# Copyright 2022 Cerebras Systems.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Pytorch GPT2/3 Dataloader"""
import logging
from cerebras.modelzoo.common.registry import registry
from cerebras.modelzoo.data.common.HDF5IterableDataProcessor import (
HDF5IterableDataProcessor,
)
from cerebras.modelzoo.data.common.HDF5IterableDataset import (
HDF5IterableDataset,
)
[docs]@registry.register_datasetprocessor("GptHDF5DataProcessor")
class GptHDF5DataProcessor(HDF5IterableDataProcessor):
"""
A HDF5 dataset processor for GPT pre-training.
Loads data from HDF5 files.
:param dict params: dict containing training
input parameters for creating dataset.
Expects the following fields:
- "data_dir" (str or list of str): Path to dataset HDF5 files
- "batch_size" (int): Batch size.
- "shuffle" (bool): Flag to enable data shuffling.
- "shuffle_buffer" (int): Size of shuffle buffer in samples.
- "shuffle_seed" (int): Shuffle seed.
- "num_workers" (int): How many subprocesses to use for data loading.
- "drop_last" (bool): If True and the dataset size is not divisible
by the batch size, the last incomplete batch will be dropped.
- "prefetch_factor" (int): Number of batches loaded in advance by each worker.
- "persistent_workers" (bool): If True, the data loader will not shutdown
the worker processes after a dataset has been consumed once.
- "use_vsl" (bool): Flag to enable variable sequence length training.
It requires the dataset to have two extra features: the
`attention_span` of keys and the `position_ids` of tokens.
Defaults to `False`.
"""
[docs] def __init__(self, params):
if params.get("max_sequence_length") is not None:
logging.warning(
"`max_sequence_length` is not used in for GptHDF5DataProcessor "
"as it expects the data to be pre tokenized to a desired MSL, "
"please remove it from the supplied config."
)
# HDF5IterableDataset yields samples with the features `input_ids`,
# `attention_mask`, and `labels`. In the case of gpt models, the name
# `attention_mask` is a misnomer. In reality it acts as a loss mask
# and its contents aren't used for any attention masking.
self.dataset = HDF5IterableDataset(params)
# The super class will take care of sharding the dataset and creating the dataloader
super().__init__(params)