Update benchmarks.yml

Author: juliagmt-google

.github/workflows/benchmarks.yml

@@ -66,22 +66,6 @@ jobs:
       #   uses: google-ml-infra/actions/ci_connection@main
       #   with:
       #     halt-dispatch-input: ${{ inputs.halt-for-connection }}
-
-      - name: Create gpu_hlo_backend.hlo
-        working-directory: xla
-        run: |
-          cat << EOF > gpu_hlo_backend.hlo
-          HloModule module
-          // CHECK: is_scheduled=true
-          
-          ENTRY computation {
-              p = f32[5000,6000]{1,0} parameter(0)
-              e = f32[5000,6000]{1,0} sqrt(p)
-              c = f32[6000,5000] transpose(p), dimensions={1,0}
-              r = f32[300,20,5000] reshape(c)
-              ROOT out = (f32[5000,6000], f32[300,20,5000]) tuple(e,r)
-          }
-          EOF
 
       # - name: Wait For Connection
       #   uses: google-ml-infra/actions/ci_connection@main
@@ -92,7 +76,7 @@ jobs:
         working-directory: xla
         run: |
           cd ..
-          ./bazel-bin/xla/tools/multihost_hlo_runner/hlo_runner_main --device_type=gpu --log_output=True --use_spmd_partitioning --xla_gpu_dump_xspace_to=xla/results/xspace.pbtxt xla/gpu_hlo_backend.hlo  &> xla/results/gpu_hlo_backend.log
+          ./bazel-bin/xla/tools/multihost_hlo_runner/hlo_runner_main --device_type=gpu --log_output=True --use_spmd_partitioning --xla_gpu_dump_xspace_to=xla/results/xspace.pbtxt xla/tools/hlo_opt/tests/gpu_hlo_pass.hlo  &> xla/results/gpu_hlo_backend.log
     
       # - name: Wait For Connection
       #   uses: google-ml-infra/actions/ci_connection@main