Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Resolve issue 308 #309

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Resolve issue 308 #309

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion pyproject.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -41,7 +41,7 @@ io = [
"pyarrow",
"python-magic",
"warcio",
"datasets>=2.18.0",
"datasets>=3.1.0",
"orjson",
"zstandard"
]
Expand Down
17 changes: 15 additions & 2 deletions src/datatrove/pipeline/readers/huggingface.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -76,7 +76,20 @@ def _get_dataset_shard(self, dst, rank: int, world_size: int):
f"Requested shard {rank} of a streaming dataset, but it only has {dst.n_shards} shards."
)
return None
ex_iterable = dst._ex_iterable.shard_data_sources(rank, world_size)

# https://github.com/huggingface/datatrove/issues/308
# huggingface/datasets@65f6eb5#diff-edc4da5f2179552e25f4f3dc9d6bf07265b68bbef048a8f712e798520a23d048L103
# Order of the arguments to shard_data_sources function in datasets lib changed. Make sure current version of the datasets is up-to-date.
import inspect
ex_iterable = dst._ex_iterable.shard_data_sources(world_size, rank)
arg_names = inspect.signature(dst._ex_iterable.shard_data_sources).parameters

# Assert that the first argument is not "worker_id"
assert list(arg_names.keys())[0]!= "worker_id", "The first argument to shard_data_sources cannot be named 'worker_id'. Make sure datasets version is up-to-date"
# Assert that the second argument is not "num_workers"
assert list(arg_names.keys())[1]!= "num_workers", "The second argument to shard_data_sources cannot be named 'num_workers'. Make sure datasets version is up-to-date"

ex_iterable = dst._ex_iterable.shard_data_sources(world_size, rank)
return IterableDataset(
ex_iterable=ex_iterable,
info=dst._info.copy(),
Expand All @@ -96,7 +109,7 @@ def run(self, data: DocumentsPipeline = None, rank: int = 0, world_size: int = 1
if data:
yield from data
ds = load_dataset(self.dataset, **self.dataset_options, streaming=self.streaming)

if self.shuffle_files:
if not self.streaming:
ds = ds.shuffle(seed=42)
Expand Down
130 changes: 116 additions & 14 deletions tests/pipeline/test_hf_reader.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,17 +47,119 @@ def test_read_streaming_dataset_shuffle(self):
self.assertEqual(len(data[0].text), 69)
self.assertEqual(len(data[1].text), 46)

def test_sharding(self):
for shards in [1, 3]:
for streaming in [True, False]:
reader = HuggingFaceDatasetReader(
"huggingface/datatrove-tests",
dataset_options={"name": f"sharding-{shards}", "split": "train"},
text_key="text",
streaming=streaming,
)
data0 = list(reader(rank=0, world_size=2))
data1 = list(reader(rank=1, world_size=2))

self.assertEqual(len(data0), 3)
self.assertEqual(len(data1), 2)
def test_sharding_1(self):
"""
>>> ds = load_dataset("huggingface/datatrove-tests",name="sharding-1",split="train",streaming=True)
>>> ds
IterableDataset({
features: ['text'],
num_shards: 1
})

>>> print(list(ds.shard(num_shards=2, index=0)))
[{'text': 'hello'}, {'text': 'world'}, {'text': 'how'}, {'text': 'are'}, {'text': 'you'}]

>>> print(list(ds.shard(num_shards=2, index=1)))
IndexError: list index out of range

>>> ds = load_dataset("huggingface/datatrove-tests",name="sharding-1",split="train",streaming=False)
>>> ds
Dataset({
features: ['text'],
num_rows: 5
})

>>> print(list(ds.shard(num_shards=2, index=0)))
>>> print(list(ds.shard(num_shards=2, index=1)))
[{'text': 'hello'}, {'text': 'world'}, {'text': 'how'}]
[{'text': 'are'}, {'text': 'you'}]

>>> print(list(ds.shard(num_shards=3, index=0)))
>>> print(list(ds.shard(num_shards=3, index=1)))
>>> print(list(ds.shard(num_shards=3, index=2)))
[{'text': 'hello'}, {'text': 'world'}]
[{'text': 'how'}, {'text': 'are'}]
[{'text': 'you'}]

"""
for streaming in [True, False]:
reader = HuggingFaceDatasetReader(
"huggingface/datatrove-tests",
dataset_options={"name": f"sharding-1", "split": "train"},
text_key="text",
streaming=streaming,
)
data0 = list(reader(rank=0, world_size=2))
data1 = list(reader(rank=1, world_size=2))

self.assertEqual(len(data0), 3)
self.assertEqual(len(data1), 2)

def test_sharding_3_stream(self):
"""
>>> ds_stream = load_dataset("huggingface/datatrove-tests",name="sharding-3",split="train",streaming=True)
>>> ds_stream
IterableDataset({
features: ['text'],
num_shards: 3
})

>>> print(list(ds_stream.shard(num_shards=2, index=0)))
>>> print(list(ds_stream.shard(num_shards=2, index=1)))
[{'text': 'hello'}, {'text': 'world'}, {'text': 'how'}, {'text': 'are'}]
[{'text': 'you'}]

>>> print(list(list(ds_stream.shard(num_shards=3, index=0))))
>>> print(list(list(ds_stream.shard(num_shards=3, index=1))))
>>> print(list(list(ds_stream.shard(num_shards=3, index=2))))
[{'text': 'hello'}, {'text': 'world'}]
[{'text': 'how'}, {'text': 'are'}]
[{'text': 'you'}]

"""
reader = HuggingFaceDatasetReader(
"huggingface/datatrove-tests",
dataset_options={"name": f"sharding-3", "split": "train"},
text_key="text",
streaming=True,
)
data0 = list(reader(rank=0, world_size=2))
data1 = list(reader(rank=1, world_size=2))

self.assertEqual(len(data0), 4)
self.assertEqual(len(data1), 1)

def test_sharding_3(self):
"""
>>> ds = load_dataset("huggingface/datatrove-tests",name="sharding-3",split="train",streaming=False)
>>> ds
Dataset({
features: ['text'],
num_rows: 5
})

>>> print(list(ds.shard(num_shards=2, index=0)))
>>> print(list(ds.shard(num_shards=2, index=1)))
[{'text': 'hello'}, {'text': 'world'}, {'text': 'how'}]
[{'text': 'are'}, {'text': 'you'}]

>>> print(list(ds.shard(num_shards=3, index=0)))
>>> print(list(ds.shard(num_shards=3, index=1)))
>>> print(list(ds.shard(num_shards=3, index=2)))
[{'text': 'hello'}, {'text': 'world'}]
[{'text': 'how'}, {'text': 'are'}]
[{'text': 'you'}]

"""
reader = HuggingFaceDatasetReader(
"huggingface/datatrove-tests",
dataset_options={"name": f"sharding-3", "split": "train"},
text_key="text",
streaming=False,
)
data0 = list(reader(rank=0, world_size=2))
data1 = list(reader(rank=1, world_size=2))

self.assertEqual(len(data0), 3)
self.assertEqual(len(data1), 2)

Loading