Update benchmarks.yml

Author: juliagmt-google

.github/workflows/benchmarks.yml

@@ -72,34 +72,40 @@ jobs:
           repository: openxla/xla  # Replace with your fork if needed
           path: xla
 
+      # - name: Build XLA with GPU support
+      #   working-directory: xla
+      #   continue-on-error: true
+      #   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/...
+     
+      # - name: Run XLA tests with GPU
+      #   working-directory: xla
+      #   continue-on-error: true
+      #   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/...
+
+      # - name: Run Profile Analysis
+      #   working-directory: xla
+      #   continue-on-error: true
+      #   run: bazel analyze-profile profile.json.gz
+        
+      # - name: Get GPU spec
+      #   working-directory: xla
+      #   continue-on-error: true
+      #   run: nvidia-smi
       # - name: Wait For Connection
       #   uses: google-ml-infra/actions/ci_connection@main
       #   with:
       #     halt-dispatch-input: ${{ inputs.halt-for-connection }}
 
-      - name: Build XLA with GPU support
+      - name: Configure XLA
         working-directory: xla
-        continue-on-error: true
-        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/...
-     
-      - name: Run XLA tests with GPU
-        working-directory: xla
-        continue-on-error: true
-        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/...
+        run: ./configure.py --backend CUDA --nccl
 
-      - name: Run Profile Analysis
+      - name: Set TF_CPP_MAX_VLOG_LEVEL
         working-directory: xla
-        continue-on-error: true
-        run: bazel analyze-profile profile.json.gz
-        
-      - name: Get GPU spec
-        working-directory: xla
-        continue-on-error: true
-        run: nvidia-smi
+        run: echo "TF_CPP_MAX_VLOG_LEVEL=1" >> $GITHUB_ENV
 
       - name: Build hlo_runner_main
         working-directory: xla
-        continue-on-error: true
         run: bazel build -c opt --config=cuda --dynamic_mode=off //xla/tools/multihost_hlo_runner:hlo_runner_main
 
       - name: Run XLA GPU microbenchmarks with hlo_runner_main