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

Jump to bottom

Feat/add support for numpy datatypes in tokensloader #401

Draft
wants to merge 7 commits into
base: main
Choose a base branch
from
Draft

Feat/add support for numpy datatypes in tokensloader #401

Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
84f0846
wip: adds support for numpy datatypes in tokensloader
bhimrazy Oct 27, 2024
3bad200
updated itemloader
bhimrazy Oct 27, 2024
defe908
update
bhimrazy Oct 27, 2024
3625fca
Merge branch 'main' into feat/add-support-for-numpy-datatypes-in-toke…
bhimrazy Oct 28, 2024
e7709d1
removed numpy array to tensor conversion in tokens loader
bhimrazy Oct 28, 2024
667a777
update
bhimrazy Oct 28, 2024
c5d6f03
Merge branch 'main' into feat/add-support-for-numpy-datatypes-in-toke…
bhimrazy Oct 28, 2024
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
23 changes: 18 additions & 5 deletions src/litdata/streaming/item_loader.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,9 +23,7 @@
import numpy as np
import torch

from litdata.constants import (
_TORCH_DTYPES_MAPPING,
)
from litdata.constants import _NUMPY_DTYPES_MAPPING, _TORCH_DTYPES_MAPPING
from litdata.streaming.serializers import Serializer
from litdata.utilities._pytree import PyTree, tree_unflatten
from litdata.utilities.encryption import Encryption, EncryptionLevel
Expand Down Expand Up @@ -281,7 +279,17 @@ def setup(
region_of_interest: Optional[List[Tuple[int, int]]] = None,
) -> None:
super().setup(config, chunks, serializers, region_of_interest)
self._dtype = _TORCH_DTYPES_MAPPING[int(config["data_format"][0].split(":")[1])]

serializer_name, dtype_index = self._data_format[0].split(":")
if serializer_name not in ["no_header_numpy", "no_header_tensor"]:
raise ValueError("The provided data format isn't supported.")

self._serializer_name = serializer_name
self._dtype = (
_TORCH_DTYPES_MAPPING[int(dtype_index)]
if serializer_name == "no_header_tensor"
else _NUMPY_DTYPES_MAPPING[int(dtype_index)]
)
if all(chunk["dim"] is None for chunk in self._chunks):
raise ValueError("The provided chunks isn't properly setup.")

Expand Down Expand Up @@ -350,7 +358,12 @@ def load_item_from_chunk(

buffer: bytes = self._buffers[chunk_index]
offset = self._dtype.itemsize * (index - begin) * self._block_size
return torch.frombuffer(buffer, dtype=self._dtype, count=self._block_size, offset=offset)

if self._serializer_name == "no_header_tensor":
data = torch.frombuffer(buffer, dtype=self._dtype, count=self._block_size, offset=offset)
else:
data = np.frombuffer(buffer, dtype=self._dtype, count=self._block_size, offset=offset) # type: ignore
return data

def delete(self, chunk_index: int, chunk_filepath: str) -> None:
if os.path.exists(chunk_filepath):
Expand Down
Loading