deepspeed

1
name: nv-inference
2

3
on:
4
  workflow_dispatch:
5
  pull_request:
6
    paths:
7
      - '.github/workflows/nv-inference.yml'
8
      - 'requirements/**'
9
      - 'deepspeed/__init__.py'
10
      - 'deepspeed/inference/**'
11
      - '!deepspeed/inference/v2/**' # exclude v2 dir
12
      - 'tests/unit/inference/**'
13
      - '!tests/unit/inference/v2/**' # exclude v2 tests dir
14
  merge_group:
15
    branches: [ master ]
16
  schedule:
17
    - cron: "0 0 * * *"
18

19
concurrency:
20
  group: ${{ github.workflow }}-${{ github.ref }}
21
  cancel-in-progress: true
22

23
jobs:
24
  unit-tests:
25
    runs-on: [self-hosted, nvidia, cu116, v100]
26

27
    steps:
28
      - uses: actions/checkout@v3
29

30
      - id: setup-venv
31
        uses: ./.github/workflows/setup-venv
32

33
      - name: Install pytorch
34
        run: |
35
          pip install -U --cache-dir $TORCH_CACHE torch==2.1.2 torchvision==0.16.2 --index-url https://download.pytorch.org/whl/cu118
36
          python -c "import torch; print('torch:', torch.__version__, torch)"
37
          python -c "import torch; print('CUDA available:', torch.cuda.is_available())"
38

39
      - name: Install transformers
40
        run: |
41
          git clone https://github.com/huggingface/transformers
42
          cd transformers
43
          #git checkout f370bebdc
44
          git rev-parse --short HEAD
45
          pip install .
46

47
      - name: Install deepspeed
48
        run: |
49
          pip install .[dev,1bit,autotuning,inf,triton]
50
          ds_report
51

52
      - name: Python environment
53
        run: |
54
          pip list
55

56
      - name: Unit tests
57
        run: |
58
          unset TORCH_CUDA_ARCH_LIST # only jit compile for current arch
59
          cd tests
60
          #pytest $PYTEST_OPTS -m 'seq_inference' unit/ --torch_ver="2.1" --cuda_ver="11.8"
61
          pytest $PYTEST_OPTS -m 'inference_ops' unit/ --torch_ver="2.1" --cuda_ver="11.8"
62
          pytest $PYTEST_OPTS --forked -n 4 -m 'inference' unit/ --torch_ver="2.1" --cuda_ver="11.8"
63