Revert default in-memory for small datasets

docs/source/loading_datasets.rst

@@ -66,11 +66,11 @@ This call to :func:`datasets.load_dataset` does the following steps under the ho
     to map blobs of data on-drive without doing any deserialization. So caching the dataset directly on disk can use
     memory-mapping and pay effectively zero cost with O(1) random access. Alternatively, you can copy it in CPU memory
     (RAM) by setting the ``keep_in_memory`` argument of :func:`datasets.load_datasets` to ``True``.
-    The default in 🤗Datasets is to memory-map the dataset on drive if its size is larger than
-    ``datasets.config.IN_MEMORY_MAX_SIZE`` (default ``250 * 2 ** 20`` B); otherwise, the dataset is copied in-memory.
-    This behavior can be disabled (i.e., the dataset will not be loaded in memory) by setting to ``0`` either the
-    configuration option ``datasets.config.IN_MEMORY_MAX_SIZE`` (higher precedence) or the environment variable
-    ``HF_DATASETS_IN_MEMORY_MAX_SIZE`` (lower precedence).
+    The default in 🤗Datasets is to memory-map the dataset on disk unless you set ``datasets.config.IN_MEMORY_MAX_SIZE``
+    different from ``0`` bytes (default). In that case, the dataset will be copied in-memory if its size is smaller than
+    ``datasets.config.IN_MEMORY_MAX_SIZE`` bytes, and memory-mapped otherwise. This behavior can be enabled by setting
+    either the configuration option ``datasets.config.IN_MEMORY_MAX_SIZE`` (higher precedence) or the environment
+    variable ``HF_DATASETS_IN_MEMORY_MAX_SIZE`` (lower precedence) to nonzero.
 
 3. Return a **dataset built from the splits** asked by the user (default: all); in the above example we create a dataset with the train split.
 
@@ -440,3 +440,15 @@ Indeed, if you've already loaded the dataset once before (when you had an intern
 You can even set the environment variable `HF_DATASETS_OFFLINE` to ``1`` to tell ``datasets`` to run in full offline mode.
 This mode disables all the network calls of the library.
 This way, instead of waiting for a dataset builder download to time out, the library looks directly at the cache.
+
+.. _load_dataset_enhancing_performance:
+
+Enhancing performance
+-----------------------------------------------------------
+
+If you would like to speed up dataset operations, you can disable caching and copy the dataset in-memory by setting
+``datasets.config.IN_MEMORY_MAX_SIZE`` to a nonzero size (in bytes) that fits in your RAM memory. In that case, the
+dataset will be copied in-memory if its size is smaller than ``datasets.config.IN_MEMORY_MAX_SIZE`` bytes, and
+memory-mapped otherwise. This behavior can be enabled by setting either the configuration option
+``datasets.config.IN_MEMORY_MAX_SIZE`` (higher precedence) or the environment variable
+``HF_DATASETS_IN_MEMORY_MAX_SIZE`` (lower precedence) to nonzero.

src/datasets/arrow_dataset.py

@@ -660,10 +660,9 @@ def load_from_disk(dataset_path: str, fs=None, keep_in_memory: Optional[bool] =
             fs (:class:`~filesystems.S3FileSystem`, ``fsspec.spec.AbstractFileSystem``, optional, default ``None``):
                 Instance of the remote filesystem used to download the files from.
             keep_in_memory (:obj:`bool`, default ``None``): Whether to copy the dataset in-memory. If `None`, the
-                dataset will be copied in-memory if its size is smaller than `datasets.config.IN_MEMORY_MAX_SIZE`
-                (default ``250 * 2 ** 20`` B). This behavior can be disabled (i.e., the dataset will not be loaded in
-                memory) by setting to ``0`` either the configuration option ``datasets.config.IN_MEMORY_MAX_SIZE``
-                (higher precedence) or the environment variable ``HF_DATASETS_IN_MEMORY_MAX_SIZE`` (lower precedence).
+                dataset will not be copied in-memory unless explicitly enabled by setting
+                `datasets.config.IN_MEMORY_MAX_SIZE` to nonzero. See more details in the
+                :ref:`load_dataset_enhancing_performance` section.
 
         Returns:
             :class:`Dataset` or :class:`DatasetDict`.

src/datasets/config.py

@@ -144,7 +144,7 @@
     HF_DATASETS_OFFLINE = False
 
 # In-memory
-DEFAULT_IN_MEMORY_MAX_SIZE = 250 * 2 ** 20  # 250 MiB
+DEFAULT_IN_MEMORY_MAX_SIZE = 0  # Disabled
 IN_MEMORY_MAX_SIZE = float(os.environ.get("HF_DATASETS_IN_MEMORY_MAX_SIZE", DEFAULT_IN_MEMORY_MAX_SIZE))
 
