diff --git a/ultralytics/models/yolo/detect/val.py b/ultralytics/models/yolo/detect/val.py
index 0823f75144..aef60d65eb 100644
--- a/ultralytics/models/yolo/detect/val.py
+++ b/ultralytics/models/yolo/detect/val.py
@@ -75,7 +75,7 @@ class DetectionValidator(BaseValidator):
         )  # is COCO
         self.is_lvis = isinstance(val, str) and "lvis" in val and not self.is_coco  # is LVIS
         self.class_map = converter.coco80_to_coco91_class() if self.is_coco else list(range(len(model.names)))
-        self.args.save_json |= (self.is_coco or self.is_lvis) and not self.training  # run on final val if training COCO
+        self.args.save_json |= self.args.val and (self.is_coco or self.is_lvis) and not self.training  # run final val
         self.names = model.names
         self.nc = len(model.names)
         self.metrics.names = self.names
diff --git a/ultralytics/utils/autobatch.py b/ultralytics/utils/autobatch.py
index 86619998a1..581fe09027 100644
--- a/ultralytics/utils/autobatch.py
+++ b/ultralytics/utils/autobatch.py
@@ -1,6 +1,7 @@
 # Ultralytics YOLO 🚀, AGPL-3.0 license
 """Functions for estimating the best YOLO batch size to use a fraction of the available CUDA memory in PyTorch."""
 
+import os
 from copy import deepcopy
 
 import numpy as np
@@ -57,7 +58,7 @@ def autobatch(model, imgsz=640, fraction=0.60, batch_size=DEFAULT_CFG.batch):
 
     # Inspect CUDA memory
     gb = 1 << 30  # bytes to GiB (1024 ** 3)
-    d = str(device).upper()  # 'CUDA:0'
+    d = f"CUDA:{os.getenv('CUDA_VISIBLE_DEVICES', '0').strip()[0]}"  # 'CUDA:0'
     properties = torch.cuda.get_device_properties(device)  # device properties
     t = properties.total_memory / gb  # GiB total
     r = torch.cuda.memory_reserved(device) / gb  # GiB reserved