diff --git a/.ci/pytorch/macos-test.sh b/.ci/pytorch/macos-test.sh
index 57549a7d63e1..f1dc9ba5a062 100755
--- a/.ci/pytorch/macos-test.sh
+++ b/.ci/pytorch/macos-test.sh
@@ -240,16 +240,6 @@ test_torchbench_smoketest() {
         dtype_arg="--float32"
     fi
     touch "$TEST_REPORTS_DIR/inductor_${backend}_torchbench_${dtype}_inference_${device}_performance.csv"
-    for model in "${models[@]}"; do
-      PYTHONPATH="$(pwd)"/torchbench python benchmarks/dynamo/torchbench.py \
-        --performance --only "$model" --backend "$backend" --inference --devices "$device" "$dtype_arg" \
-        --output "$TEST_REPORTS_DIR/inductor_${backend}_torchbench_${dtype}_inference_${device}_performance.csv" || true
-      if [ "$backend" == "inductor" ]; then
-        PYTHONPATH="$(pwd)"/torchbench python benchmarks/dynamo/torchbench.py \
-          --accuracy --only "$model" --backend "$backend" --inference --devices "$device" "$dtype_arg" \
-          --output "$TEST_REPORTS_DIR/inductor_${backend}_torchbench_${dtype}_inference_${device}_accuracy.csv" || true
-      fi
-    done
     if [ "$backend" == "inductor" ]; then
       PYTHONPATH="$(pwd)"/torchbench python benchmarks/dynamo/huggingface.py \
         --performance --backend "$backend" --inference --devices "$device" "$dtype_arg" \
@@ -257,6 +247,18 @@ test_torchbench_smoketest() {
       PYTHONPATH="$(pwd)"/torchbench python benchmarks/dynamo/huggingface.py \
         --accuracy --backend "$backend" --inference --devices "$device" "$dtype_arg" \
         --output "$TEST_REPORTS_DIR/inductor_${backend}_huggingface_${dtype}_inference_${device}_accuracy.csv" || true
+      PYTHONPATH="$(pwd)"/torchbench python benchmarks/dynamo/torchbench.py \
+        --performance --backend "$backend" --inference --devices "$device" "$dtype_arg" \
+        --output "$TEST_REPORTS_DIR/inductor_${backend}_torchbench_${dtype}_inference_${device}_performance.csv" || true
+      PYTHONPATH="$(pwd)"/torchbench python benchmarks/dynamo/torchbench.py \
+        --accuracy --backend "$backend" --inference --devices "$device" "$dtype_arg" \
+        --output "$TEST_REPORTS_DIR/inductor_${backend}_torchbench_${dtype}_inference_${device}_accuracy.csv" || true
+    else
+      for model in "${models[@]}"; do
+        PYTHONPATH="$(pwd)"/torchbench python benchmarks/dynamo/torchbench.py \
+          --performance --only "$model" --backend "$backend" --inference --devices "$device" "$dtype_arg" \
+          --output "$TEST_REPORTS_DIR/inductor_${backend}_torchbench_${dtype}_inference_${device}_performance.csv" || true
+      done
     fi
 
     if [ "$dtype" == notset ]; then
diff --git a/.github/workflows/inductor-perf-test-nightly-macos.yml b/.github/workflows/inductor-perf-test-nightly-macos.yml
index 0d92455a8f3c..2571c296c878 100644
--- a/.github/workflows/inductor-perf-test-nightly-macos.yml
+++ b/.github/workflows/inductor-perf-test-nightly-macos.yml
@@ -63,5 +63,6 @@ jobs:
       disable-monitor: false
       monitor-log-interval: 15
       monitor-data-collect-interval: 4
+      timeout: 360
 
     secrets: inherit