 # File names

src/datasets/dataset_dict.py

@@ -706,10 +706,9 @@ def load_from_disk(dataset_dict_path: str, fs=None, keep_in_memory: Optional[boo
             fs (:class:`~filesystems.S3FileSystem` or ``fsspec.spec.AbstractFileSystem``, optional, default ``None``):
                 Instance of the remote filesystem used to download the files from.
             keep_in_memory (:obj:`bool`, default ``None``): Whether to copy the dataset in-memory. If `None`, the
-                dataset will be copied in-memory if its size is smaller than `datasets.config.IN_MEMORY_MAX_SIZE`
-                (default ``250 * 2 ** 20`` B). This behavior can be disabled (i.e., the dataset will not be loaded in
-                memory) by setting to ``0`` either the configuration option ``datasets.config.IN_MEMORY_MAX_SIZE``
-                (higher precedence) or the environment variable ``HF_DATASETS_IN_MEMORY_MAX_SIZE`` (lower precedence).
+                dataset will not be copied in-memory unless explicitly enabled by setting
+                `datasets.config.IN_MEMORY_MAX_SIZE` to nonzero. See more details in the
+                :ref:`load_dataset_enhancing_performance` section.
 
         Returns:
             :class:`DatasetDict`

src/datasets/load.py

@@ -683,10 +683,8 @@ def load_dataset(
         download_mode (:class:`GenerateMode`, optional): Select the download/generate mode - Default to REUSE_DATASET_IF_EXISTS
         ignore_verifications (:obj:`bool`, default ``False``): Ignore the verifications of the downloaded/processed dataset information (checksums/size/splits/...).
         keep_in_memory (:obj:`bool`, default ``None``): Whether to copy the dataset in-memory. If `None`, the dataset
-            will be copied in-memory if its size is smaller than `datasets.config.IN_MEMORY_MAX_SIZE` (default
-            ``250 * 2 ** 20`` B). This behavior can be disabled (i.e., the dataset will not be loaded in memory) by
-            setting to ``0`` either the configuration option ``datasets.config.IN_MEMORY_MAX_SIZE`` (higher precedence)
-            or the environment variable ``HF_DATASETS_IN_MEMORY_MAX_SIZE`` (lower precedence).
+            will not be copied in-memory unless explicitly enabled by setting `datasets.config.IN_MEMORY_MAX_SIZE` to
+            nonzero. See more details in the :ref:`load_dataset_enhancing_performance` section.
         save_infos (:obj:`bool`, default ``False``): Save the dataset information (checksums/size/splits/...).
         script_version (:class:`~utils.Version` or :obj:`str`, optional): Version of the dataset script to load:
 
@@ -775,10 +773,8 @@ def load_from_disk(dataset_path: str, fs=None, keep_in_memory: Optional[bool] =
         fs (:class:`~filesystems.S3FileSystem` or ``fsspec.spec.AbstractFileSystem``, optional, default ``None``):
             Instance of of the remote filesystem used to download the files from.
         keep_in_memory (:obj:`bool`, default ``None``): Whether to copy the dataset in-memory. If `None`, the dataset
-            will be copied in-memory if its size is smaller than `datasets.config.IN_MEMORY_MAX_SIZE` (default
-            ``250 * 2 ** 20`` B). This behavior can be disabled (i.e., the dataset will not be loaded in memory) by
-            setting to ``0`` either the configuration option ``datasets.config.IN_MEMORY_MAX_SIZE`` (higher precedence)
-            or the environment variable ``HF_DATASETS_IN_MEMORY_MAX_SIZE`` (lower precedence).
+            will not be copied in-memory unless explicitly enabled by setting `datasets.config.IN_MEMORY_MAX_SIZE` to
+            nonzero. See more details in the :ref:`load_dataset_enhancing_performance` section.
 
     Returns:
         ``datasets.Dataset`` or ``datasets.DatasetDict``

tests/test_info_utils.py

@@ -23,7 +23,7 @@ def test_is_small_dataset(
         if env_max_in_memory_dataset_size:
             assert max_in_memory_dataset_size == env_max_in_memory_dataset_size
         else:
-            assert max_in_memory_dataset_size == 250 * 2 ** 20
+            assert max_in_memory_dataset_size == 0
     else:
         assert max_in_memory_dataset_size == config_max_in_memory_dataset_size
     if dataset_size and max_in_memory_dataset_size:

tests/test_load.py

@@ -233,8 +233,7 @@ def test_load_dataset_local_with_default_in_memory(
 ):
     current_dataset_size = 148
     if max_in_memory_dataset_size == "default":
-        # default = 250 * 2 ** 20
-        max_in_memory_dataset_size = datasets.config.IN_MEMORY_MAX_SIZE
+        max_in_memory_dataset_size = 0  # default
     else:
         monkeypatch.setattr(datasets.config, "IN_MEMORY_MAX_SIZE", max_in_memory_dataset_size)
     if max_in_memory_dataset_size:
@@ -253,8 +252,7 @@ def test_load_from_disk_with_default_in_memory(
 ):
     current_dataset_size = 512  # arrow file size = 512, in-memory dataset size = 148
     if max_in_memory_dataset_size == "default":
-        # default = 250 * 2 ** 20
-        max_in_memory_dataset_size = datasets.config.IN_MEMORY_MAX_SIZE
+        max_in_memory_dataset_size = 0  # default
     else:
         monkeypatch.setattr(datasets.config, "IN_MEMORY_MAX_SIZE", max_in_memory_dataset_size)
     if max_in_memory_dataset_size: