Ensure GPU utilization and benchmark vectorized computations

.github/workflows/gpu_benchmark.yml

@@ -38,7 +38,7 @@ jobs:
 
       - name: Run benchmark and store result
         run: |
-            pytest tests/test_benchmark.py -k 'not test_rigid_contact_model and not test_soft_contact_model' --benchmark-only --benchmark-json output.json
+            pytest tests/test_benchmark.py -k 'not test_rigid_contact_model and not test_soft_contact_model' --gpu-only --batch-size 128 --benchmark-only --benchmark-json output.json
 
       - name: Download previous benchmark data
         uses: actions/cache@v4