Commit 925a6c9f authored by Jan Kocoń's avatar Jan Kocoń

Megatron-LM adjusted to Polish texts

parent 78e7e33c
# Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions
# are met:
# * Redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer.
# * Redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution.
# * Neither the name of NVIDIA CORPORATION nor the names of its
# contributors may be used to endorse or promote products derived
# from this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
# EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
# PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
# CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
# EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
# PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
# PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
# OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
------------- LICENSE FOR huggingface(transformer) repository --------------
Apache License
Version 2.0, January 2004
http://www.apache.org/licenses/
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
1. Definitions.
"License" shall mean the terms and conditions for use, reproduction,
and distribution as defined by Sections 1 through 9 of this document.
"Licensor" shall mean the copyright owner or entity authorized by
the copyright owner that is granting the License.
"Legal Entity" shall mean the union of the acting entity and all
other entities that control, are controlled by, or are under common
control with that entity. For the purposes of this definition,
"control" means (i) the power, direct or indirect, to cause the
direction or management of such entity, whether by contract or
otherwise, or (ii) ownership of fifty percent (50%) or more of the
outstanding shares, or (iii) beneficial ownership of such entity.
"You" (or "Your") shall mean an individual or Legal Entity
exercising permissions granted by this License.
"Source" form shall mean the preferred form for making modifications,
including but not limited to software source code, documentation
source, and configuration files.
"Object" form shall mean any form resulting from mechanical
transformation or translation of a Source form, including but
not limited to compiled object code, generated documentation,
and conversions to other media types.
"Work" shall mean the work of authorship, whether in Source or
Object form, made available under the License, as indicated by a
copyright notice that is included in or attached to the work
(an example is provided in the Appendix below).
"Derivative Works" shall mean any work, whether in Source or Object
form, that is based on (or derived from) the Work and for which the
editorial revisions, annotations, elaborations, or other modifications
represent, as a whole, an original work of authorship. For the purposes
of this License, Derivative Works shall not include works that remain
separable from, or merely link (or bind by name) to the interfaces of,
the Work and Derivative Works thereof.
"Contribution" shall mean any work of authorship, including
the original version of the Work and any modifications or additions
to that Work or Derivative Works thereof, that is intentionally
submitted to Licensor for inclusion in the Work by the copyright owner
or by an individual or Legal Entity authorized to submit on behalf of
the copyright owner. For the purposes of this definition, "submitted"
means any form of electronic, verbal, or written communication sent
to the Licensor or its representatives, including but not limited to
communication on electronic mailing lists, source code control systems,
and issue tracking systems that are managed by, or on behalf of, the
Licensor for the purpose of discussing and improving the Work, but
excluding communication that is conspicuously marked or otherwise
designated in writing by the copyright owner as "Not a Contribution."
"Contributor" shall mean Licensor and any individual or Legal Entity
on behalf of whom a Contribution has been received by Licensor and
subsequently incorporated within the Work.
2. Grant of Copyright License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
copyright license to reproduce, prepare Derivative Works of,
publicly display, publicly perform, sublicense, and distribute the
Work and such Derivative Works in Source or Object form.
3. Grant of Patent License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
(except as stated in this section) patent license to make, have made,
use, offer to sell, sell, import, and otherwise transfer the Work,
where such license applies only to those patent claims licensable
by such Contributor that are necessarily infringed by their
Contribution(s) alone or by combination of their Contribution(s)
with the Work to which such Contribution(s) was submitted. If You
institute patent litigation against any entity (including a
cross-claim or counterclaim in a lawsuit) alleging that the Work
or a Contribution incorporated within the Work constitutes direct
or contributory patent infringement, then any patent licenses
granted to You under this License for that Work shall terminate
as of the date such litigation is filed.
4. Redistribution. You may reproduce and distribute copies of the
Work or Derivative Works thereof in any medium, with or without
modifications, and in Source or Object form, provided that You
meet the following conditions:
(a) You must give any other recipients of the Work or
Derivative Works a copy of this License; and
(b) You must cause any modified files to carry prominent notices
stating that You changed the files; and
(c) You must retain, in the Source form of any Derivative Works
that You distribute, all copyright, patent, trademark, and
attribution notices from the Source form of the Work,
excluding those notices that do not pertain to any part of
the Derivative Works; and
(d) If the Work includes a "NOTICE" text file as part of its
distribution, then any Derivative Works that You distribute must
include a readable copy of the attribution notices contained
within such NOTICE file, excluding those notices that do not
pertain to any part of the Derivative Works, in at least one
of the following places: within a NOTICE text file distributed
as part of the Derivative Works; within the Source form or
documentation, if provided along with the Derivative Works; or,
within a display generated by the Derivative Works, if and
wherever such third-party notices normally appear. The contents
of the NOTICE file are for informational purposes only and
do not modify the License. You may add Your own attribution
notices within Derivative Works that You distribute, alongside
or as an addendum to the NOTICE text from the Work, provided
that such additional attribution notices cannot be construed
as modifying the License.
You may add Your own copyright statement to Your modifications and
may provide additional or different license terms and conditions
for use, reproduction, or distribution of Your modifications, or
for any such Derivative Works as a whole, provided Your use,
reproduction, and distribution of the Work otherwise complies with
the conditions stated in this License.
5. Submission of Contributions. Unless You explicitly state otherwise,
any Contribution intentionally submitted for inclusion in the Work
by You to the Licensor shall be under the terms and conditions of
this License, without any additional terms or conditions.
Notwithstanding the above, nothing herein shall supersede or modify
the terms of any separate license agreement you may have executed
with Licensor regarding such Contributions.
6. Trademarks. This License does not grant permission to use the trade
names, trademarks, service marks, or product names of the Licensor,
except as required for reasonable and customary use in describing the
origin of the Work and reproducing the content of the NOTICE file.
7. Disclaimer of Warranty. Unless required by applicable law or
agreed to in writing, Licensor provides the Work (and each
Contributor provides its Contributions) on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
implied, including, without limitation, any warranties or conditions
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
PARTICULAR PURPOSE. You are solely responsible for determining the
appropriateness of using or redistributing the Work and assume any
risks associated with Your exercise of permissions under this License.
8. Limitation of Liability. In no event and under no legal theory,
whether in tort (including negligence), contract, or otherwise,
unless required by applicable law (such as deliberate and grossly
negligent acts) or agreed to in writing, shall any Contributor be
liable to You for damages, including any direct, indirect, special,
incidental, or consequential damages of any character arising as a
result of this License or out of the use or inability to use the
Work (including but not limited to damages for loss of goodwill,
work stoppage, computer failure or malfunction, or any and all
other commercial damages or losses), even if such Contributor
has been advised of the possibility of such damages.
9. Accepting Warranty or Additional Liability. While redistributing
the Work or Derivative Works thereof, You may choose to offer,
and charge a fee for, acceptance of support, warranty, indemnity,
or other liability obligations and/or rights consistent with this
License. However, in accepting such obligations, You may act only
on Your own behalf and on Your sole responsibility, not on behalf
of any other Contributor, and only if You agree to indemnify,
defend, and hold each Contributor harmless for any liability
incurred by, or claims asserted against, such Contributor by reason
of your accepting any such warranty or additional liability.
END OF TERMS AND CONDITIONS
APPENDIX: How to apply the Apache License to your work.
To apply the Apache License to your work, attach the following
boilerplate notice, with the fields enclosed by brackets "[]"
replaced with your own identifying information. (Don't include
the brackets!) The text should be enclosed in the appropriate
comment syntax for the file format. We also recommend that a
file or class name and description of purpose be included on the
same "printed page" as the copyright notice for easier
identification within third-party archives.
Copyright [yyyy] [name of copyright owner]
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.
# coding=utf-8
# Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
#
# 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.
"""argparser configuration"""
import argparse
import os
import torch
def add_model_config_args(parser):
"""Model arguments"""
group = parser.add_argument_group('model', 'model configuration')
group.add_argument('--pretrained-bert', action='store_true',
help='use a pretrained bert-large-uncased model instead'
'of initializing from scratch. See '
'--tokenizer-model-type to specify which pretrained '
'BERT model to use', default=False)
group.add_argument('--attention-dropout', type=float, default=0.1,
help='dropout probability for attention weights')
group.add_argument('--num-attention-heads', type=int, default=16,
help='num of transformer attention heads')
group.add_argument('--hidden-size', type=int, default=1024,
help='tansformer hidden size')
group.add_argument('--intermediate-size', type=int, default=None,
help='transformer embedding dimension for FFN'
'set to 4*`--hidden-size` if it is None')
group.add_argument('--num-layers', type=int, default=24,
help='num decoder layers')
group.add_argument('--layernorm-epsilon', type=float, default=1e-5,
help='layer norm epsilon')
group.add_argument('--hidden-dropout', type=float, default=0.1,
help='dropout probability for hidden state transformer')
group.add_argument('--max-position-embeddings', type=int, default=512,
help='maximum number of position embeddings to use')
group.add_argument('--vocab-size', type=int, default=32296,
help='vocab size to use for non-character-level '
'tokenization. This value will only be used when '
'creating a tokenizer')
group.add_argument('--deep-init', action='store_true',
help='initialize bert model similar to gpt2 model.'
'scales initialization of projection layers by a '
'factor of 1/sqrt(2N). Necessary to train bert '
'models larger than BERT-Large.')
group.add_argument('--make-vocab-size-divisible-by', type=int, default=128,
help='Pad the vocab size to be divisible by this value.'
'This is added for computational efficieny reasons.')
return parser
def add_fp16_config_args(parser):
"""Mixed precision arguments."""
group = parser.add_argument_group('fp16', 'fp16 configurations')
group.add_argument('--fp16', action='store_true',
help='Run model in fp16 mode')
group.add_argument('--fp32-embedding', action='store_true',
help='embedding in fp32')
group.add_argument('--fp32-layernorm', action='store_true',
help='layer norm in fp32')
group.add_argument('--fp32-tokentypes', action='store_true',
help='embedding token types in fp32')
group.add_argument('--fp32-allreduce', action='store_true',
help='all-reduce in fp32')
group.add_argument('--hysteresis', type=int, default=2,
help='hysteresis for dynamic loss scaling')
group.add_argument('--loss-scale', type=float, default=None,
help='Static loss scaling, positive power of 2 '
'values can improve fp16 convergence. If None, dynamic'
'loss scaling is used.')
group.add_argument('--loss-scale-window', type=float, default=1000,
help='Window over which to raise/lower dynamic scale')
group.add_argument('--min-scale', type=float, default=1,
help='Minimum loss scale for dynamic loss scale')
return parser
def add_training_args(parser):
"""Training arguments."""
group = parser.add_argument_group('train', 'training configurations')
group.add_argument('--batch-size', type=int, default=4,
help='Data Loader batch size')
group.add_argument('--weight-decay', type=float, default=0.01,
help='weight decay coefficient for L2 regularization')
group.add_argument('--checkpoint-activations', action='store_true',
help='checkpoint activation to allow for training '
'with larger models and sequences')
group.add_argument('--checkpoint-num-layers', type=int, default=1,
help='chunk size (number of layers) for checkpointing')
group.add_argument('--clip-grad', type=float, default=1.0,
help='gradient clipping')
group.add_argument('--train-iters', type=int, default=1000000,
help='total number of iterations to train over all training runs')
group.add_argument('--log-interval', type=int, default=100,
help='report interval')
group.add_argument('--exit-interval', type=int, default=None,
help='Exit the program after this many new iterations.')
group.add_argument('--tensorboard-dir', type=str, default=None,
help='Write TensorBoard logs to this directory')
group.add_argument('--seed', type=int, default=1234,
help='random seed')
# Batch prodecuer arguments
group.add_argument('--reset-position-ids', action='store_true',
help='Reset posistion ids after end-of-document token.')
group.add_argument('--reset-attention-mask', action='store_true',
help='Reset self attention maske after '
'end-of-document token.')
group.add_argument('--eod-mask-loss', action='store_true',
help='Mask loss for the end of document tokens')
# Learning rate.
group.add_argument('--lr-decay-iters', type=int, default=None,
help='number of iterations to decay LR over,'
' If None defaults to `--train-iters`*`--epochs`')
group.add_argument('--lr-decay-style', type=str, default='linear',
choices=['constant', 'linear', 'cosine', 'exponential'],
help='learning rate decay function')
group.add_argument('--lr', type=float, default=1.0e-4,
help='initial learning rate')
group.add_argument('--min-lr', type=float, default=0.0,
help='Minumum value for learning rate. The scheduler'
'clip values below this threshold.')
group.add_argument('--warmup', type=float, default=0.01,
help='percentage of data to warmup on (.01 = 1% of all '
'training iters). Default 0.01')
group.add_argument('--override-lr-scheduler', action='store_true',
help='Reset the values of the scheduler (learning rate,'
'warmup iterations, minimum learning rate, maximum '
'number of iterations, and decay style from input '
'arguments and ignore values from checkpoints. Note'
'that all the above values will be reset.')
group.add_argument('--use-checkpoint-lr-scheduler', action='store_true',
help='Use checkpoint to set the values of the scheduler '
'(learning rate, warmup iterations, minimum learning '
'rate, maximum number of iterations, and decay style '
'from input arguments and ignore values from '
'checkpoints. Notethat all the above values will be '
'reset.')
