150+
- name: Install Python deps
151+
run: pip install pytest numpy
152+
153+
- name: Configure & Build (Release)
154+
shell: pwsh
155+
run: |
156+
cmake -B build -DCMAKE_BUILD_TYPE=Release `
157+
-DCMAKE_TOOLCHAIN_FILE="$env:VCPKG_ROOT\scripts\buildsystems\vcpkg.cmake"
158+
cmake --build build --target llama-embedding --config Release -j 2
159+
160+
- name: Pre-download tiny model (retry x3)
161+
shell: bash
162+
run: |
163+
set -e
164+
tries=0
165+
until ./build/bin/Release/llama-embedding.exe \
166+
-hfr ggml-org/embeddinggemma-300M-qat-q4_0-GGUF \
167+
-hff embeddinggemma-300M-qat-Q4_0.gguf \
168+
--ctx-size 16 --embd-output-format json --no-warmup --threads 1 --seed 42 <<< "ok"; do
169+
tries=$((tries+1))
170+
if [ $tries -ge 3 ]; then
171+
echo "Pre-download failed after $tries attempts"; exit 1
172+
fi
173+
echo "Retrying download ($tries/3)..."; sleep 3
174+
done
175+
176+
- name: Run smoke tests
177+
shell: bash
178+
run: |
179+
pytest -q tests/e2e/embedding -k raw_vs_json_consistency
180+
181+
182+
183+
embedding-cli-tests-macos:
184+
runs-on: macos-latest
185+
continue-on-error: true
186+
env:
187+
LLAMA_CACHE: tmp
188+
EMBD_TEST_DEBUG: "1"
189+
steps:
190+
- uses: actions/checkout@v4
191+
- uses: actions/setup-python@v5
192+
with: { python-version: '3.11' }
193+
194+
- name: Install Python deps
195+
run: pip install pytest numpy
196+
197+
- name: Build
198+
run: |
199+
cmake -B build -DCMAKE_BUILD_TYPE=Release
200+
cmake --build build --target llama-embedding -j 3
201+
202+
- name: Pre-download tiny model (retry x3)
203+
run: |
204+
set -e
205+
tries=0
206+
until ./build/bin/llama-embedding \
207+
-hfr ggml-org/embeddinggemma-300M-qat-q4_0-GGUF \
208+
-hff embeddinggemma-300M-qat-Q4_0.gguf \
209+
--ctx-size 16 --embd-output-format json --no-warmup --threads 1 --seed 42 <<< "ok"; do
210+
tries=$((tries+1))
211+
if [ $tries -ge 3 ]; then
212+
echo "Pre-download failed after $tries attempts"; exit 1
213+
fi
214+
echo "Retrying download ($tries/3)..."; sleep 3
215+
done
216+
217+
- name: Warm cache & run a tiny smoke
64218
run: |
65-
pytest -v tests/e2e/embedding
219+
./build/bin/llama-embedding --help >/dev/null 2>&1
220+
pytest -q tests/e2e/embedding -k raw_vs_json_consistency

0 commit comments

Comments
 (0)