|
8 | 8 | pull_request: |
9 | 9 | workflow_dispatch: |
10 | 10 | jobs: |
11 | | - core: |
12 | | - name: 'Core' |
13 | | - runs-on: ubuntu-latest |
14 | | - container: |
15 | | - image: nvcr.io/nvidia/cuda:12.1.0-devel-ubuntu22.04 |
16 | | - options: --user root |
17 | | - steps: |
18 | | - - name: 'Dependencies' |
19 | | - run: | |
20 | | - apt-get update |
21 | | - apt-get install -y git python3.9 pip cudnn9-cuda-12 |
22 | | - pip install cmake==3.21.0 pybind11[global] ninja nvidia-mathdx==25.1.1 |
23 | | - - name: 'Checkout' |
24 | | - uses: actions/checkout@v3 |
25 | | - with: |
26 | | - submodules: recursive |
27 | | - - name: 'Build' |
28 | | - run: pip install --no-build-isolation . -v |
29 | | - env: |
30 | | - NVTE_FRAMEWORK: none |
31 | | - MAX_JOBS: 1 |
32 | | - - name: 'Sanity check' |
33 | | - run: python3 -c "import transformer_engine" |
34 | | - working-directory: / |
35 | 11 | pytorch: |
36 | 12 | name: 'PyTorch' |
37 | | - runs-on: ubuntu-latest |
| 13 | + runs-on: [ self-hosted, Linux, X64, nvidia, gpu-8 ] |
| 14 | + defaults: |
| 15 | + run: |
| 16 | + shell: bash |
38 | 17 | container: |
39 | | - image: nvcr.io/nvidia/cuda:12.8.0-devel-ubuntu22.04 |
| 18 | + image: harbor.baai.ac.cn/flagscale/cuda12.8.1-torch2.7.1-python3.10-te2.9:20260209 |
40 | 19 | options: --user root |
41 | 20 | steps: |
42 | | - - name: 'Dependencies' |
43 | | - run: | |
44 | | - apt-get update |
45 | | - apt-get install -y git python3.9 pip cudnn9-cuda-12 |
46 | | - pip install cmake torch ninja pydantic importlib-metadata>=1.0 packaging pybind11 numpy einops onnxscript nvidia-mathdx==25.1.1 |
47 | 21 | - name: 'Checkout' |
48 | 22 | uses: actions/checkout@v3 |
49 | 23 | with: |
50 | 24 | submodules: recursive |
51 | 25 | - name: 'Build' |
52 | | - run: pip install --no-build-isolation . -v --no-deps |
| 26 | + run: |
| 27 | + source /opt/miniconda3/etc/profile.d/conda.sh |
| 28 | + conda activate flagscale-train |
| 29 | + pip install --no-build-isolation . -v --no-deps |
53 | 30 | env: |
54 | 31 | NVTE_FRAMEWORK: pytorch |
55 | | - MAX_JOBS: 1 |
56 | | - - name: 'Sanity check' |
57 | | - run: python3 tests/pytorch/test_sanity_import.py |
58 | | - jax: |
59 | | - name: 'JAX' |
60 | | - runs-on: ubuntu-latest |
61 | | - container: |
62 | | - image: ghcr.io/nvidia/jax:jax |
63 | | - options: --user root |
64 | | - steps: |
65 | | - - name: 'Dependencies' |
66 | | - run: pip install pybind11[global] nvidia-mathdx==25.1.1 |
67 | | - - name: 'Checkout' |
68 | | - uses: actions/checkout@v3 |
69 | | - with: |
70 | | - submodules: recursive |
71 | | - - name: 'Build' |
72 | | - run: pip install --no-build-isolation . -v |
73 | | - env: |
74 | | - NVTE_FRAMEWORK: jax |
75 | | - MAX_JOBS: 1 |
76 | | - - name: 'Sanity check' |
77 | | - run: python3 tests/jax/test_sanity_import.py |
78 | | - all: |
79 | | - name: 'All' |
80 | | - runs-on: ubuntu-latest |
81 | | - container: |
82 | | - image: ghcr.io/nvidia/jax:jax |
83 | | - options: --user root |
84 | | - steps: |
85 | | - - name: 'Dependencies' |
86 | | - run: pip install torch pybind11[global] einops onnxscript nvidia-mathdx==25.1.1 |
87 | | - - name: 'Checkout' |
88 | | - uses: actions/checkout@v3 |
89 | | - with: |
90 | | - submodules: recursive |
91 | | - - name: 'Build' |
92 | | - run: pip install --no-build-isolation . -v --no-deps |
93 | | - env: |
94 | | - NVTE_FRAMEWORK: all |
95 | | - MAX_JOBS: 1 |
| 32 | + TE_WITH_NCCL: 1 |
96 | 33 | - name: 'Sanity check' |
97 | | - run: python3 tests/pytorch/test_sanity_import.py && python3 tests/jax/test_sanity_import.py |
| 34 | + run: |
| 35 | + source /opt/miniconda3/etc/profile.d/conda.sh |
| 36 | + conda activate flagscale-train |
| 37 | + python3 tests/pytorch/test_sanity_import.py |
0 commit comments