# model checkpointing
group.add_argument('--save', type=str, default=None,
help='Output directory to save checkpoints to.')
group.add_argument('--save-interval', type=int, default=5000,
help='number of iterations between saves')
group.add_argument('--no-save-optim', action='store_true',
help='Do not save current optimizer.')
group.add_argument('--no-save-rng', action='store_true',
help='Do not save current rng state.')
group.add_argument('--load', type=str, default=None,
help='Path to a directory containing a model checkpoint.')
group.add_argument('--no-load-optim', action='store_true',
help='Do not load optimizer when loading checkpoint.')
group.add_argument('--no-load-rng', action='store_true',
help='Do not load rng state when loading checkpoint.')
group.add_argument('--finetune', action='store_true',
help='Load model for finetuning. Do not load optimizer '
'or rng state from checkpoint and set iteration to 0. '
'Assumed when loading a release checkpoint.')
group.add_argument('--resume-dataloader', action='store_true',
help='Resume the dataloader when resuming training. '
'Does not apply to tfrecords dataloader, try resuming'
'with a different seed in this case.')
# distributed training args
group.add_argument('--distributed-backend', default='nccl',
help='which backend to use for distributed '
'training. One of [gloo, nccl]')
group.add_argument('--DDP-impl', default='local',
help='which DistributedDataParallel implementation '
'to use. One of [local, torch]')
group.add_argument('--local_rank', type=int, default=None,
help='local rank passed from distributed launcher')
# autoresume
group.add_argument('--adlr-autoresume', action='store_true',
help='enable autoresume on adlr cluster.')
group.add_argument('--adlr-autoresume-interval', type=int, default=1000,
help='intervals over which check for autoresume'
'termination signal')
return parser
def add_evaluation_args(parser):
"""Evaluation arguments."""
group = parser.add_argument_group('validation', 'validation configurations')
group.add_argument('--eval-batch-size', type=int, default=None,
help='Data Loader batch size for evaluation datasets.'
'Defaults to `--batch-size`')
group.add_argument('--eval-iters', type=int, default=100,
help='number of iterations to run for evaluation'
'validation/test for')
group.add_argument('--eval-interval', type=int, default=1000,
help='interval between running evaluation on validation set')
group.add_argument('--eval-seq-length', type=int, default=None,
help='Maximum sequence length to process for '
'evaluation. Defaults to `--seq-length`')
group.add_argument('--eval-max-preds-per-seq', type=int, default=None,
help='Maximum number of predictions to use for '
'evaluation. Defaults to '
'math.ceil(`--eval-seq-length`*.15/10)*10')
group.add_argument('--overlapping-eval', type=int, default=32,
help='sliding window for overlapping eval ')
group.add_argument('--cloze-eval', action='store_true',
help='Evaluation dataset from `--valid-data` is a cloze task')
group.add_argument('--strict-lambada', action='store_true',
help='use more difficult formulation of lambada')
group.add_argument('--eval-hf', action='store_true',
help='perform evaluation with huggingface openai model.'
'use `--load` to specify weights path to be loaded')
group.add_argument('--load-openai', action='store_true',
help='load openai weights into our model. Use `--load` '
'to specify weights path to be loaded')
return parser
def add_text_generate_args(parser):
"""Text generate arguments."""
group = parser.add_argument_group('Text generation', 'configurations')
group.add_argument("--temperature", type=float, default=1.0)
group.add_argument("--greedy", action='store_true', default=False)
group.add_argument("--top_p", type=float, default=0.0)
group.add_argument("--top_k", type=int, default=0)
group.add_argument("--out-seq-length", type=int, default=1024)
group.add_argument("--sample-input-file", type=str, default="",
help='get input from file instead of interactive mode, '
'each line is an input' )
group.add_argument("--sample-output-file", type=str, default="",
help='output file got from --sample-input-file')
group.add_argument("--num-samples", type=int, default=0,
help='number of samples to generate unconditionally, '
'defaults to 0 and interactive conditional sampling')
group.add_argument("--genfile", type=str,
help='output file when generating unconditionally')
group.add_argument("--recompute", action='store_true',
help='during generation recompute all attention '
'instead of using previously computed keys/values.')
return parser
def add_data_args(parser):
"""Train/valid/test data arguments."""
group = parser.add_argument_group('data', 'data configurations')
group.add_argument('--model-parallel-size', type=int, default=1,
help='size of the model parallel.')
group.add_argument('--shuffle', action='store_true',
help='Shuffle data. Shuffling is deterministic '
'based on seed and current epoch.')
group.add_argument('--train-data', nargs='+', default=None,
help='Whitespace separated filenames or corpora names '
'for training.')
group.add_argument('--use-npy-data-loader', action='store_true',
help='Use the numpy data loader. If set, then'
'train-data-path, val-data-path, and test-data-path'
'should also be provided.')
group.add_argument('--train-data-path', type=str, default='',
help='path to the training data')
group.add_argument('--val-data-path', type=str, default='',
help='path to the validation data')
group.add_argument('--test-data-path', type=str, default='',
help='path to the test data')
group.add_argument('--input-data-sizes-file', type=str, default='sizes.txt',
help='the filename containing all the shards sizes')
group.add_argument('--delim', default=',',
help='delimiter used to parse csv data files')
group.add_argument('--text-key', default='sentence',
help='key to use to extract text from json/csv')
group.add_argument('--eval-text-key', default=None,
help='key to use to extract text from '
'json/csv evaluation datasets')
group.add_argument('--valid-data', nargs='*', default=None,
help="""Filename for validation data.""")
group.add_argument('--split', default='1000,1,1',
help='comma-separated list of proportions for training,'
' validation, and test split')
group.add_argument('--test-data', nargs='*', default=None,
help="""Filename for testing""")
group.add_argument('--lazy-loader', action='store_true',
help='whether to lazy read the data set')
group.add_argument('--loose-json', action='store_true',
help='Use loose json (one json-formatted string per '
'newline), instead of tight json (data file is one '
'json string)')
group.add_argument('--presplit-sentences', action='store_true',
help='Dataset content consists of documents where '
'each document consists of newline separated sentences')
group.add_argument('--num-workers', type=int, default=2,
help="""Number of workers to use for dataloading""")
group.add_argument('--tokenizer-model-type', type=str,
default='bert-large-uncased',
help="Model type to use for sentencepiece tokenization \
(one of ['bpe', 'char', 'unigram', 'word']) or \
bert vocab to use for BertWordPieceTokenizer (one of \
['bert-large-uncased', 'bert-large-cased', etc.])")
group.add_argument('--tokenizer-path', type=str, default='tokenizer.model',
help='path used to save/load sentencepiece tokenization '
'models')
group.add_argument('--tokenizer-type', type=str,
default='BertWordPieceTokenizer',
choices=['CharacterLevelTokenizer',
'SentencePieceTokenizer',
'BertWordPieceTokenizer',
'GPT2BPETokenizer'],
help='what type of tokenizer to use')
group.add_argument("--cache-dir", default=None, type=str,
help="Where to store pre-trained BERT downloads")
group.add_argument('--use-tfrecords', action='store_true',
help='load `--train-data`, `--valid-data`, '
'`--test-data` from BERT tf records instead of '
'normal data pipeline')
group.add_argument('--seq-length', type=int, default=512,
help="Maximum sequence length to process")
group.add_argument('--max-preds-per-seq', type=int, default=None,
help='Maximum number of predictions to use per sequence.'
'Defaults to math.ceil(`--seq-length`*.15/10)*10.'
'MUST BE SPECIFIED IF `--use-tfrecords` is True.')
return parser
def get_args():
"""Parse all the args."""
parser = argparse.ArgumentParser(description='PyTorch BERT Model')
parser = add_model_config_args(parser)
parser = add_fp16_config_args(parser)
parser = add_training_args(parser)
parser = add_evaluation_args(parser)
parser = add_text_generate_args(parser)
parser = add_data_args(parser)
args = parser.parse_args()
if not args.train_data and not args.train_data_path:
print('WARNING: No training data specified')
args.cuda = torch.cuda.is_available()