Skip to content

Commit 33ce83c

Browse files
juliagmt-googlejuliagmt-google
authored
Update benchmarks.yml
1 parent f6bb4d4 commit 33ce83c

File tree

1 file changed

+53
-23
lines changed

1 file changed

+53
-23
lines changed

‎.github/workflows/benchmarks.yml‎

Lines changed: 53 additions & 23 deletions
Original file line numberDiff line numberDiff line change
@@ -71,38 +71,68 @@ jobs:
7171
with:
7272
repository: openxla/xla # Replace with your fork if needed
7373
path: xla
74-
75-
# - name: Build XLA with GPU support
76-
# working-directory: xla
77-
# continue-on-error: true
78-
# run: bazel build --build_tag_filters=-no_oss,requires-gpu-nvidia,gpu,-rocm-only --test_tag_filters=-no_oss,requires-gpu-nvidia,gpu,-rocm-only,requires-gpu-sm75-only,requires-gpu-sm60,requires-gpu-sm70,-requires-gpu-sm80,-requires-gpu-sm80-only,-requires-gpu-sm90,-requires-gpu-sm90-only,-requires-gpu-sm100,-requires-gpu-sm100-only,-requires-gpu-amd --config=warnings --config=rbe_linux_cuda_nvcc --run_under=//tools/ci_build/gpu_build:parallel_gpu_execute --repo_env=TF_CUDA_COMPUTE_CAPABILITIES=7.5 --@cuda_driver//:enable_forward_compatibility=true --test_output=errors --verbose_failures --keep_going --nobuild_tests_only --profile=profile.json.gz --flaky_test_attempts=3 --jobs=150 --bes_upload_mode=fully_async --nobuild -- //xla/... //build_tools/... @tsl//tsl/...
79-
80-
# - name: Run XLA tests with GPU
81-
# working-directory: xla
82-
# continue-on-error: true
83-
# run: bazel test --build_tag_filters=-no_oss,requires-gpu-nvidia,gpu,-rocm-only --test_tag_filters=-no_oss,requires-gpu-nvidia,gpu,-rocm-only,requires-gpu-sm75-only,requires-gpu-sm60,requires-gpu-sm70,-requires-gpu-sm80,-requires-gpu-sm80-only,-requires-gpu-sm90,-requires-gpu-sm90-only,-requires-gpu-sm100,-requires-gpu-sm100-only,-requires-gpu-amd --config=warnings --config=rbe_linux_cuda_nvcc --run_under=//tools/ci_build/gpu_build:parallel_gpu_execute --repo_env=TF_CUDA_COMPUTE_CAPABILITIES=7.5 --@cuda_driver//:enable_forward_compatibility=true --test_output=errors --verbose_failures --keep_going --nobuild_tests_only --profile=profile.json.gz --flaky_test_attempts=3 --jobs=150 --bes_upload_mode=fully_async -- //xla/... //build_tools/... @tsl//tsl/...
74+
75+
- name: Configure XLA
76+
working-directory: xla
77+
run: ./configure.py --backend CUDA --nccl
8478

85-
# - name: Run Profile Analysis
86-
# working-directory: xla
87-
# continue-on-error: true
88-
# run: bazel analyze-profile profile.json.gz
79+
- name: Set TF_CPP_MAX_VLOG_LEVEL
80+
working-directory: xla
81+
run: echo "TF_CPP_MAX_VLOG_LEVEL=1" >> $GITHUB_ENV
8982

90-
# - name: Get GPU spec
91-
# working-directory: xla
92-
# continue-on-error: true
93-
# run: nvidia-smi
9483
- name: Wait For Connection
9584
uses: google-ml-infra/actions/ci_connection@main
9685
with:
9786
halt-dispatch-input: ${{ inputs.halt-for-connection }}
98-
99-
- name: Configure XLA
87+
88+
- name: Build XLA with GPU support with RBE
10089
working-directory: xla
101-
run: ./configure.py --backend CUDA --nccl
90+
continue-on-error: true
91+
run: bazel build --build_tag_filters=-no_oss,requires-gpu-nvidia,gpu,-rocm-only --test_tag_filters=-no_oss,requires-gpu-nvidia,gpu,-rocm-only,requires-gpu-sm75-only,requires-gpu-sm60,requires-gpu-sm70,-requires-gpu-sm80,-requires-gpu-sm80-only,-requires-gpu-sm90,-requires-gpu-sm90-only,-requires-gpu-sm100,-requires-gpu-sm100-only,-requires-gpu-amd --config=warnings --config=rbe_linux_cuda_nvcc --run_under=//tools/ci_build/gpu_build:parallel_gpu_execute --repo_env=TF_CUDA_COMPUTE_CAPABILITIES=7.5 --@cuda_driver//:enable_forward_compatibility=true --test_output=errors --verbose_failures --keep_going --nobuild_tests_only --profile=profile.json.gz --flaky_test_attempts=3 --jobs=150 --bes_upload_mode=fully_async --nobuild -- //xla/... //build_tools/... @tsl//tsl/...
92+
93+
- name: Run XLA tests with GPU support with RBE
94+
working-directory: xla
95+
continue-on-error: true
96+
run: bazel test --build_tag_filters=-no_oss,requires-gpu-nvidia,gpu,-rocm-only --test_tag_filters=-no_oss,requires-gpu-nvidia,gpu,-rocm-only,requires-gpu-sm75-only,requires-gpu-sm60,requires-gpu-sm70,-requires-gpu-sm80,-requires-gpu-sm80-only,-requires-gpu-sm90,-requires-gpu-sm90-only,-requires-gpu-sm100,-requires-gpu-sm100-only,-requires-gpu-amd --config=warnings --config=rbe_linux_cuda_nvcc --run_under=//tools/ci_build/gpu_build:parallel_gpu_execute --repo_env=TF_CUDA_COMPUTE_CAPABILITIES=7.5 --@cuda_driver//:enable_forward_compatibility=true --test_output=errors --verbose_failures --keep_going --nobuild_tests_only --profile=profile.json.gz --flaky_test_attempts=3 --jobs=150 --bes_upload_mode=fully_async -- //xla/... //build_tools/... @tsl//tsl/...
10297

103-
- name: Set TF_CPP_MAX_VLOG_LEVEL
98+
- name: Run Profile Analysis
10499
working-directory: xla
105-
run: echo "TF_CPP_MAX_VLOG_LEVEL=1" >> $GITHUB_ENV
100+
continue-on-error: true
101+
run: bazel analyze-profile profile.json.gz
102+
103+
- name: Get GPU spec
104+
working-directory: xla
105+
continue-on-error: true
106+
run: nvidia-smi
107+
108+
- name: Build run_hlo_module with GPU
109+
working-directory: openxla
110+
run: bazel build -c opt --dynamic_mode=off xla/tools:run_hlo_module --config=cuda
111+
112+
- name: Run HLO Module Benchmarks withg GPU in xla/tests/fuzz
113+
working-directory: openxla
114+
continue-on-error: true
115+
run: |
116+
for file in xla/tests/fuzz/*.hlo; do
117+
filename=$(basename "$file")
118+
# Skip expected failed hlo files.
119+
if [[ "$filename" == "rand_000060.hlo" || "$filename" == "rand_000067.hlo" || "$filename" == "rand_000072.hlo" ]]; then
120+
echo "Skipping benchmark on $file"
121+
continue
122+
fi
123+
echo "Running benchmark on $file"
124+
./bazel-bin/xla/tools/run_hlo_module --input_format=hlo --platform=GPU "$file"
125+
done
126+
127+
- name: Run HLO Module Benchmarks withg GPU in xla/service/gpu
128+
working-directory: openxla
129+
continue-on-error: true
130+
run: |
131+
find xla/service/gpu -name "*.hlo" -print0 | while IFS= read -r -d $'\0' hlo_file; do
132+
echo "Running: $hlo_file"
133+
./bazel-bin/xla/tools/run_hlo_module --input_format=hlo --platform=GPU "$hlo_file"
134+
echo "--------------------------------------------------"
135+
done
106136
107137
- name: Build hlo_runner_main
108138
working-directory: xla

0 commit comments

Comments
 (0)