add flag to disable gpu capture (#54762)

python/paddle/distributed/launch/context/args_envs.py

@@ -89,6 +89,13 @@ def parse_args():
         help="rank node by ip. Default False",
     )
 
+    base_group.add_argument(
+        "--enable_gpu_log",
+        type=strtobool,
+        default=True,
+        help="enable capture gpu log while running. Default True",
+    )
+
     base_group.add_argument(
         "--nnodes",
         type=str,

python/paddle/distributed/launch/controllers/watcher.py

@@ -27,6 +27,9 @@ class Watcher:
 
         self.gpu_util = []
 
+        if not self.ctx.args.enable_gpu_log:
+            return
+
         # gpu log file
         self.gpus = self.ctx.args.devices or self.ctx.node.device.labels
         if len(self.gpus) > 0: