Merge branch 'main' into main

src/datasets/iterable_dataset.py

@@ -1431,10 +1431,18 @@ def n_shards(self) -> int:
 
 
 class SkipExamplesIterable(_BaseExamplesIterable):
-    def __init__(self, ex_iterable: _BaseExamplesIterable, n: int):
+    def __init__(
+        self,
+        ex_iterable: _BaseExamplesIterable,
+        n: int,
+        block_sources_order_when_shuffling: bool = True,
+        split_when_sharding: bool = True,
+    ):
         super().__init__()
         self.ex_iterable = ex_iterable
         self.n = n
+        self.block_sources_order_when_shuffling = block_sources_order_when_shuffling
+        self.split_when_sharding = split_when_sharding
         # TODO(QL): implement iter_arrow
 
     def _init_state_dict(self) -> dict:
@@ -1447,20 +1455,57 @@ def __iter__(self):
             self._state_dict["skipped"] = True
         yield from islice(self.ex_iterable, ex_iterable_idx_start, None)
 
+    @staticmethod
+    def split_number(num, n):
+        quotient = num // n
+        remainder = num % n
+        result = [quotient] * n
+        for i in range(remainder):
+            result[i] += 1
+        return result
+
     def shuffle_data_sources(self, generator: np.random.Generator) -> "SkipExamplesIterable":
-        """Doesn't shuffle the wrapped examples iterable since it would skip examples from other shards instead."""
-        return self
+        """May not shuffle the wrapped examples iterable since it would skip examples from other shards instead."""
+        if self.block_sources_order_when_shuffling:
+            return self
+        else:
+            return SkipExamplesIterable(
+                self.ex_iterable.shuffle_data_sources(generator),
+                n=self.n,
+                block_sources_order_when_shuffling=self.block_sources_order_when_shuffling,
+                split_when_sharding=self.split_when_sharding,
+            )
+
+    def shard_data_sources(self, worker_id: int, num_workers: int) -> "SkipExamplesIterable":
+        """Keep only the requested shard."""
+        if self.split_when_sharding:
+            return SkipExamplesIterable(
+                self.ex_iterable.shard_data_sources(worker_id, num_workers),
+                n=self.split_number(self.n, num_workers)[worker_id],
+                block_sources_order_when_shuffling=self.block_sources_order_when_shuffling,
+                split_when_sharding=self.split_when_sharding,
+            )
+        else:
+            return self
 
     @property
     def n_shards(self) -> int:
         return self.ex_iterable.n_shards
 
 
 class TakeExamplesIterable(_BaseExamplesIterable):
-    def __init__(self, ex_iterable: _BaseExamplesIterable, n: int):
+    def __init__(
+        self,
+        ex_iterable: _BaseExamplesIterable,
+        n: int,
+        block_sources_order_when_shuffling: bool = True,
+        split_when_sharding: bool = True,
+    ):
         super().__init__()
         self.ex_iterable = ex_iterable
         self.n = n
+        self.block_sources_order_when_shuffling = block_sources_order_when_shuffling
+        self.split_when_sharding = split_when_sharding
         # TODO(QL): implement iter_arrow
 
     def _init_state_dict(self) -> dict:
@@ -1474,10 +1519,6 @@ def __iter__(self):
                 self._state_dict["num_taken"] += 1
             yield key_example
 
-    def shuffle_data_sources(self, generator: np.random.Generator) -> "TakeExamplesIterable":
-        """Doesn't shuffle the wrapped examples iterable since it would take examples from other shards instead."""
-        return self
-
     @staticmethod
     def split_number(num, n):
         quotient = num // n
@@ -1487,12 +1528,34 @@ def split_number(num, n):
             result[i] += 1
         return result
 
+    def shuffle_data_sources(self, generator: np.random.Generator) -> "TakeExamplesIterable":
+        """May not shuffle the wrapped examples iterable since it would take examples from other shards instead."""
+        if self.block_sources_order_when_shuffling:
+            return self
+        else:
+            return TakeExamplesIterable(
+                self.ex_iterable.shuffle_data_sources(generator),
+                n=self.n,
+                block_sources_order_when_shuffling=self.block_sources_order_when_shuffling,
+                split_when_sharding=self.split_when_sharding,
+            )
+
     def shard_data_sources(self, worker_id: int, num_workers: int) -> "TakeExamplesIterable":
         """Keep only the requested shard."""
-        return TakeExamplesIterable(
-            self.ex_iterable.shard_data_sources(worker_id, num_workers),
-            n=self.split_number(self.n, num_workers)[worker_id],
-        )
+        if self.split_when_sharding:
+            return TakeExamplesIterable(
+                self.ex_iterable.shard_data_sources(worker_id, num_workers),
+                n=self.split_number(self.n, num_workers)[worker_id],
+                block_sources_order_when_shuffling=self.block_sources_order_when_shuffling,
+                split_when_sharding=self.split_when_sharding,
+            )
+        else:
+            return TakeExamplesIterable(
+                self.ex_iterable.shard_data_sources(worker_id, num_workers),
+                n=self.n,
+                block_sources_order_when_shuffling=self.block_sources_order_when_shuffling,
+                split_when_sharding=self.split_when_sharding,
+            )
 
     @property
     def n_shards(self) -> int:
@@ -2392,7 +2455,7 @@ def shuffle(
         return IterableDataset(
             ex_iterable=BufferShuffledExamplesIterable(
                 self._ex_iterable, buffer_size=buffer_size, generator=generator
-            ).shuffle_data_sources(generator),
+            ),
             info=self._info.copy(),
             split=self._split,
             formatting=self._formatting,
@@ -2432,7 +2495,12 @@ def skip(self, n: int) -> "IterableDataset":
          'text': 'if you sometimes like to go to the movies to have fun , wasabi is a good place to start .'}]
         ```
         """
-        ex_iterable = SkipExamplesIterable(self._ex_iterable, n)
+        ex_iterable = SkipExamplesIterable(
+            self._ex_iterable,
+            n,
+            block_sources_order_when_shuffling=self._shuffling is None,
+            split_when_sharding=self._distributed is None,
+        )
         return IterableDataset(
             ex_iterable=ex_iterable,
             info=self._info.copy(),
@@ -2464,7 +2532,12 @@ def take(self, n: int) -> "IterableDataset":
          'text': 'the gorgeously elaborate continuation of " the lord of the rings " trilogy is so huge that a column of words cannot adequately describe co-writer/director peter jackson\'s expanded vision of j . r . r . tolkien\'s middle-earth .'}]
         ```
         """
-        ex_iterable = TakeExamplesIterable(self._ex_iterable, n)
+        ex_iterable = TakeExamplesIterable(
+            self._ex_iterable,
+            n,
+            block_sources_order_when_shuffling=self._shuffling is None,
+            split_when_sharding=self._distributed is None,
+        )
         return IterableDataset(
             ex_iterable=ex_iterable,
             info=self._info.copy(),

tests/test_inspect.py

@@ -72,7 +72,7 @@ def test_get_dataset_config_info_error(path, config_name, expected_exception):
         ("squad", ["plain_text"]),
         ("hf-internal-testing/dataset_with_script", ["default"]),
         ("dalle-mini/wit", ["default"]),
-        ("hf-internal-testing/librispeech_asr_dummy", ["clean", "other"]),
+        ("hf-internal-testing/librispeech_asr_dummy", ["clean"]),
         ("hf-internal-testing/audiofolder_no_configs_in_metadata", ["default"]),
         ("hf-internal-testing/audiofolder_single_config_in_metadata", ["custom"]),
         ("hf-internal-testing/audiofolder_two_configs_in_metadata", ["v1", "v2"]),
@@ -90,7 +90,7 @@ def test_get_dataset_config_names(path, expected):
         ("squad", "plain_text"),
         ("hf-internal-testing/dataset_with_script", "default"),
         ("dalle-mini/wit", "default"),
-        ("hf-internal-testing/librispeech_asr_dummy", None),
+        ("hf-internal-testing/librispeech_asr_dummy", "clean"),
         ("hf-internal-testing/audiofolder_no_configs_in_metadata", "default"),
         ("hf-internal-testing/audiofolder_single_config_in_metadata", "custom"),
         ("hf-internal-testing/audiofolder_two_configs_in_metadata", None),

tests/test_iterable_dataset.py

@@ -10,6 +10,7 @@
 
 from datasets import Dataset, load_dataset
 from datasets.combine import concatenate_datasets, interleave_datasets
+from datasets.distributed import split_dataset_by_node
 from datasets.features import (
     ClassLabel,
     Features,
@@ -1657,17 +1658,60 @@ def test_iterable_dataset_take(dataset: IterableDataset, n):
 
 
 @pytest.mark.parametrize("method", ["skip", "take"])
-def test_iterable_dataset_shuffle_after_skip_or_take(method):
+@pytest.mark.parametrize("after_shuffle", [False, True])
+@pytest.mark.parametrize("count", [2, 5, 11])
+def test_iterable_dataset_skip_or_take_after_shuffle(method, after_shuffle, count):
     seed = 42
     n, n_shards = 3, 10
-    count = 7
     ex_iterable = ExamplesIterable(generate_examples_fn, {"n": n, "filepaths": [f"{i}.txt" for i in range(n_shards)]})
     dataset = IterableDataset(ex_iterable)
-    dataset = dataset.skip(n) if method == "skip" else dataset.take(count)
-    shuffled_dataset = dataset.shuffle(seed, buffer_size=DEFAULT_N_EXAMPLES)
-    # shuffling a skip/take dataset should keep the same examples and don't shuffle the shards
-    key = lambda x: f"{x['filepath']}_{x['id']}"  # noqa: E731
-    assert sorted(dataset, key=key) == sorted(shuffled_dataset, key=key)
+    shuffled_dataset = dataset
+    if after_shuffle:
+        shuffled_dataset = shuffled_dataset.shuffle(seed, buffer_size=DEFAULT_N_EXAMPLES)
+        shuffled_dataset = shuffled_dataset.skip(count) if method == "skip" else shuffled_dataset.take(count)
+        # skip/take a shuffled dataset should not keep the same examples and shuffle the shards
+        key = lambda x: f"{x['filepath']}_{x['id']}"  # noqa: E731
+        assert (len(list(dataset)) - count if method == "skip" else count) == len(list(shuffled_dataset))
+        assert sorted(list(dataset)[count:] if method == "skip" else list(dataset)[:count], key=key) != sorted(
+            shuffled_dataset, key=key
+        )
+    else:
+        shuffled_dataset = shuffled_dataset.skip(count) if method == "skip" else shuffled_dataset.take(count)
+        shuffled_dataset = shuffled_dataset.shuffle(seed, buffer_size=DEFAULT_N_EXAMPLES)
+        # shuffling a skip/take dataset should keep the same examples and don't shuffle the shards
+        key = lambda x: f"{x['filepath']}_{x['id']}"  # noqa: E731
+        assert (len(list(dataset)) - count if method == "skip" else count) == len(list(shuffled_dataset))
+        assert sorted(list(dataset)[count:] if method == "skip" else list(dataset)[:count], key=key) == sorted(
+            shuffled_dataset, key=key
+        )
+
+
+@pytest.mark.parametrize("method", ["skip", "take"])
+@pytest.mark.parametrize("after_split_by_node", [False, True])
+@pytest.mark.parametrize("count", [2, 5, 11])
+def test_iterable_dataset_skip_or_take_after_split_by_node(method, after_split_by_node, count):
+    n, n_shards = 3, 10
+    rank, world_size = 1, 2
+    ex_iterable = ExamplesIterable(generate_examples_fn, {"n": n, "filepaths": [f"{i}.txt" for i in range(n_shards)]})
+    dataset = IterableDataset(ex_iterable)
+    distributed_dataset = dataset
+    true_distributed_dataset = split_dataset_by_node(dataset, rank=rank, world_size=world_size)
+    if after_split_by_node:
+        distributed_dataset = split_dataset_by_node(distributed_dataset, rank=rank, world_size=world_size)
+        distributed_dataset = distributed_dataset.skip(count) if method == "skip" else distributed_dataset.take(count)
+        assert (
+            list(true_distributed_dataset)[count:]
+            if method == "skip"
+            else list(true_distributed_dataset)[:count] == list(distributed_dataset)
+        )
+    else:
+        distributed_dataset = distributed_dataset.skip(count) if method == "skip" else distributed_dataset.take(count)
+        distributed_dataset = split_dataset_by_node(distributed_dataset, rank=rank, world_size=world_size)
+        assert len(
+            list(true_distributed_dataset)[count // world_size :]
+            if method == "skip"
+            else list(true_distributed_dataset)[: count // world_size]
+        ) == len(list(distributed_dataset))
 
 
 def test_iterable_dataset_add_column(dataset_with_several_columns: IterableDataset):