added a check for autocast

classify/val.py

@@ -108,7 +108,8 @@ def run(
     action = "validating" if dataloader.dataset.root.stem == "val" else "testing"
     desc = f"{pbar.desc[:-36]}{action:>36}" if pbar else f"{action}"
     bar = tqdm(dataloader, desc, n, not training, bar_format=TQDM_BAR_FORMAT, position=0)
-    with torch.amp.autocast("cuda", enabled=device.type != "cpu"):
+    device_amp = torch.is_autocast_available(device_type=device.type)
+    with torch.amp.autocast(device_type=device.type, enabled=device_amp):
         for images, labels in bar:
             with dt[0]:
                 images, labels = images.to(device, non_blocking=True), labels.to(device)

train.py

@@ -352,7 +352,11 @@ def train(hyp, opt, device, callbacks):
     maps = np.zeros(nc)  # mAP per class
     results = (0, 0, 0, 0, 0, 0, 0)  # P, R, mAP@.5, mAP@.5-.95, val_loss(box, obj, cls)
     scheduler.last_epoch = start_epoch - 1  # do not move
-    scaler = amp.GradScaler(enabled=device.type != "cpu")
+    
+    #checking if autocast is available
+    device_amp = torch.is_autocast_available(device_type=device.type)
+
+    scaler = torch.amp.GradScaler(enabled=(device_amp and device.type != "cpu"))
     stopper, stop = EarlyStopping(patience=opt.patience), False
     compute_loss = ComputeLoss(model)  # init loss class
     callbacks.run("on_train_start")
@@ -409,7 +413,7 @@ def train(hyp, opt, device, callbacks):
                     imgs = nn.functional.interpolate(imgs, size=ns, mode="bilinear", align_corners=False)
 
             # Forward
-            with torch.amp.autocast("cuda", enabled=device.type != "cpu"):
+            with torch.amp.autocast(device_type=device.type, enabled=device_amp):
                 pred = model(imgs)  # forward
                 loss, loss_items = compute_loss(pred, targets.to(device))  # loss scaled by batch_size
                 if RANK != -1: