Skip to content

GitLab

"...git@developer.sourcefind.cn:chenpangpang/transformers.git" did not exist on "fb056494e50be6ff729166734eff96f9f26d9e96"
Commit f8276008 authored by thomwolf's avatar thomwolf
Browse files

update readme, file names, removing TF code, moving tests

parent 3c24e4be
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 343 additions and 4413 deletions
Comparing TF and PT models.ipynb
View file @ f8276008
...@@ -6,15 +6,15 @@ ...@@ -6,15 +6,15 @@
"source": [ "source": [
"# Comparing TensorFlow (original) and PyTorch models\n", "# Comparing TensorFlow (original) and PyTorch models\n",
"\n", "\n",
"We use this small notebook to test the conversion of the model's weights and to make sure both the TensorFlow and PyTorch are coherent. In particular, we compare the weights of the last layer on a simple example (in `input.txt`).\n", "You can use this small notebook to check the conversion of the model's weights from the TensorFlow model to the PyTorch model. In the following, we compare the weights of the last layer on a simple example (in `input.txt`) but both models returns all the hidden layers so you can check every stage of the model.\n",
"\n", "\n",
"To run this notebook, please make sure that your Python environment has both TensorFlow and PyTorch.\n", "To run this notebook, follow these instructions:\n",
"You should follow the instructions in the `README.md` and make sure that you have:\n", "- make sure that your Python environment has both TensorFlow and PyTorch installed,\n",
"- the original TensorFlow implementation\n", "- download the original TensorFlow implementation,\n",
"- the `BERT-base, Uncased` model\n", "- download a pre-trained TensorFlow model as indicaded in the TensorFlow implementation readme,\n",
"- run the script `convert_tf_checkpoint_to_pytorch.py` to convert the weights to PyTorch\n", "- run the script `convert_tf_checkpoint_to_pytorch.py` as indicated in the `README` to convert the pre-trained TensorFlow model to PyTorch.\n",
"\n", "\n",
"Please modify the relative paths accordingly (at the beggining of Sections 1 and 2)." "If needed change the relative paths indicated in this notebook (at the beggining of Sections 1 and 2) to point to the relevent models and code."
] ]
}, },
{ {
...@@ -37,7 +37,7 @@ ...@@ -37,7 +37,7 @@
"bert_config_file = model_dir + \"bert_config.json\"\n", "bert_config_file = model_dir + \"bert_config.json\"\n",
"init_checkpoint = model_dir + \"bert_model.ckpt\"\n", "init_checkpoint = model_dir + \"bert_model.ckpt\"\n",
"\n", "\n",
"input_file = \"input.txt\"\n", "input_file = \"./samples/input.txt\"\n",
"max_seq_length = 128" "max_seq_length = 128"
] ]
}, },
...@@ -296,8 +296,8 @@ ...@@ -296,8 +296,8 @@
}, },
"outputs": [], "outputs": [],
"source": [ "source": [
"import extract_features_pytorch\n", "import extract_features\n",
"from extract_features_pytorch import *" "from extract_features import *"
] ]
}, },
{ {
...@@ -625,7 +625,7 @@ ...@@ -625,7 +625,7 @@
], ],
"source": [ "source": [
"device = torch.device(\"cpu\")\n", "device = torch.device(\"cpu\")\n",
"model = extract_features_pytorch.BertModel(bert_config)\n", "model = extract_features.BertModel(bert_config)\n",
"model.load_state_dict(torch.load(init_checkpoint_pt, map_location='cpu'))\n", "model.load_state_dict(torch.load(init_checkpoint_pt, map_location='cpu'))\n",
"model.to(device)" "model.to(device)"
] ]
...@@ -1196,7 +1196,7 @@ ...@@ -1196,7 +1196,7 @@
"metadata": { "metadata": {
"hide_input": false, "hide_input": false,
"kernelspec": { "kernelspec": {
"display_name": "Python 3", "display_name": "Python [default]",
"language": "python", "language": "python",
"name": "python3" "name": "python3"
}, },
...@@ -1210,7 +1210,7 @@ ...@@ -1210,7 +1210,7 @@
"name": "python", "name": "python",
"nbconvert_exporter": "python", "nbconvert_exporter": "python",
"pygments_lexer": "ipython3", "pygments_lexer": "ipython3",
"version": "3.6.5" "version": "3.6.7"
}, },
"toc": { "toc": {
"colors": { "colors": {
......
README.md
View file @ f8276008
# PyTorch implementation of Google AI's BERT # PyTorch implementation of Google AI's BERT
## Introduction ## Introduction
This is a PyTorch implementation of the [TensorFlow code](https://github.com/google-research/bert) released by Google AI with the paper [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805). This is a PyTorch implementation of the [TensorFlow code](https://github.com/google-research/bert) released by Google AI with the paper [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805).
It is op-for-op reimplementation that can load any pre-trained TensorFlow checkpoint in a PyTorch model (see below).
There are a few differences with the TensorFlow model:
- the PyTorch model has multi-GPU and distributed training capabilities (see below),
- there is not TPU support in the current stable version of PyTorch (0.4.1) and as a consequence, the pre-training script are not included in this repo. TPU support is supposed to be available in PyTorch v1.0 that will be released in the coming weeks. We will update the repository with TPU-adapted pre-training scripts when PyTorch will have TPU support. In the meantime, you can use the TensorFlow version to train a model on TPU and import the checkpoint using the following script.
## Converting a TensorFlow checkpoint (in particular Google's pre-trained models) to Pytorch
## Converting the TensorFlow pre-trained models to Pytorch You can convert any TensorFlow checkpoint, and in particular the pre-trained weights released by GoogleAI, by using `convert_tf_checkpoint_to_pytorch.py`.
You can convert the pre-trained weights released by GoogleAI by calling the script `convert_tf_checkpoint_to_pytorch.py`. This script takes as input a TensorFlow checkpoint (`bert_model.ckpt`) and converts it in a PyTorch dump as a `.bin` that can be imported using the usual `torch.load()` command.
It takes a TensorFlow checkpoint (`bert_model.ckpt`) containg the pre-trained weights and converts it to a `.bin` file readable for PyTorch.
TensorFlow pre-trained models can be found in the [original TensorFlow code](https://github.com/google-research/bert). We give an example with the `BERT-Base Uncased` model: TensorFlow pre-trained models can be found in the [original TensorFlow code](https://github.com/google-research/bert). Here give an example with the `BERT-Base Uncased` model:
```shell ```shell
export BERT_BASE_DIR=/path/to/bert/uncased_L-12_H-768_A-12 export BERT_BASE_DIR=/path/to/bert/uncased_L-12_H-768_A-12
export BERT_PYTORCH_DIR=/path/to/pytorch/bert/uncased_L-12_H-768_A-12
python convert_tf_checkpoint_to_pytorch.py \ python convert_tf_checkpoint_to_pytorch.py \
--tf_checkpoint_path=$BERT_BASE_DIR/bert_model.ckpt \ --tf_checkpoint_path=$BERT_BASE_DIR/bert_model.ckpt \
--bert_config_file=$BERT_BASE_DIR/bert_config.json \ --bert_config_file=$BERT_BASE_DIR/bert_config.json \
--pytorch_dump_path=$BERT_PYTORCH_DIR/pytorch_model.bin --pytorch_dump_path=$BERT_BASE_DIR/pytorch_model.bin
``` ```
## Fine-tuning with BERT: running the examples ## Fine-tuning with BERT: running the examples
We showcase the same examples as in the original implementation: fine-tuning on the MRPC classification corpus and the question answering dataset SQUAD. We showcase the same examples as in the original implementation: fine-tuning on the MRPC classification corpus and the question answering dataset SQUAD.
...@@ -40,7 +44,7 @@ Corpus (MRPC) corpus and runs in less than 10 minutes on a single K-80. ...@@ -40,7 +44,7 @@ Corpus (MRPC) corpus and runs in less than 10 minutes on a single K-80.
```shell ```shell
export GLUE_DIR=/path/to/glue export GLUE_DIR=/path/to/glue
python run_classifier_pytorch.py \ python run_classifier.py \
--task_name MRPC \ --task_name MRPC \
--do_train \ --do_train \
--do_eval \ --do_eval \
...@@ -53,21 +57,21 @@ python run_classifier_pytorch.py \ ...@@ -53,21 +57,21 @@ python run_classifier_pytorch.py \
--train_batch_size 32 \ --train_batch_size 32 \
--learning_rate 2e-5 \ --learning_rate 2e-5 \
--num_train_epochs 3.0 \ --num_train_epochs 3.0 \
--output_dir /tmp/mrpc_output_pytorch/ --output_dir /tmp/mrpc_output/
``` ```
The next example fine-tunes `BERT-Base` on the SQuAD question answering task. The next example fine-tunes `BERT-Base` on the SQuAD question answering task.
The data for SQuAD can be downloaded with the following links and should be saved in a `$SQUAD_DIR` directory. The data for SQuAD can be downloaded with the following links and should be saved in a `$SQUAD_DIR` directory.
* [train-v1.1.json](https://rajpurkar.github.io/SQuAD-explorer/dataset/train-v1.1.json) * [train-v1.1.json](https://rajpurkar.github.io/SQuAD-explorer/dataset/train-v1.1.json)
* [dev-v1.1.json](https://rajpurkar.github.io/SQuAD-explorer/dataset/dev-v1.1.json) * [dev-v1.1.json](https://rajpurkar.github.io/SQuAD-explorer/dataset/dev-v1.1.json)
* [evaluate-v1.1.py](https://github.com/allenai/bi-att-flow/blob/master/squad/evaluate-v1.1.py) * [evaluate-v1.1.py](https://github.com/allenai/bi-att-flow/blob/master/squad/evaluate-v1.1.py)
```shell ```shell
export SQUAD_DIR=/path/to/SQUAD export SQUAD_DIR=/path/to/SQUAD
python run_squad_pytorch.py \ python run_squad.py \
--vocab_file=$BERT_BASE_DIR/vocab.txt \ --vocab_file=$BERT_BASE_DIR/vocab.txt \
--bert_config_file=$BERT_BASE_DIR/bert_config.json \ --bert_config_file=$BERT_BASE_DIR/bert_config.json \
--init_checkpoint=$BERT_PYTORCH_DIR/pytorch_model.bin \ --init_checkpoint=$BERT_PYTORCH_DIR/pytorch_model.bin \
...@@ -83,13 +87,11 @@ python run_squad_pytorch.py \ ...@@ -83,13 +87,11 @@ python run_squad_pytorch.py \
--output_dir=../debug_squad/ --output_dir=../debug_squad/
``` ```
## Comparing TensorFlow and PyTorch models ## Comparing TensorFlow and PyTorch models
We also include [a small Notebook](https://github.com/huggingface/pytorch-pretrained-BERT/blob/master/Comparing%20TF%20and%20PT%20models.ipynb) we used to verify that the conversion of the weights to PyTorch are consistent with the original TensorFlow weights. We also include [a small Notebook](https://github.com/huggingface/pytorch-pretrained-BERT/blob/master/Comparing%20TF%20and%20PT%20models.ipynb) we used to verify that the conversion of the weights to PyTorch are consistent with the original TensorFlow weights.
Please follow the instructions in the Notebook to run it. Please follow the instructions in the Notebook to run it.
## Note on pre-training ## Note on pre-training
The original TensorFlow code also release two scripts for pre-training BERT: [create_pretraining_data.py](https://github.com/google-research/bert/blob/master/create_pretraining_data.py) and [run_pretraining.py](https://github.com/google-research/bert/blob/master/run_pretraining.py). The original TensorFlow code also release two scripts for pre-training BERT: [create_pretraining_data.py](https://github.com/google-research/bert/blob/master/create_pretraining_data.py) and [run_pretraining.py](https://github.com/google-research/bert/blob/master/run_pretraining.py).
...@@ -97,9 +99,15 @@ As the authors notice, pre-training BERT is particularly expensive and requires ...@@ -97,9 +99,15 @@ As the authors notice, pre-training BERT is particularly expensive and requires
We have decided **not** to port these scripts for now and wait for the TPU support on PyTorch (see the recent [official announcement](https://cloud.google.com/blog/products/ai-machine-learning/introducing-pytorch-across-google-cloud)). We have decided **not** to port these scripts for now and wait for the TPU support on PyTorch (see the recent [official announcement](https://cloud.google.com/blog/products/ai-machine-learning/introducing-pytorch-across-google-cloud)).
## Requirements ## Requirements
The main dependencies of this code are: The main dependencies of this code are:
- PyTorch (>= 0.4.0) - PyTorch (>= 0.4.0)
- tqdm - tqdm
\ No newline at end of file
To install the dependencies:
````bash
pip install -r ./requirements.txt
````
convert_tf_checkpoint_to_pytorch.py
View file @ f8276008
...@@ -11,7 +11,7 @@ import tensorflow as tf ...@@ -11,7 +11,7 @@ import tensorflow as tf
import torch import torch
import numpy as np import numpy as np
from modeling_pytorch import BertConfig, BertModel from modeling import BertConfig, BertModel
parser = argparse.ArgumentParser() parser = argparse.ArgumentParser()
......
convert_tf_checkpoint_to_pytorch_special_edition.py deleted 100644 → 0
View file @ 3c24e4be
# coding=utf-8
"""Convert BERT checkpoint."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import re
import argparse
import tensorflow as tf
import torch
import numpy as np
from modeling_pytorch import BertConfig, BertForSequenceClassification
parser = argparse.ArgumentParser()
## Required parameters
parser.add_argument("--tf_checkpoint_path",
default = None,
type = str,
required = True,
help = "Path the TensorFlow checkpoint path.")
parser.add_argument("--bert_config_file",
default = None,
type = str,
required = True,
help = "The config json file corresponding to the pre-trained BERT model. \n"
"This specifies the model architecture.")
parser.add_argument("--pytorch_dump_path",
default = None,
type = str,
required = True,
help = "Path to the output PyTorch model.")
args = parser.parse_args()
def convert():
# Initialise PyTorch model
config = BertConfig.from_json_file(args.bert_config_file)
model = BertForSequenceClassification(config, num_labels=2)
# Load weights from TF model
path = args.tf_checkpoint_path
print("Converting TensorFlow checkpoint from {}".format(path))
init_vars = tf.train.list_variables(path)
names = []
arrays = []
for name, shape in init_vars:
print("Loading {} with shape {}".format(name, shape))
array = tf.train.load_variable(path, name)
print("Numpy array shape {}".format(array.shape))
names.append(name)
arrays.append(array)
for name, array in zip(names, arrays):
# name = name[5:] # skip "bert/"
print("Loading {} or shape {}".format(name, array.shape))
name = name.split('/')
if name[0] in ['cls']:
if name[1] in ['predictions']:
print("Skipping")
continue
elif name[1] in ['seq_relationship']:
name = name[2:]
assert len(name) == 1
name[0] = name[0][7:]
pointer = model.classifier
else:
pointer = model
for m_name in name:
if re.fullmatch(r'[A-Za-z]+_\d+', m_name):
l = re.split(r'_(\d+)', m_name)
else:
l = [m_name]
if l[0] in ['kernel', 'weights']:
pointer = getattr(pointer, 'weight')
else:
pointer = getattr(pointer, l[0])
if len(l) >= 2:
num = int(l[1])
pointer = pointer[num]
if m_name[-11:] == '_embeddings':
pointer = getattr(pointer, 'weight')
elif m_name == 'kernel':
array = np.transpose(array)
try:
assert pointer.shape == array.shape
except AssertionError as e:
e.args += (pointer.shape, array.shape)
raise
pointer.data = torch.from_numpy(array)
# Save pytorch-model
torch.save(model.state_dict(), args.pytorch_dump_path)
if __name__ == "__main__":
convert()
extract_features_pytorch.py → extract_features.py
View file @ f8276008
...@@ -31,7 +31,7 @@ import torch ...@@ -31,7 +31,7 @@ import torch
from torch.utils.data import TensorDataset, DataLoader, RandomSampler, SequentialSampler from torch.utils.data import TensorDataset, DataLoader, RandomSampler, SequentialSampler
from torch.utils.data.distributed import DistributedSampler from torch.utils.data.distributed import DistributedSampler
from modeling_pytorch import BertConfig, BertModel from modeling import BertConfig, BertModel
logging.basicConfig(format = '%(asctime)s - %(levelname)s - %(name)s - %(message)s', logging.basicConfig(format = '%(asctime)s - %(levelname)s - %(name)s - %(message)s',
datefmt = '%m/%d/%Y %H:%M:%S', datefmt = '%m/%d/%Y %H:%M:%S',
......
modeling_test_pytorch.py deleted 100644 → 0
View file @ 3c24e4be
# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors.
#
# 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.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import collections
import json
import random
import re
import modeling_pytorch as modeling
import six
import unittest
import torch
class BertModelTest(unittest.TestCase):
class BertModelTester(object):
def __init__(self,
parent,
batch_size=13,
seq_length=7,
is_training=True,
use_input_mask=True,
use_token_type_ids=True,
vocab_size=99,
hidden_size=32,
num_hidden_layers=5,
num_attention_heads=4,
intermediate_size=37,
hidden_act="gelu",
hidden_dropout_prob=0.1,
attention_probs_dropout_prob=0.1,
max_position_embeddings=512,
type_vocab_size=16,
initializer_range=0.02,
scope=None):
self.parent = parent
self.batch_size = batch_size
self.seq_length = seq_length
self.is_training = is_training
self.use_input_mask = use_input_mask
self.use_token_type_ids = use_token_type_ids
self.vocab_size = vocab_size
self.hidden_size = hidden_size
self.num_hidden_layers = num_hidden_layers
self.num_attention_heads = num_attention_heads
self.intermediate_size = intermediate_size
self.hidden_act = hidden_act
self.hidden_dropout_prob = hidden_dropout_prob
self.attention_probs_dropout_prob = attention_probs_dropout_prob
self.max_position_embeddings = max_position_embeddings
self.type_vocab_size = type_vocab_size
self.initializer_range = initializer_range
self.scope = scope
def create_model(self):
input_ids = BertModelTest.ids_tensor([self.batch_size, self.seq_length], self.vocab_size)
input_mask = None
if self.use_input_mask:
input_mask = BertModelTest.ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
token_type_ids = None
if self.use_token_type_ids:
token_type_ids = BertModelTest.ids_tensor([self.batch_size, self.seq_length], self.type_vocab_size)
config = modeling.BertConfig(
vocab_size=self.vocab_size,
hidden_size=self.hidden_size,
num_hidden_layers=self.num_hidden_layers,
num_attention_heads=self.num_attention_heads,
intermediate_size=self.intermediate_size,
hidden_act=self.hidden_act,
hidden_dropout_prob=self.hidden_dropout_prob,
attention_probs_dropout_prob=self.attention_probs_dropout_prob,
max_position_embeddings=self.max_position_embeddings,
type_vocab_size=self.type_vocab_size,
initializer_range=self.initializer_range)
model = modeling.BertModel(config=config)
all_encoder_layers, pooled_output = model(input_ids, token_type_ids, input_mask)
outputs = {
"sequence_output": all_encoder_layers[-1],
"pooled_output": pooled_output,
"all_encoder_layers": all_encoder_layers,
}
return outputs
def check_output(self, result):
self.parent.assertListEqual(
list(result["sequence_output"].size()),
[self.batch_size, self.seq_length, self.hidden_size])
self.parent.assertListEqual(list(result["pooled_output"].size()), [self.batch_size, self.hidden_size])
def test_default(self):
self.run_tester(BertModelTest.BertModelTester(self))
def test_config_to_json_string(self):
config = modeling.BertConfig(vocab_size=99, hidden_size=37)
obj = json.loads(config.to_json_string())
self.assertEqual(obj["vocab_size"], 99)
self.assertEqual(obj["hidden_size"], 37)
def run_tester(self, tester):
output_result = tester.create_model()
tester.check_output(output_result)
# TODO Find PyTorch equivalent of assert_all_tensors_reachable() if necessary
# self.assert_all_tensors_reachable(sess, [init_op, ops])
@classmethod
def ids_tensor(cls, shape, vocab_size, rng=None, name=None):
"""Creates a random int32 tensor of the shape within the vocab size."""
if rng is None:
rng = random.Random()
total_dims = 1
for dim in shape:
total_dims *= dim
values = []
for _ in range(total_dims):
values.append(rng.randint(0, vocab_size - 1))
# TODO Solve : the returned tensors provoke index out of range errors when passed to the model
return torch.tensor(data=values, dtype=torch.int32)
def assert_all_tensors_reachable(self, sess, outputs):
"""Checks that all the tensors in the graph are reachable from outputs."""
graph = sess.graph
ignore_strings = [
"^.*/dilation_rate$",
"^.*/Tensordot/concat$",
"^.*/Tensordot/concat/axis$",
"^testing/.*$",
]
ignore_regexes = [re.compile(x) for x in ignore_strings]
unreachable = self.get_unreachable_ops(graph, outputs)
filtered_unreachable = []
for x in unreachable:
do_ignore = False
for r in ignore_regexes:
m = r.match(x.name)
if m is not None:
do_ignore = True
if do_ignore:
continue
filtered_unreachable.append(x)
unreachable = filtered_unreachable
self.assertEqual(
len(unreachable), 0, "The following ops are unreachable: %s" %
(" ".join([x.name for x in unreachable])))
@classmethod
def get_unreachable_ops(cls, graph, outputs):
"""Finds all of the tensors in graph that are unreachable from outputs."""
outputs = cls.flatten_recursive(outputs)
output_to_op = collections.defaultdict(list)
op_to_all = collections.defaultdict(list)
assign_out_to_in = collections.defaultdict(list)
for op in graph.get_operations():
for x in op.inputs:
op_to_all[op.name].append(x.name)
for y in op.outputs:
output_to_op[y.name].append(op.name)
op_to_all[op.name].append(y.name)
if str(op.type) == "Assign":
for y in op.outputs:
for x in op.inputs:
assign_out_to_in[y.name].append(x.name)
assign_groups = collections.defaultdict(list)
for out_name in assign_out_to_in.keys():
name_group = assign_out_to_in[out_name]
for n1 in name_group:
assign_groups[n1].append(out_name)
for n2 in name_group:
if n1 != n2:
assign_groups[n1].append(n2)
seen_tensors = {}
stack = [x.name for x in outputs]
while stack:
name = stack.pop()
if name in seen_tensors:
continue
seen_tensors[name] = True
if name in output_to_op:
for op_name in output_to_op[name]:
if op_name in op_to_all:
for input_name in op_to_all[op_name]:
if input_name not in stack:
stack.append(input_name)
expanded_names = []
if name in assign_groups:
for assign_name in assign_groups[name]:
expanded_names.append(assign_name)
for expanded_name in expanded_names:
if expanded_name not in stack:
stack.append(expanded_name)
unreachable_ops = []
for op in graph.get_operations():
is_unreachable = False
all_names = [x.name for x in op.inputs] + [x.name for x in op.outputs]
for name in all_names:
if name not in seen_tensors:
is_unreachable = True
if is_unreachable:
unreachable_ops.append(op)
return unreachable_ops
@classmethod
def flatten_recursive(cls, item):
"""Flattens (potentially nested) a tuple/dictionary/list to a list."""
output = []
if isinstance(item, list):
output.extend(item)
elif isinstance(item, tuple):
output.extend(list(item))
elif isinstance(item, dict):
for (_, v) in six.iteritems(item):
output.append(v)
else:
return [item]
flat_output = []
for x in output:
flat_output.extend(cls.flatten_recursive(x))
return flat_output
if __name__ == "__main__":
unittest.main()
run_classifier_pytorch.py → run_classifier.py
View file @ f8276008
...@@ -30,9 +30,9 @@ import torch ...@@ -30,9 +30,9 @@ import torch
from torch.utils.data import TensorDataset, DataLoader, RandomSampler, SequentialSampler from torch.utils.data import TensorDataset, DataLoader, RandomSampler, SequentialSampler
from torch.utils.data.distributed import DistributedSampler from torch.utils.data.distributed import DistributedSampler
import tokenization_pytorch import tokenization
from modeling_pytorch import BertConfig, BertForSequenceClassification from modeling import BertConfig, BertForSequenceClassification
from optimization_pytorch import BERTAdam from optimization import BERTAdam
logging.basicConfig(format = '%(asctime)s - %(levelname)s - %(name)s - %(message)s', logging.basicConfig(format = '%(asctime)s - %(levelname)s - %(name)s - %(message)s',
datefmt = '%m/%d/%Y %H:%M:%S', datefmt = '%m/%d/%Y %H:%M:%S',
...@@ -122,9 +122,9 @@ class MrpcProcessor(DataProcessor): ...@@ -122,9 +122,9 @@ class MrpcProcessor(DataProcessor):
if i == 0: if i == 0:
continue continue
guid = "%s-%s" % (set_type, i) guid = "%s-%s" % (set_type, i)
text_a = tokenization_pytorch.convert_to_unicode(line[3]) text_a = tokenization.convert_to_unicode(line[3])
text_b = tokenization_pytorch.convert_to_unicode(line[4]) text_b = tokenization.convert_to_unicode(line[4])
label = tokenization_pytorch.convert_to_unicode(line[0]) label = tokenization.convert_to_unicode(line[0])
examples.append( examples.append(
InputExample(guid=guid, text_a=text_a, text_b=text_b, label=label)) InputExample(guid=guid, text_a=text_a, text_b=text_b, label=label))
return examples return examples
...@@ -154,10 +154,10 @@ class MnliProcessor(DataProcessor): ...@@ -154,10 +154,10 @@ class MnliProcessor(DataProcessor):
for (i, line) in enumerate(lines): for (i, line) in enumerate(lines):
if i == 0: if i == 0:
continue continue
guid = "%s-%s" % (set_type, tokenization_pytorch.convert_to_unicode(line[0])) guid = "%s-%s" % (set_type, tokenization.convert_to_unicode(line[0]))
text_a = tokenization_pytorch.convert_to_unicode(line[8]) text_a = tokenization.convert_to_unicode(line[8])
text_b = tokenization_pytorch.convert_to_unicode(line[9]) text_b = tokenization.convert_to_unicode(line[9])
label = tokenization_pytorch.convert_to_unicode(line[-1]) label = tokenization.convert_to_unicode(line[-1])
examples.append( examples.append(
InputExample(guid=guid, text_a=text_a, text_b=text_b, label=label)) InputExample(guid=guid, text_a=text_a, text_b=text_b, label=label))
return examples return examples
...@@ -185,8 +185,8 @@ class ColaProcessor(DataProcessor): ...@@ -185,8 +185,8 @@ class ColaProcessor(DataProcessor):
examples = [] examples = []
for (i, line) in enumerate(lines): for (i, line) in enumerate(lines):
guid = "%s-%s" % (set_type, i) guid = "%s-%s" % (set_type, i)
text_a = tokenization_pytorch.convert_to_unicode(line[3]) text_a = tokenization.convert_to_unicode(line[3])
label = tokenization_pytorch.convert_to_unicode(line[1]) label = tokenization.convert_to_unicode(line[1])
examples.append( examples.append(
InputExample(guid=guid, text_a=text_a, text_b=None, label=label)) InputExample(guid=guid, text_a=text_a, text_b=None, label=label))
return examples return examples
...@@ -274,7 +274,7 @@ def convert_examples_to_features(examples, label_list, max_seq_length, ...@@ -274,7 +274,7 @@ def convert_examples_to_features(examples, label_list, max_seq_length,
logger.info("*** Example ***") logger.info("*** Example ***")
logger.info("guid: %s" % (example.guid)) logger.info("guid: %s" % (example.guid))
logger.info("tokens: %s" % " ".join( logger.info("tokens: %s" % " ".join(
[tokenization_pytorch.printable_text(x) for x in tokens])) [tokenization.printable_text(x) for x in tokens]))
logger.info("input_ids: %s" % " ".join([str(x) for x in input_ids])) logger.info("input_ids: %s" % " ".join([str(x) for x in input_ids]))
logger.info("input_mask: %s" % " ".join([str(x) for x in input_mask])) logger.info("input_mask: %s" % " ".join([str(x) for x in input_mask]))
logger.info( logger.info(
...@@ -477,7 +477,7 @@ def main(): ...@@ -477,7 +477,7 @@ def main():
label_list = processor.get_labels() label_list = processor.get_labels()
tokenizer = tokenization_pytorch.FullTokenizer( tokenizer = tokenization.FullTokenizer(
vocab_file=args.vocab_file, do_lower_case=args.do_lower_case) vocab_file=args.vocab_file, do_lower_case=args.do_lower_case)
train_examples = None train_examples = None
......
tensorflow_code/create_pretraining_data.py deleted 100644 → 0
View file @ 3c24e4be
# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors.
#
# 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.
"""Create masked LM/next sentence masked_lm TF examples for BERT."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import collections
import random
from tensorflow_code import tokenization
import tensorflow as tf
flags = tf.flags
FLAGS = flags.FLAGS
flags.DEFINE_string("input_file", None,
"Input raw text file (or comma-separated list of files).")
flags.DEFINE_string(
"output_file", None,
"Output TF example file (or comma-separated list of files).")
flags.DEFINE_string("vocab_file", None,
"The vocabulary file that the BERT model was trained on.")
flags.DEFINE_bool(
"do_lower_case", True,
"Whether to lower case the input text. Should be True for uncased "
"models and False for cased models.")
flags.DEFINE_integer("max_seq_length", 128, "Maximum sequence length.")
flags.DEFINE_integer("max_predictions_per_seq", 20,
"Maximum number of masked LM predictions per sequence.")
flags.DEFINE_integer("random_seed", 12345, "Random seed for data generation.")
flags.DEFINE_integer(
"dupe_factor", 10,
"Number of times to duplicate the input data (with different masks).")
flags.DEFINE_float("masked_lm_prob", 0.15, "Masked LM probability.")
flags.DEFINE_float(
"short_seq_prob", 0.1,
"Probability of creating sequences which are shorter than the "
"maximum length.")
class TrainingInstance(object):
"""A single training instance (sentence pair)."""
def __init__(self, tokens, segment_ids, masked_lm_positions, masked_lm_labels,
is_random_next):
self.tokens = tokens
self.segment_ids = segment_ids
self.is_random_next = is_random_next
self.masked_lm_positions = masked_lm_positions
self.masked_lm_labels = masked_lm_labels
def __str__(self):
s = ""
s += "tokens: %s\n" % (" ".join(
[tokenization.printable_text(x) for x in self.tokens]))
s += "segment_ids: %s\n" % (" ".join([str(x) for x in self.segment_ids]))
s += "is_random_next: %s\n" % self.is_random_next
s += "masked_lm_positions: %s\n" % (" ".join(
[str(x) for x in self.masked_lm_positions]))
s += "masked_lm_labels: %s\n" % (" ".join(
[tokenization.printable_text(x) for x in self.masked_lm_labels]))
s += "\n"
return s
def __repr__(self):
return self.__str__()
def write_instance_to_example_files(instances, tokenizer, max_seq_length,
max_predictions_per_seq, output_files):
"""Create TF example files from `TrainingInstance`s."""
writers = []
for output_file in output_files:
writers.append(tf.python_io.TFRecordWriter(output_file))
writer_index = 0
total_written = 0
for (inst_index, instance) in enumerate(instances):
input_ids = tokenizer.convert_tokens_to_ids(instance.tokens)
input_mask = [1] * len(input_ids)
segment_ids = list(instance.segment_ids)
assert len(input_ids) <= max_seq_length
while len(input_ids) < max_seq_length:
input_ids.append(0)
input_mask.append(0)
segment_ids.append(0)
assert len(input_ids) == max_seq_length
assert len(input_mask) == max_seq_length
assert len(segment_ids) == max_seq_length
masked_lm_positions = list(instance.masked_lm_positions)
masked_lm_ids = tokenizer.convert_tokens_to_ids(instance.masked_lm_labels)
masked_lm_weights = [1.0] * len(masked_lm_ids)
while len(masked_lm_positions) < max_predictions_per_seq:
masked_lm_positions.append(0)
masked_lm_ids.append(0)
masked_lm_weights.append(0.0)
next_sentence_label = 1 if instance.is_random_next else 0
features = collections.OrderedDict()
features["input_ids"] = create_int_feature(input_ids)
features["input_mask"] = create_int_feature(input_mask)
features["segment_ids"] = create_int_feature(segment_ids)
features["masked_lm_positions"] = create_int_feature(masked_lm_positions)
features["masked_lm_ids"] = create_int_feature(masked_lm_ids)
features["masked_lm_weights"] = create_float_feature(masked_lm_weights)
features["next_sentence_labels"] = create_int_feature([next_sentence_label])
tf_example = tf.train.Example(features=tf.train.Features(feature=features))
writers[writer_index].write(tf_example.SerializeToString())
writer_index = (writer_index + 1) % len(writers)
total_written += 1
if inst_index < 20:
tf.logging.info("*** Example ***")
tf.logging.info("tokens: %s" % " ".join(
[tokenization.printable_text(x) for x in instance.tokens]))
for feature_name in features.keys():
feature = features[feature_name]
values = []
if feature.int64_list.value:
values = feature.int64_list.value
elif feature.float_list.value:
values = feature.float_list.value
tf.logging.info(
"%s: %s" % (feature_name, " ".join([str(x) for x in values])))
for writer in writers:
writer.close()
tf.logging.info("Wrote %d total instances", total_written)
def create_int_feature(values):
feature = tf.train.Feature(int64_list=tf.train.Int64List(value=list(values)))
return feature
def create_float_feature(values):
feature = tf.train.Feature(float_list=tf.train.FloatList(value=list(values)))
return feature
def create_training_instances(input_files, tokenizer, max_seq_length,
dupe_factor, short_seq_prob, masked_lm_prob,
max_predictions_per_seq, rng):
"""Create `TrainingInstance`s from raw text."""
all_documents = [[]]
# Input file format:
# (1) One sentence per line. These should ideally be actual sentences, not
# entire paragraphs or arbitrary spans of text. (Because we use the
# sentence boundaries for the "next sentence prediction" task).
# (2) Blank lines between documents. Document boundaries are needed so
# that the "next sentence prediction" task doesn't span between documents.
for input_file in input_files:
with tf.gfile.GFile(input_file, "r") as reader:
while True:
line = tokenization.convert_to_unicode(reader.readline())
if not line:
break
line = line.strip()
# Empty lines are used as document delimiters
if not line:
all_documents.append([])
tokens = tokenizer.tokenize(line)
if tokens:
all_documents[-1].append(tokens)
# Remove empty documents
all_documents = [x for x in all_documents if x]
rng.shuffle(all_documents)
vocab_words = list(tokenizer.vocab.keys())
instances = []
for _ in range(dupe_factor):
for document_index in range(len(all_documents)):
instances.extend(
create_instances_from_document(
all_documents, document_index, max_seq_length, short_seq_prob,
masked_lm_prob, max_predictions_per_seq, vocab_words, rng))
rng.shuffle(instances)
return instances
def create_instances_from_document(
all_documents, document_index, max_seq_length, short_seq_prob,
masked_lm_prob, max_predictions_per_seq, vocab_words, rng):
"""Creates `TrainingInstance`s for a single document."""
document = all_documents[document_index]
# Account for [CLS], [SEP], [SEP]
max_num_tokens = max_seq_length - 3
# We *usually* want to fill up the entire sequence since we are padding
# to `max_seq_length` anyways, so short sequences are generally wasted
# computation. However, we *sometimes*
# (i.e., short_seq_prob == 0.1 == 10% of the time) want to use shorter
# sequences to minimize the mismatch between pre-training and fine-tuning.
# The `target_seq_length` is just a rough target however, whereas
# `max_seq_length` is a hard limit.
target_seq_length = max_num_tokens
if rng.random() < short_seq_prob:
target_seq_length = rng.randint(2, max_num_tokens)
# We DON'T just concatenate all of the tokens from a document into a long
# sequence and choose an arbitrary split point because this would make the
# next sentence prediction task too easy. Instead, we split the input into
# segments "A" and "B" based on the actual "sentences" provided by the user
# input.
instances = []
current_chunk = []
current_length = 0
i = 0
while i < len(document):
segment = document[i]
current_chunk.append(segment)
current_length += len(segment)
if i == len(document) - 1 or current_length >= target_seq_length:
if current_chunk:
# `a_end` is how many segments from `current_chunk` go into the `A`
# (first) sentence.
a_end = 1
if len(current_chunk) >= 2:
a_end = rng.randint(1, len(current_chunk) - 1)
tokens_a = []
for j in range(a_end):
tokens_a.extend(current_chunk[j])
tokens_b = []
# Random next
is_random_next = False
if len(current_chunk) == 1 or rng.random() < 0.5:
is_random_next = True
target_b_length = target_seq_length - len(tokens_a)
# This should rarely go for more than one iteration for large
# corpora. However, just to be careful, we try to make sure that
# the random document is not the same as the document
# we're processing.
for _ in range(10):
random_document_index = rng.randint(0, len(all_documents) - 1)
if random_document_index != document_index:
break
random_document = all_documents[random_document_index]
random_start = rng.randint(0, len(random_document) - 1)
for j in range(random_start, len(random_document)):
tokens_b.extend(random_document[j])
if len(tokens_b) >= target_b_length:
break
# We didn't actually use these segments so we "put them back" so
# they don't go to waste.
num_unused_segments = len(current_chunk) - a_end
i -= num_unused_segments
# Actual next
else:
is_random_next = False
for j in range(a_end, len(current_chunk)):
tokens_b.extend(current_chunk[j])
truncate_seq_pair(tokens_a, tokens_b, max_num_tokens, rng)
assert len(tokens_a) >= 1
assert len(tokens_b) >= 1
tokens = []
segment_ids = []
tokens.append("[CLS]")
segment_ids.append(0)
for token in tokens_a:
tokens.append(token)
segment_ids.append(0)
tokens.append("[SEP]")
segment_ids.append(0)
for token in tokens_b:
tokens.append(token)
segment_ids.append(1)
tokens.append("[SEP]")
segment_ids.append(1)
(tokens, masked_lm_positions,
masked_lm_labels) = create_masked_lm_predictions(
tokens, masked_lm_prob, max_predictions_per_seq, vocab_words, rng)
instance = TrainingInstance(
tokens=tokens,
segment_ids=segment_ids,
is_random_next=is_random_next,
masked_lm_positions=masked_lm_positions,
masked_lm_labels=masked_lm_labels)
instances.append(instance)
current_chunk = []
current_length = 0
i += 1
return instances
def create_masked_lm_predictions(tokens, masked_lm_prob,
max_predictions_per_seq, vocab_words, rng):
"""Creates the predictis for the masked LM objective."""
