Skip to content

Commit a7246da

Browse files
q10facebook-github-bot
q10
authored and
facebook-github-bot
committed
Disable GenAI builds against CUDA 11.8 (#4173)
Summary: X-link: facebookresearch/FBGEMM#1255 - Disable GenAI builds against CUDA 11.8 since it is no longr possible to support GenAI builds against CUDA 11.8.0 as of #4138 Pull Request resolved: #4173 Reviewed By: jiawenliu64 Differential Revision: D75229752 Pulled By: q10 fbshipit-source-id: e9626799d371ee2671f9062df1933d3caea65087
1 parent daf6b8b commit a7246da

File tree

1 file changed

+23
-15
lines changed

1 file changed

+23
-15
lines changed

‎.github/workflows/fbgemm_gpu_ci_cuda.yml

Lines changed: 23 additions & 15 deletions
Original file line numberDiff line numberDiff line change
@@ -64,18 +64,21 @@ jobs:
6464
BUILD_ENV: build_binary
6565
BUILD_TARGET: ${{ matrix.host-machine.build-target }}
6666
BUILD_VARIANT: cuda
67-
BUILD_CUDA_VERSION: ${{ matrix.cuda-version }}
67+
BUILD_CUDA_VERSION: ${{ matrix.host-machine.cuda-version }}
6868
continue-on-error: true
6969
strategy:
7070
# Don't fast-fail all the other builds if one of the them fails
7171
fail-fast: false
7272
matrix:
7373
host-machine: [
74-
{ arch: x86, instance: "linux.24xlarge", build-target: "default" },
75-
{ arch: x86, instance: "linux.8xlarge.memory", build-target: "genai" },
74+
{ arch: x86, instance: "linux.24xlarge", build-target: "default", cuda-version: "11.8.0" },
75+
{ arch: x86, instance: "linux.24xlarge", build-target: "default", cuda-version: "12.6.3" },
76+
{ arch: x86, instance: "linux.24xlarge", build-target: "default", cuda-version: "12.8.0" },
77+
# GenAI is unable to support 11.8.0 anymore as of https://github.com/pytorch/FBGEMM/pull/4138
78+
{ arch: x86, instance: "linux.8xlarge.memory", build-target: "genai", cuda-version: "12.6.3" },
79+
{ arch: x86, instance: "linux.8xlarge.memory", build-target: "genai", cuda-version: "12.8.0" },
7680
]
7781
python-version: [ "3.9", "3.10", "3.11", "3.12", "3.13" ]
78-
cuda-version: [ "11.8.0", "12.6.3", "12.8.0" ]
7982
compiler: [ "gcc", "clang" ]
8083

8184
steps:
@@ -106,18 +109,18 @@ jobs:
106109
run: . $PRELUDE; install_build_tools $BUILD_ENV
107110

108111
- name: Install CUDA
109-
run: . $PRELUDE; install_cuda $BUILD_ENV ${{ matrix.cuda-version }}
112+
run: . $PRELUDE; install_cuda $BUILD_ENV ${{ matrix.host-machine.cuda-version }}
110113

111114
# Install via PIP to avoid defaulting to the CPU variant if the GPU variant of the day is not ready
112115
- name: Install PyTorch Nightly
113-
run: . $PRELUDE; install_pytorch_pip $BUILD_ENV ${{ (github.event_name == 'workflow_dispatch' && github.event.inputs.pytorch_channel_version) || 'nightly' }} cuda/${{ matrix.cuda-version }}
116+
run: . $PRELUDE; install_pytorch_pip $BUILD_ENV ${{ (github.event_name == 'workflow_dispatch' && github.event.inputs.pytorch_channel_version) || 'nightly' }} cuda/${{ matrix.host-machine.cuda-version }}
114117

115118
- name: Collect PyTorch Environment Info
116119
if: ${{ success() || failure() }}
117120
run: if . $PRELUDE && which conda; then collect_pytorch_env_info $BUILD_ENV; fi
118121

119122
- name: Install cuDNN
120-
run: . $PRELUDE; install_cudnn $BUILD_ENV "$(pwd)/build_only/cudnn" ${{ matrix.cuda-version }}
123+
run: . $PRELUDE; install_cudnn $BUILD_ENV "$(pwd)/build_only/cudnn" ${{ matrix.host-machine.cuda-version }}
121124

122125
- name: Prepare FBGEMM_GPU Build
123126
run: . $PRELUDE; cd fbgemm_gpu; prepare_fbgemm_gpu_build $BUILD_ENV
@@ -128,7 +131,7 @@ jobs:
128131
- name: Upload Built Wheel as GHA Artifact
129132
uses: actions/upload-artifact@v4
130133
with:
131-
name: fbgemm_${{ matrix.host-machine.build-target }}_${{ matrix.host-machine.arch }}_${{ matrix.compiler }}_py${{ matrix.python-version }}_cu${{ matrix.cuda-version }}.whl
134+
name: fbgemm_${{ matrix.host-machine.build-target }}_${{ matrix.host-machine.arch }}_${{ matrix.compiler }}_py${{ matrix.python-version }}_cu${{ matrix.host-machine.cuda-version }}.whl
132135
path: fbgemm_gpu/dist/*.whl
133136
if-no-files-found: error
134137

@@ -147,7 +150,7 @@ jobs:
147150
BUILD_ENV: build_binary
148151
BUILD_TARGET: ${{ matrix.build-target }}
149152
BUILD_VARIANT: cuda
150-
BUILD_CUDA_VERSION: ${{ matrix.cuda-version }}
153+
BUILD_CUDA_VERSION: ${{ matrix.build.cuda-version }}
151154
ENFORCE_CUDA_DEVICE: 1
152155
strategy:
153156
fail-fast: false
@@ -158,9 +161,14 @@ jobs:
158161
# https://hud.pytorch.org/metrics
159162
# { arch: x86, instance: "linux.gcp.a100" },
160163
]
161-
build-target: [ "default", "genai" ]
164+
build: [
165+
{ build-target: "default", cuda-version: "11.8.0" },
166+
{ build-target: "default", cuda-version: "12.6.3" },
167+
{ build-target: "default", cuda-version: "12.8.0" },
168+
{ build-target: "genai", cuda-version: "12.6.3" },
169+
{ build-target: "genai", cuda-version: "12.8.0" },
170+
]
162171
python-version: [ "3.9", "3.10", "3.11", "3.12", "3.13" ]
163-
cuda-version: [ "11.8.0", "12.6.3", "12.8.0" ]
164172
# Specify exactly ONE CUDA version for artifact publish
165173
cuda-version-publish: [ "12.6.3" ]
166174
compiler: [ "gcc", "clang" ]
@@ -177,7 +185,7 @@ jobs:
177185
# Cannot upgrade to actions/download-artifact@v4 yet because GLIBC on the instance is too old
178186
uses: actions/download-artifact@v4
179187
with:
180-
name: fbgemm_${{ matrix.build-target }}_${{ matrix.host-machine.arch }}_${{ matrix.compiler }}_py${{ matrix.python-version }}_cu${{ matrix.cuda-version }}.whl
188+
name: fbgemm_${{ matrix.build.build-target }}_${{ matrix.host-machine.arch }}_${{ matrix.compiler }}_py${{ matrix.python-version }}_cu${{ matrix.build.cuda-version }}.whl
181189

182190
# Use PyTorch test infrastructure action - https://github.com/pytorch/test-infra/blob/main/.github/actions/setup-nvidia/action.yml
183191
- name: Install NVIDIA Drivers and NVIDIA-Docker Runtime
@@ -203,11 +211,11 @@ jobs:
203211
run: . $PRELUDE; install_cxx_compiler $BUILD_ENV gcc
204212

205213
- name: Install CUDA
206-
run: . $PRELUDE; install_cuda $BUILD_ENV ${{ matrix.cuda-version }}
214+
run: . $PRELUDE; install_cuda $BUILD_ENV ${{ matrix.build.cuda-version }}
207215

208216
# Install via PIP to avoid defaulting to the CPU variant if the GPU variant of the day is not ready
209217
- name: Install PyTorch Nightly
210-
run: . $PRELUDE; install_pytorch_pip $BUILD_ENV ${{ (github.event_name == 'workflow_dispatch' && github.event.inputs.pytorch_channel_version) || 'nightly' }} cuda/${{ matrix.cuda-version }}
218+
run: . $PRELUDE; install_pytorch_pip $BUILD_ENV ${{ (github.event_name == 'workflow_dispatch' && github.event.inputs.pytorch_channel_version) || 'nightly' }} cuda/${{ matrix.build.cuda-version }}
211219

212220
- name: Collect PyTorch Environment Info
213221
if: ${{ success() || failure() }}
@@ -224,7 +232,7 @@ jobs:
224232
run: . $PRELUDE; test_all_fbgemm_gpu_modules $BUILD_ENV
225233

226234
- name: Push Wheel to PyPI
227-
if: ${{ (github.event_name == 'schedule' && matrix.cuda-version == matrix.cuda-version-publish) || (github.event_name == 'workflow_dispatch' && github.event.inputs.publish_to_pypi == 'true' && matrix.cuda-version == matrix.cuda-version-publish) }}
235+
if: ${{ (github.event_name == 'schedule' && matrix.build.cuda-version == matrix.cuda-version-publish) || (github.event_name == 'workflow_dispatch' && github.event.inputs.publish_to_pypi == 'true' && matrix.build.cuda-version == matrix.cuda-version-publish) }}
228236
env:
229237
PYPI_TOKEN: ${{ secrets.PYPI_TOKEN }}
230238
run: . $PRELUDE; publish_to_pypi $BUILD_ENV "$PYPI_TOKEN" *.whl

0 commit comments

Comments
 (0)