forked from pytorch/FBGEMM
-
Notifications
You must be signed in to change notification settings - Fork 0
231 lines (187 loc) · 8.02 KB
/
fbgemm_gpu_pip.yml
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
# This source code is licensed under the BSD-style license found in the
# LICENSE file in the root directory of this source tree.
# This workflow is used for testing the download and installation of FBGEMM_GPU
# nightly releases published to PyTorch PyPI.
name: FBGEMM_GPU PIP Install + Test
on:
# Cron Trigger (UTC)
#
# Based on the the nightly releases schedule in PyTorch infrastructure, the
# wheels are published to PyTorch PIP at around 11:30 UTC every day. After
# publication, it can take up to 30 minutes for the wheels to be published, as
# the re-indexing job is scheduled to run every 30 minutes. As such, we set
# the PIP install + test workflow to be kicked off 4 hours after the publish
# job is kicked off to give ample time for the nightly wheel to be available
# in PyTorch PIP.
#
schedule:
- cron: '30 15 * * *'
# Manual Trigger
#
workflow_dispatch:
inputs:
pytorch_version:
description: PyTorch Version (e.g. '2.1.0', 'nightly', 'test')
type: string
required: true
default: "nightly"
fbgemm_gpu_channel_version:
description: FBGEMM-GPU Channel + Version (e.g. '0.5.0', 'nightly', 'test/0.8.0r0')
type: string
required: true
default: "nightly"
fbgemm_gpu_variant_type:
description: FBGEMM-GPU Variant
type: choice
required: true
options: [ "cpu", "cuda", "rocm" ]
default: "cpu"
jobs:
test_pypi_install_cpu:
if: ${{ github.repository_owner == 'pytorch' && (github.event_name == 'schedule' || (github.event_name == 'workflow_dispatch' && github.event.inputs.fbgemm_gpu_variant_type == 'cpu')) }}
runs-on: ${{ matrix.host-machine.instance }}
container:
image: amazonlinux:2023
options: --user root
defaults:
run:
shell: bash
env:
PRELUDE: .github/scripts/setup_env.bash
BUILD_ENV: test_install
BUILD_VARIANT: cpu
strategy:
fail-fast: false
matrix:
host-machine: [
{ arch: x86, instance: "linux.4xlarge", timeout: 20 },
{ arch: arm, instance: "linux.arm64.2xlarge", timeout: 30 },
]
python-version: [ "3.9", "3.10", "3.11", "3.12" ]
steps:
- name: Setup Build Container
run: yum update -y; yum install -y binutils findutils git pciutils sudo wget which
- name: Checkout the Repository
uses: actions/checkout@v4
- name: Display System Info
run: . $PRELUDE; print_system_info; print_ec2_info
- name: Display GPU Info
run: . $PRELUDE; print_gpu_info
- name: Setup Miniconda
run: . $PRELUDE; setup_miniconda $HOME/miniconda
- name: Create Conda Environment
run: . $PRELUDE; create_conda_environment $BUILD_ENV ${{ matrix.python-version }}
- name: Install Build Tools
run: . $PRELUDE; install_build_tools $BUILD_ENV
- name: Install PyTorch-CPU
run: . $PRELUDE; install_pytorch_pip $BUILD_ENV ${{ github.event.inputs.pytorch_version || 'nightly' }} cpu
- name: Collect PyTorch Environment Info
if: ${{ success() || failure() }}
run: if . $PRELUDE && which conda; then collect_pytorch_env_info $BUILD_ENV; fi
- name: Install FBGEMM_GPU-CPU
run: . $PRELUDE; install_fbgemm_gpu_pip $BUILD_ENV ${{ github.event.inputs.fbgemm_gpu_channel_version || 'nightly' }} cpu
- name: Test with PyTest
timeout-minutes: ${{ matrix.host-machine.timeout }}
run: . $PRELUDE; test_all_fbgemm_gpu_modules $BUILD_ENV
test_pypi_install_cuda:
if: ${{ github.repository_owner == 'pytorch' && (github.event_name == 'schedule' || (github.event_name == 'workflow_dispatch' && github.event.inputs.fbgemm_gpu_variant_type == 'cuda') }}
runs-on: ${{ matrix.host-machine.instance }}
defaults:
run:
shell: bash
env:
PRELUDE: .github/scripts/setup_env.bash
BUILD_ENV: test_install
BUILD_VARIANT: cuda
ENFORCE_CUDA_DEVICE: 1
strategy:
fail-fast: false
matrix:
host-machine: [
{ instance: "linux.g5.4xlarge.nvidia.gpu" },
]
python-version: [ "3.9", "3.10", "3.11", "3.12" ]
cuda-version: [ "11.8.0", "12.1.1", "12.4.1" ]
steps:
# Cannot upgrade to actions/checkout@v4 yet because GLIBC on the instance is too old
- name: Checkout the Repository
uses: actions/checkout@v3
- name: Install NVIDIA Drivers and NVIDIA-Docker Runtime
uses: pytorch/test-infra/.github/actions/setup-nvidia@main
- name: Display System Info
run: . $PRELUDE; print_system_info; print_ec2_info
- name: Display GPU Info
run: . $PRELUDE; print_gpu_info
- name: Setup Miniconda
run: . $PRELUDE; setup_miniconda $HOME/miniconda
- name: Create Conda Environment
run: . $PRELUDE; create_conda_environment $BUILD_ENV ${{ matrix.python-version }}
- name: Install Build Tools
run: . $PRELUDE; install_build_tools $BUILD_ENV
- name: Install CUDA
run: . $PRELUDE; install_cuda $BUILD_ENV ${{ matrix.cuda-version }}
- name: Install PyTorch-CUDA
run: . $PRELUDE; install_pytorch_pip $BUILD_ENV ${{ github.event.inputs.pytorch_version || 'nightly' }} cuda/${{ matrix.cuda-version }}
- name: Collect PyTorch Environment Info
if: ${{ success() || failure() }}
run: if . $PRELUDE && which conda; then collect_pytorch_env_info $BUILD_ENV; fi
- name: Install FBGEMM_GPU-CUDA
run: . $PRELUDE; install_fbgemm_gpu_pip $BUILD_ENV ${{ github.event.inputs.fbgemm_gpu_channel_version || 'nightly' }} cuda/${{ matrix.cuda-version }}
- name: Test with PyTest
timeout-minutes: 40
run: . $PRELUDE; test_all_fbgemm_gpu_modules $BUILD_ENV
test_pypi_install_rocm:
if: ${{ github.repository_owner == 'pytorch' && (github.event_name == 'schedule' || (github.event_name == 'workflow_dispatch' && github.event.inputs.fbgemm_gpu_variant_type == 'rocm') }}
runs-on: ${{ matrix.host-machine.instance }}
container:
image: "rocm/dev-ubuntu-20.04:${{ matrix.rocm-version }}-complete"
options: --user root --device=/dev/kfd --device=/dev/dri --ipc=host --shm-size 16G --group-add video --cap-add=SYS_PTRACE --security-opt seccomp=unconfined
defaults:
run:
shell: bash
env:
PRELUDE: .github/scripts/setup_env.bash
BUILD_ENV: test_install
BUILD_VARIANT: rocm
ENFORCE_ROCM_DEVICE: 1
strategy:
fail-fast: false
matrix:
host-machine: [
{ arch: x86, instance: "rocm" },
]
# ROCm machines are limited, so we only test a subset of Python versions
python-version: [ "3.11", "3.12" ]
rocm-version: [ "6.1" ]
steps:
- name: Setup Build Container
run: |
apt update -y
apt install -y git wget
git config --global --add safe.directory '*'
- name: Checkout the Repository
uses: actions/checkout@v4
- name: Display System Info
run: . $PRELUDE; print_system_info
- name: Display GPU Info
run: . $PRELUDE; print_gpu_info
- name: Free Disk Space
run: . $PRELUDE; free_disk_space
- name: Setup Miniconda
run: . $PRELUDE; setup_miniconda $HOME/miniconda
- name: Create Conda Environment
run: . $PRELUDE; create_conda_environment $BUILD_ENV ${{ matrix.python-version }}
- name: Install Build Tools
run: . $PRELUDE; install_build_tools $BUILD_ENV
- name: Install PyTorch-ROCm
run: . $PRELUDE; install_pytorch_pip $BUILD_ENV ${{ github.event.inputs.pytorch_version || 'nightly' }} rocm/${{ matrix.rocm-version }}
- name: Collect PyTorch Environment Info
if: ${{ success() || failure() }}
run: if . $PRELUDE && which conda; then collect_pytorch_env_info $BUILD_ENV; fi
- name: Install FBGEMM_GPU-ROCm
run: . $PRELUDE; install_fbgemm_gpu_pip $BUILD_ENV ${{ github.event.inputs.fbgemm_gpu_channel_version || 'nightly' }} rocm/${{ matrix.rocm-version }}
- name: Test with PyTest
timeout-minutes: 20
run: . $PRELUDE; test_all_fbgemm_gpu_modules $BUILD_ENV