cand_indexes = []
for (i, token) in enumerate(tokens):
if token == "[CLS]" or token == "[SEP]":
continue
cand_indexes.append(i)
rng.shuffle(cand_indexes)
output_tokens = list(tokens)
masked_lm = collections.namedtuple("masked_lm", ["index", "label"]) # pylint: disable=invalid-name
num_to_predict = min(max_predictions_per_seq,
max(1, int(round(len(tokens) * masked_lm_prob))))
masked_lms = []
covered_indexes = set()
for index in cand_indexes:
if len(masked_lms) >= num_to_predict:
break
if index in covered_indexes:
continue
covered_indexes.add(index)
masked_token = None
# 80% of the time, replace with [MASK]
if rng.random() < 0.8:
masked_token = "[MASK]"
else:
# 10% of the time, keep original
if rng.random() < 0.5:
masked_token = tokens[index]
# 10% of the time, replace with random word
else:
masked_token = vocab_words[rng.randint(0, len(vocab_words) - 1)]
output_tokens[index] = masked_token
masked_lms.append(masked_lm(index=index, label=tokens[index]))
masked_lms = sorted(masked_lms, key=lambda x: x.index)
masked_lm_positions = []
masked_lm_labels = []
for p in masked_lms:
masked_lm_positions.append(p.index)
masked_lm_labels.append(p.label)
return (output_tokens, masked_lm_positions, masked_lm_labels)
def truncate_seq_pair(tokens_a, tokens_b, max_num_tokens, rng):
"""Truncates a pair of sequences to a maximum sequence length."""
while True:
total_length = len(tokens_a) + len(tokens_b)
if total_length <= max_num_tokens:
break
trunc_tokens = tokens_a if len(tokens_a) > len(tokens_b) else tokens_b
assert len(trunc_tokens) >= 1
# We want to sometimes truncate from the front and sometimes from the
# back to add more randomness and avoid biases.
if rng.random() < 0.5:
del trunc_tokens[0]
else:
trunc_tokens.pop()
def main(_):
tf.logging.set_verbosity(tf.logging.INFO)
tokenizer = tokenization.FullTokenizer(
vocab_file=FLAGS.vocab_file, do_lower_case=FLAGS.do_lower_case)
input_files = []
for input_pattern in FLAGS.input_file.split(","):
input_files.extend(tf.gfile.Glob(input_pattern))
tf.logging.info("*** Reading from input files ***")
for input_file in input_files:
tf.logging.info(" %s", input_file)
rng = random.Random(FLAGS.random_seed)
instances = create_training_instances(
input_files, tokenizer, FLAGS.max_seq_length, FLAGS.dupe_factor,
FLAGS.short_seq_prob, FLAGS.masked_lm_prob, FLAGS.max_predictions_per_seq,
rng)
output_files = FLAGS.output_file.split(",")
tf.logging.info("*** Writing to output files ***")
for output_file in output_files:
tf.logging.info(" %s", output_file)
write_instance_to_example_files(instances, tokenizer, FLAGS.max_seq_length,
FLAGS.max_predictions_per_seq, output_files)
if __name__ == "__main__":
flags.mark_flag_as_required("input_file")
flags.mark_flag_as_required("output_file")
flags.mark_flag_as_required("vocab_file")
tf.app.run()
tensorflow_code/extract_features.py deleted 100644 → 0
View file @ 3c24e4be
# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors.
#
# 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.
"""Extract pre-computed feature vectors from BERT."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import codecs
import collections
import json
import re
from tensorflow_code import modeling
from tensorflow_code import tokenization
import tensorflow as tf
flags = tf.flags
FLAGS = flags.FLAGS
flags.DEFINE_string("input_file", None, "")
flags.DEFINE_string("output_file", None, "")
flags.DEFINE_string("layers", "-1,-2,-3,-4", "")
flags.DEFINE_string(
"bert_config_file", None,
"The config json file corresponding to the pre-trained BERT model. "
"This specifies the model architecture.")
flags.DEFINE_integer(
"max_seq_length", 128,
"The maximum total input sequence length after WordPiece tokenization. "
"Sequences longer than this will be truncated, and sequences shorter "
"than this will be padded.")
flags.DEFINE_string(
"init_checkpoint", None,
"Initial checkpoint (usually from a pre-trained BERT model).")
flags.DEFINE_string("vocab_file", None,
"The vocabulary file that the BERT model was trained on.")
flags.DEFINE_bool(
"do_lower_case", True,
"Whethre to lower case the input text. Should be True for uncased "
"models and False for cased models.")
flags.DEFINE_integer("batch_size", 32, "Batch size for predictions.")
flags.DEFINE_bool("use_tpu", False, "Whether to use TPU or GPU/CPU.")
flags.DEFINE_string("master", None,
"If using a TPU, the address of the master.")
flags.DEFINE_integer(
"num_tpu_cores", 8,
"Only used if `use_tpu` is True. Total number of TPU cores to use.")
flags.DEFINE_bool(
"use_one_hot_embeddings", False,
"If True, tf.one_hot will be used for embedding lookups, otherwise "
"tf.nn.embedding_lookup will be used. On TPUs, this should be True "
"since it is much faster.")
class InputExample(object):
def __init__(self, unique_id, text_a, text_b):
self.unique_id = unique_id
self.text_a = text_a
self.text_b = text_b
class InputFeatures(object):
"""A single set of features of data."""
def __init__(self, unique_id, tokens, input_ids, input_mask, input_type_ids):
self.unique_id = unique_id
self.tokens = tokens
self.input_ids = input_ids
self.input_mask = input_mask
self.input_type_ids = input_type_ids
def input_fn_builder(features, seq_length):
"""Creates an `input_fn` closure to be passed to TPUEstimator."""
all_unique_ids = []
all_input_ids = []
all_input_mask = []
all_input_type_ids = []
for feature in features:
all_unique_ids.append(feature.unique_id)
all_input_ids.append(feature.input_ids)
all_input_mask.append(feature.input_mask)
all_input_type_ids.append(feature.input_type_ids)
def input_fn(params):
"""The actual input function."""
batch_size = params["batch_size"]
num_examples = len(features)
# This is for demo purposes and does NOT scale to large data sets. We do
# not use Dataset.from_generator() because that uses tf.py_func which is
# not TPU compatible. The right way to load data is with TFRecordReader.
d = tf.data.Dataset.from_tensor_slices({
"unique_ids":
tf.constant(all_unique_ids, shape=[num_examples], dtype=tf.int32),
"input_ids":
tf.constant(
all_input_ids, shape=[num_examples, seq_length],
dtype=tf.int32),
"input_mask":
tf.constant(
all_input_mask,
shape=[num_examples, seq_length],
dtype=tf.int32),
"input_type_ids":
tf.constant(
all_input_type_ids,
shape=[num_examples, seq_length],
dtype=tf.int32),
})
d = d.batch(batch_size=batch_size, drop_remainder=False)
return d
return input_fn
def model_fn_builder(bert_config, init_checkpoint, layer_indexes, use_tpu,
use_one_hot_embeddings):
"""Returns `model_fn` closure for TPUEstimator."""
def model_fn(features, labels, mode, params): # pylint: disable=unused-argument
"""The `model_fn` for TPUEstimator."""
unique_ids = features["unique_ids"]
input_ids = features["input_ids"]
input_mask = features["input_mask"]
input_type_ids = features["input_type_ids"]
model = modeling.BertModel(
config=bert_config,
is_training=False,
input_ids=input_ids,
input_mask=input_mask,
token_type_ids=input_type_ids,
use_one_hot_embeddings=use_one_hot_embeddings)
if mode != tf.estimator.ModeKeys.PREDICT:
raise ValueError("Only PREDICT modes are supported: %s" % (mode))
tvars = tf.trainable_variables()
scaffold_fn = None
(assignment_map, _) = modeling.get_assigment_map_from_checkpoint(
tvars, init_checkpoint)
if use_tpu:
def tpu_scaffold():
tf.train.init_from_checkpoint(init_checkpoint, assignment_map)
return tf.train.Scaffold()
scaffold_fn = tpu_scaffold
else:
tf.train.init_from_checkpoint(init_checkpoint, assignment_map)
all_layers = model.get_all_encoder_layers()
predictions = {
"unique_id": unique_ids,
}
for (i, layer_index) in enumerate(layer_indexes):
predictions["layer_output_%d" % i] = all_layers[layer_index]
output_spec = tf.contrib.tpu.TPUEstimatorSpec(
mode=mode, predictions=predictions, scaffold_fn=scaffold_fn)
return output_spec
return model_fn
def convert_examples_to_features(examples, seq_length, tokenizer):
"""Loads a data file into a list of `InputBatch`s."""
features = []
for (ex_index, example) in enumerate(examples):
tokens_a = tokenizer.tokenize(example.text_a)
tokens_b = None
if example.text_b:
tokens_b = tokenizer.tokenize(example.text_b)
if tokens_b:
# Modifies `tokens_a` and `tokens_b` in place so that the total
# length is less than the specified length.
# Account for [CLS], [SEP], [SEP] with "- 3"
_truncate_seq_pair(tokens_a, tokens_b, seq_length - 3)
else:
# Account for [CLS] and [SEP] with "- 2"
if len(tokens_a) > seq_length - 2:
tokens_a = tokens_a[0:(seq_length - 2)]
# The convention in BERT is:
# (a) For sequence pairs:
# tokens: [CLS] is this jack ##son ##ville ? [SEP] no it is not . [SEP]
# type_ids: 0 0 0 0 0 0 0 0 1 1 1 1 1 1
# (b) For single sequences:
# tokens: [CLS] the dog is hairy . [SEP]
# type_ids: 0 0 0 0 0 0 0
#
# Where "type_ids" are used to indicate whether this is the first
# sequence or the second sequence. The embedding vectors for `type=0` and
# `type=1` were learned during pre-training and are added to the wordpiece
# embedding vector (and position vector). This is not *strictly* necessary
# since the [SEP] token unambigiously separates the sequences, but it makes
# it easier for the model to learn the concept of sequences.
#
# For classification tasks, the first vector (corresponding to [CLS]) is
# used as as the "sentence vector". Note that this only makes sense because
# the entire model is fine-tuned.
tokens = []
input_type_ids = []
tokens.append("[CLS]")
input_type_ids.append(0)
for token in tokens_a:
tokens.append(token)
input_type_ids.append(0)
tokens.append("[SEP]")
input_type_ids.append(0)
if tokens_b:
for token in tokens_b:
tokens.append(token)
input_type_ids.append(1)
tokens.append("[SEP]")
input_type_ids.append(1)
input_ids = tokenizer.convert_tokens_to_ids(tokens)
# The mask has 1 for real tokens and 0 for padding tokens. Only real
# tokens are attended to.
input_mask = [1] * len(input_ids)
# Zero-pad up to the sequence length.
while len(input_ids) < seq_length:
input_ids.append(0)
input_mask.append(0)
input_type_ids.append(0)
assert len(input_ids) == seq_length
assert len(input_mask) == seq_length
assert len(input_type_ids) == seq_length
if ex_index < 5:
tf.logging.info("*** Example ***")
tf.logging.info("unique_id: %s" % (example.unique_id))
tf.logging.info("tokens: %s" % " ".join([str(x) for x in tokens]))
tf.logging.info("input_ids: %s" % " ".join([str(x) for x in input_ids]))
tf.logging.info("input_mask: %s" % " ".join([str(x) for x in input_mask]))
tf.logging.info(
"input_type_ids: %s" % " ".join([str(x) for x in input_type_ids]))
features.append(
InputFeatures(
unique_id=example.unique_id,
tokens=tokens,
input_ids=input_ids,
input_mask=input_mask,
input_type_ids=input_type_ids))
return features
def _truncate_seq_pair(tokens_a, tokens_b, max_length):
"""Truncates a sequence pair in place to the maximum length."""
# This is a simple heuristic which will always truncate the longer sequence
# one token at a time. This makes more sense than truncating an equal percent
# of tokens from each, since if one sequence is very short then each token
# that's truncated likely contains more information than a longer sequence.
while True:
total_length = len(tokens_a) + len(tokens_b)
if total_length <= max_length:
break
if len(tokens_a) > len(tokens_b):
tokens_a.pop()
else:
tokens_b.pop()
def read_examples(input_file):
"""Read a list of `InputExample`s from an input file."""
examples = []
unique_id = 0
with tf.gfile.GFile(input_file, "r") as reader:
while True:
line = tokenization.convert_to_unicode(reader.readline())
if not line:
break
line = line.strip()
text_a = None
text_b = None
m = re.match(r"^(.*) \|\|\| (.*)$", line)
if m is None:
text_a = line
else:
text_a = m.group(1)
text_b = m.group(2)
examples.append(
InputExample(unique_id=unique_id, text_a=text_a, text_b=text_b))
unique_id += 1
return examples
def main(_):
tf.logging.set_verbosity(tf.logging.INFO)
layer_indexes = [int(x) for x in FLAGS.layers.split(",")]
bert_config = modeling.BertConfig.from_json_file(FLAGS.bert_config_file)
tokenizer = tokenization.FullTokenizer(
vocab_file=FLAGS.vocab_file, do_lower_case=FLAGS.do_lower_case)
is_per_host = tf.contrib.tpu.InputPipelineConfig.PER_HOST_V2
run_config = tf.contrib.tpu.RunConfig(
master=FLAGS.master,
tpu_config=tf.contrib.tpu.TPUConfig(
num_shards=FLAGS.num_tpu_cores,
per_host_input_for_training=is_per_host))
examples = read_examples(FLAGS.input_file)
features = convert_examples_to_features(
examples=examples, seq_length=FLAGS.max_seq_length, tokenizer=tokenizer)
unique_id_to_feature = {}
for feature in features:
unique_id_to_feature[feature.unique_id] = feature
model_fn = model_fn_builder(
bert_config=bert_config,
init_checkpoint=FLAGS.init_checkpoint,
layer_indexes=layer_indexes,
use_tpu=FLAGS.use_tpu,
use_one_hot_embeddings=FLAGS.use_one_hot_embeddings)
# If TPU is not available, this will fall back to normal Estimator on CPU
# or GPU.
estimator = tf.contrib.tpu.TPUEstimator(
use_tpu=FLAGS.use_tpu,
model_fn=model_fn,
config=run_config,
predict_batch_size=FLAGS.batch_size)
input_fn = input_fn_builder(
features=features, seq_length=FLAGS.max_seq_length)
with codecs.getwriter("utf-8")(tf.gfile.Open(FLAGS.output_file,
"w")) as writer:
for result in estimator.predict(input_fn, yield_single_examples=True):
unique_id = int(result["unique_id"])
feature = unique_id_to_feature[unique_id]
output_json = collections.OrderedDict()
output_json["linex_index"] = unique_id
all_features = []
for (i, token) in enumerate(feature.tokens):
all_layers = []
for (j, layer_index) in enumerate(layer_indexes):
layer_output = result["layer_output_%d" % j]
layers = collections.OrderedDict()
layers["index"] = layer_index
layers["values"] = [
round(float(x), 6) for x in layer_output[i:(i + 1)].flat
]
all_layers.append(layers)
features = collections.OrderedDict()
features["token"] = token
features["layers"] = all_layers
all_features.append(features)
output_json["features"] = all_features
writer.write(json.dumps(output_json) + "\n")
if __name__ == "__main__":
flags.mark_flag_as_required("input_file")
flags.mark_flag_as_required("vocab_file")
flags.mark_flag_as_required("bert_config_file")
flags.mark_flag_as_required("init_checkpoint")
flags.mark_flag_as_required("output_file")
tf.app.run()
tensorflow_code/modeling.py deleted 100644 → 0
View file @ 3c24e4be
This diff is collapsed.
tensorflow_code/optimization.py deleted 100644 → 0
View file @ 3c24e4be
# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors.
