Add warning when cache_ram works with classify (#14650)

Co-authored-by: UltralyticsAssistant <web@ultralytics.com>
Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>

ultralytics/data/dataset.py

@@ -431,6 +431,12 @@ class ClassificationDataset:
             self.samples = self.samples[: round(len(self.samples) * args.fraction)]
         self.prefix = colorstr(f"{prefix}: ") if prefix else ""
         self.cache_ram = args.cache is True or str(args.cache).lower() == "ram"  # cache images into RAM
+        if self.cache_ram:
+            LOGGER.warning(
+                "WARNING ⚠️ Classification `cache_ram` training has known memory leak in "
+                "https://github.com/ultralytics/ultralytics/issues/9824, setting `cache_ram=False`."
+            )
+            self.cache_ram = False
         self.cache_disk = str(args.cache).lower() == "disk"  # cache images on hard drive as uncompressed *.npy files
         self.samples = self.verify_images()  # filter out bad images
         self.samples = [list(x) + [Path(x[0]).with_suffix(".npy"), None] for x in self.samples]  # file, index, npy, im