diff --git a/run.py b/run.py
index 927f1472a..c41d90d83 100644
--- a/run.py
+++ b/run.py
@@ -477,15 +477,15 @@ def main() -> None:
     )
     parser.add_argument(
         "--metrics-gpu-backend",
-        choices=["dcgm", "default", "nvml"],
-        default="default",
+        choices=["torch", "nvml", "dcgm"],
+        default="torch",
         help="""
-        Specify the backend [dcgm, default, nvml] to collect metrics. In all modes,
+        Specify the backend [torch, nvml, dcgm] to collect metrics. In all modes,
         the latency (execution time) is always collected using `time.time_ns()`. The CPU
         and GPU peak memory usage metrics are optional. The CPU peak memory usage is
         collected by `psutil.Process()` in all modes. In nvml mode, the GPU peak memory
         usage is collected by the `nvml` library. In dcgm mode, the GPU peak memory usage is
-        collected by the `dcgm` library. In default mode, the GPU peak memory usage is collected
+        collected by the `dcgm` library. In torch mode, the GPU peak memory usage is collected
         by `torch.cuda.max_memory_allocated()`.
         """,
     )
diff --git a/torchbenchmark/util/experiment/metrics.py b/torchbenchmark/util/experiment/metrics.py
index 7e7684caf..fd206b22a 100644
--- a/torchbenchmark/util/experiment/metrics.py
+++ b/torchbenchmark/util/experiment/metrics.py
@@ -71,7 +71,7 @@ def get_peak_memory(
     num_iter=MEMPROF_ITER,
     export_metrics_file="",
     metrics_needed=[],
-    metrics_gpu_backend="default",
+    metrics_gpu_backend="torch",
     cpu_monitored_pid=None,
 ) -> Tuple[Optional[float], Optional[str], Optional[float]]:
     "Run one step of the model, and return the peak memory in MB."
diff --git a/userbenchmark/triton/run.py b/userbenchmark/triton/run.py
index 033d160b3..f79ebb685 100644
--- a/userbenchmark/triton/run.py
+++ b/userbenchmark/triton/run.py
@@ -94,13 +94,13 @@ def get_parser(args=None):
     )
     parser.add_argument(
         "--metrics-gpu-backend",
-        choices=["default", "nvml"],
-        default="default",
+        choices=["torch", "nvml"],
+        default="torch",
         help=(
-            "Specify the backend [default, nvml] to collect metrics. In all modes, the latency "
+            "Specify the backend [torch, nvml] to collect metrics. In all modes, the latency "
             "(execution time) is always collected using `time.time_ns()`. The CPU peak memory "
             "usage is collected by `psutil.Process()`. In nvml mode, the GPU peak memory usage "
-            "is collected by the `nvml` library. In default mode, the GPU peak memory usage is "
+            "is collected by the `nvml` library. In torch mode, the GPU peak memory usage is "
             "collected by `torch.cuda.max_memory_allocated()`."
         ),
     )