#
# 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.
"""Functions and classes related to optimization (weight updates)."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import re
import tensorflow as tf
def create_optimizer(loss, init_lr, num_train_steps, num_warmup_steps, use_tpu):
"""Creates an optimizer training op."""
global_step = tf.train.get_or_create_global_step()
learning_rate = tf.constant(value=init_lr, shape=[], dtype=tf.float32)
# Implements linear decay of the learning rate.
learning_rate = tf.train.polynomial_decay(
learning_rate,
global_step,
num_train_steps,
end_learning_rate=0.0,
power=1.0,
cycle=False)
# Implements linear warmup. I.e., if global_step < num_warmup_steps, the
# learning rate will be `global_step/num_warmup_steps * init_lr`.
if num_warmup_steps:
global_steps_int = tf.cast(global_step, tf.int32)
warmup_steps_int = tf.constant(num_warmup_steps, dtype=tf.int32)
global_steps_float = tf.cast(global_steps_int, tf.float32)
warmup_steps_float = tf.cast(warmup_steps_int, tf.float32)
warmup_percent_done = global_steps_float / warmup_steps_float
warmup_learning_rate = init_lr * warmup_percent_done
is_warmup = tf.cast(global_steps_int < warmup_steps_int, tf.float32)
learning_rate = (
(1.0 - is_warmup) * learning_rate + is_warmup * warmup_learning_rate)
# It is recommended that you use this optimizer for fine tuning, since this
# is how the model was trained (note that the Adam m/v variables are NOT
# loaded from init_checkpoint.)
optimizer = AdamWeightDecayOptimizer(
learning_rate=learning_rate,
weight_decay_rate=0.01,
beta_1=0.9,
beta_2=0.999,
epsilon=1e-6,
exclude_from_weight_decay=["LayerNorm", "layer_norm", "bias"])
if use_tpu:
optimizer = tf.contrib.tpu.CrossShardOptimizer(optimizer)
tvars = tf.trainable_variables()
grads = tf.gradients(loss, tvars)
# This is how the model was pre-trained.
(grads, _) = tf.clip_by_global_norm(grads, clip_norm=1.0)
train_op = optimizer.apply_gradients(
zip(grads, tvars), global_step=global_step)
new_global_step = global_step + 1
train_op = tf.group(train_op, [global_step.assign(new_global_step)])
return train_op
class AdamWeightDecayOptimizer(tf.train.Optimizer):
"""A basic Adam optimizer that includes "correct" L2 weight decay."""
def __init__(self,
learning_rate,
weight_decay_rate=0.0,
beta_1=0.9,
beta_2=0.999,
epsilon=1e-6,
exclude_from_weight_decay=None,
name="AdamWeightDecayOptimizer"):
"""Constructs a AdamWeightDecayOptimizer."""
super(AdamWeightDecayOptimizer, self).__init__(False, name)
self.learning_rate = learning_rate
self.weight_decay_rate = weight_decay_rate
self.beta_1 = beta_1
self.beta_2 = beta_2
self.epsilon = epsilon
self.exclude_from_weight_decay = exclude_from_weight_decay
def apply_gradients(self, grads_and_vars, global_step=None, name=None):
"""See base class."""
assignments = []
for (grad, param) in grads_and_vars:
if grad is None or param is None:
continue
param_name = self._get_variable_name(param.name)
m = tf.get_variable(
name=param_name + "/adam_m",
shape=param.shape.as_list(),
dtype=tf.float32,
trainable=False,
initializer=tf.zeros_initializer())
v = tf.get_variable(
name=param_name + "/adam_v",
shape=param.shape.as_list(),
dtype=tf.float32,
trainable=False,
initializer=tf.zeros_initializer())
# Standard Adam update.
next_m = (
tf.multiply(self.beta_1, m) + tf.multiply(1.0 - self.beta_1, grad))
next_v = (
tf.multiply(self.beta_2, v) + tf.multiply(1.0 - self.beta_2,
tf.square(grad)))
update = next_m / (tf.sqrt(next_v) + self.epsilon)
# Just adding the square of the weights to the loss function is *not*
# the correct way of using L2 regularization/weight decay with Adam,
# since that will interact with the m and v parameters in strange ways.
#
# Instead we want ot decay the weights in a manner that doesn't interact
# with the m/v parameters. This is equivalent to adding the square
# of the weights to the loss with plain (non-momentum) SGD.
if self._do_use_weight_decay(param_name):
update += self.weight_decay_rate * param
update_with_lr = self.learning_rate * update
next_param = param - update_with_lr
assignments.extend(
[param.assign(next_param),
m.assign(next_m),
v.assign(next_v)])
return tf.group(*assignments, name=name)
def _do_use_weight_decay(self, param_name):
"""Whether to use L2 weight decay for `param_name`."""
if not self.weight_decay_rate:
return False
if self.exclude_from_weight_decay:
for r in self.exclude_from_weight_decay:
if re.search(r, param_name) is not None:
return False
return True
def _get_variable_name(self, param_name):
"""Get the variable name from the tensor name."""
m = re.match("^(.*):\\d+$", param_name)
if m is not None:
param_name = m.group(1)
return param_name
tensorflow_code/optimization_test.py deleted 100644 → 0
View file @ 3c24e4be
# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors.
#
# 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.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow_code import optimization
import tensorflow as tf
class OptimizationTest(tf.test.TestCase):
def test_adam(self):
with self.test_session() as sess:
w = tf.get_variable(
"w",
shape=[3],
initializer=tf.constant_initializer([0.1, -0.2, -0.1]))
x = tf.constant([0.4, 0.2, -0.5])
loss = tf.reduce_mean(tf.square(x - w))
tvars = tf.trainable_variables()
grads = tf.gradients(loss, tvars)
global_step = tf.train.get_or_create_global_step()
optimizer = optimization.AdamWeightDecayOptimizer(learning_rate=0.2)
train_op = optimizer.apply_gradients(zip(grads, tvars), global_step)
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
sess.run(init_op)
np_w = sess.run(w)
np_loss = sess.run(loss)
np_grad = sess.run(grads)[0]
for i in range(100):
print(i)
sess.run(train_op)
np_w = sess.run(w)
np_loss = sess.run(loss)
np_grad = sess.run(grads)[0]
self.assertAllClose(np_w.flat, [0.4, 0.2, -0.5], rtol=1e-2, atol=1e-2)
if __name__ == "__main__":
tf.test.main()
tensorflow_code/run_classifier.py deleted 100644 → 0
View file @ 3c24e4be
This diff is collapsed.
tensorflow_code/run_pretraining.py deleted 100644 → 0
View file @ 3c24e4be
This diff is collapsed.
tensorflow_code/tokenization.py deleted 100644 → 0
View file @ 3c24e4be
This diff is collapsed.
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment