Warning -> info about defaults from v0.7.0

streaming/base/dataset.py

@@ -414,10 +414,8 @@ def __init__(self,
                           f'This may result in slower batch time. Recommendation is to set ' +
                           f'predownload to at-least batch_size.')
         elif self.predownload is None:
-            logger.warning(f'Because `predownload` was not specified, it will default to ' +
-                           f'8*batch_size if batch_size is not None, otherwise 64. Prior to ' +
-                           f'Streaming v0.7.0, `predownload` defaulted to ' +
-                           f'max(batch_size, 256 * batch_size // num_canonical_nodes).')
+            logger.info(f'Because `predownload` was not specified, it will default to ' +
+                        f'8*batch_size if batch_size is not None, otherwise 64.')
             self.predownload = 8 * self.batch_size if self.batch_size is not None else 64
 
         # Convert epoch size from string to int, if needed. Cannot be negative.
@@ -669,10 +667,9 @@ def _set_shuffle_block_size(self, world: World):
         """Set the shuffle block size value."""
         if self.shuffle_block_size is None:
             if not world.worker_of_rank:
-                logger.warning(f'Because `shuffle_block_size` was not specified, it will ' +
-                               f'default to max(4_000_000 // num_canonical_nodes, 1 << 18) if ' +
-                               f'num_canonical_nodes is not None, otherwise 262144. Prior to ' +
-                               f'Streaming v0.7.0, `shuffle_block_size` defaulted to 262144.')
+                logger.info(f'Because `shuffle_block_size` was not specified, it will ' +
+                            f'default to max(4_000_000 // num_canonical_nodes, 1 << 18) if ' +
+                            f'num_canonical_nodes is not None, otherwise 262144.')
             self.shuffle_block_size = max(4_000_000 // self.num_canonical_nodes, 1 << 18) \
                 if self.num_canonical_nodes is not None else 1 << 18
 
@@ -697,12 +694,9 @@ def _resume(self, world: World, epoch: int) -> tuple[int, int]:
                     self.num_canonical_nodes = 64 * world.num_nodes
                 else:
                     if not world.worker_of_rank:
-                        logger.warning(
-                            f'Because `num_canonical_nodes` was not specified, and ' +
-                            f'`shuffle_algo` is {self.shuffle_algo}, it will default to ' +
-                            f'be equal to physical nodes. Prior to Streaming ' +
-                            f'v0.7.0, `num_canonical_nodes` defaulted to 64 * physical ' +
-                            f'nodes.')
+                        logger.info(f'Because `num_canonical_nodes` was not specified, and ' +
+                                    f'`shuffle_algo` is {self.shuffle_algo}, it will default to ' +
+                                    f'be equal to physical nodes.')
                     self.num_canonical_nodes = world.num_nodes
             self._set_shuffle_block_size(world)
             return epoch, 0
@@ -723,9 +717,7 @@ def _resume(self, world: World, epoch: int) -> tuple[int, int]:
                         logger.warning(
                             f'Because `num_canonical_nodes` was not specified, and ' +
                             f'`shuffle_algo` is {self.shuffle_algo}, it will default to ' +
-                            f'be equal to physical nodes. Prior to Streaming ' +
-                            f'v0.7.0, `num_canonical_nodes` defaulted to 64 * physical ' +
-                            f'nodes.')
+                            f'be equal to physical nodes.')
                     self.num_canonical_nodes = world.num_nodes
             self._set_shuffle_block_size(world)
             return epoch, 0

tests/test_streaming.py

@@ -1,12 +1,11 @@
 # Copyright 2022-2024 MosaicML Streaming authors
 # SPDX-License-Identifier: Apache-2.0
 
-import logging
 import math
 import os
 import shutil
 from multiprocessing import Process
-from typing import Any, Callable
+from typing import Any
 
 import pytest
 from torch.utils.data import DataLoader
@@ -65,26 +64,6 @@ def test_no_batch_size_exception(local_remote_dir: tuple[str, str]):
             pass
 
 
-@pytest.mark.usefixtures('local_remote_dir')
-def test_new_defaults_warning(local_remote_dir: tuple[str, str], caplog: Callable):
-    caplog.set_level(logging.WARNING)
-    local, remote = local_remote_dir
-    convert_to_mds(out_root=remote,
-                   dataset_name='sequencedataset',
-                   num_samples=100,
-                   size_limit=1 << 8)
-
-    # Build a StreamingDataset with new defaults. Should warn about the new defaults changes.
-    dataset = StreamingDataset(local=local, remote=remote, shuffle=True, batch_size=4)
-    dataloader = StreamingDataLoader(dataset=dataset, batch_size=4)
-    for _ in dataloader:
-        pass
-
-    assert 'Because `predownload` was not specified,' in caplog.text
-    assert 'Because `shuffle_block_size` was not specified,' in caplog.text
-    assert 'Because `num_canonical_nodes` was not specified,' in caplog.text
-
-
 @pytest.mark.parametrize('batch_size', [4, 7])
 @pytest.mark.parametrize('seed', [2222, 1000])
 @pytest.mark.parametrize('num_canonical_nodes', [2, 4])