README.md 30.4 KB
Newer Older
thomwolf's avatar
thomwolf committed
1
2
<p align="center">
    <br>
thomwolf's avatar
thomwolf committed
3
    <img src="https://raw.githubusercontent.com/huggingface/transformers/master/docs/source/imgs/transformers_logo_name.png" width="400"/>
thomwolf's avatar
thomwolf committed
4
5
6
    <br>
<p>
<p align="center">
Lysandre Debut's avatar
Lysandre Debut committed
7
    <a href="https://circleci.com/gh/huggingface/transformers">
thomwolf's avatar
thomwolf committed
8
        <img alt="Build" src="https://img.shields.io/circleci/build/github/huggingface/transformers/master">
thomwolf's avatar
thomwolf committed
9
10
    </a>
    <a href="https://github.com/huggingface/transformers/blob/master/LICENSE">
thomwolf's avatar
thomwolf committed
11
        <img alt="GitHub" src="https://img.shields.io/github/license/huggingface/transformers.svg?color=blue">
thomwolf's avatar
thomwolf committed
12
13
    </a>
    <a href="https://huggingface.co/transformers/index.html">
thomwolf's avatar
thomwolf committed
14
        <img alt="Documentation" src="https://img.shields.io/website/http/huggingface.co/transformers/index.html.svg?down_color=red&down_message=offline&up_message=online">
thomwolf's avatar
thomwolf committed
15
16
    </a>
    <a href="https://github.com/huggingface/transformers/releases">
thomwolf's avatar
thomwolf committed
17
        <img alt="GitHub release" src="https://img.shields.io/github/release/huggingface/transformers.svg">
thomwolf's avatar
thomwolf committed
18
19
20
    </a>
</p>

thomwolf's avatar
thomwolf committed
21
22
23
<h3 align="center">
<p>State-of-the-art Natural Language Processing for TensorFlow 2.0 and PyTorch
</h3>
thomwolf's avatar
thomwolf committed
24

keskarnitish's avatar
keskarnitish committed
25
🤗 Transformers (formerly known as `pytorch-transformers` and `pytorch-pretrained-bert`) provides state-of-the-art general-purpose architectures (BERT, GPT-2, RoBERTa, XLM, DistilBert, XLNet, CTRL...) for Natural Language Understanding (NLU) and Natural Language Generation (NLG) with over 32+ pretrained models in 100+ languages and deep interoperability between TensorFlow 2.0 and PyTorch.
thomwolf's avatar
thomwolf committed
26
27

### Features
thomwolf's avatar
thomwolf committed
28
29
30
31
32
33
34
35
36
37
38
39
40
41

- As easy to use as pytorch-transformers
- As powerful and concise as Keras
- High performance on NLU and NLG tasks
- Low barrier to entry for educators and practitioners

State-of-the-art NLP for everyone
- Deep learning researchers
- Hands-on practitioners
- AI/ML/NLP teachers and educators

Lower compute costs, smaller carbon footprint
- Researchers can share trained models instead of always retraining
- Practitioners can reduce compute time and production costs
42
- 10 architectures with over 30 pretrained models, some in more than 100 languages
thomwolf's avatar
thomwolf committed
43
44
45

Choose the right framework for every part of a model's lifetime
- Train state-of-the-art models in 3 lines of code
thomwolf's avatar
thomwolf committed
46
47
- Deep interoperability between TensorFlow 2.0 and PyTorch models
- Move a single model between TF2.0/PyTorch frameworks at will
thomwolf's avatar
thomwolf committed
48
- Seamlessly pick the right framework for training, evaluation, production
Julien Chaumond's avatar
Julien Chaumond committed
49

thomwolf's avatar
indeed  
thomwolf committed
50

thomwolf's avatar
thomwolf committed
51
52
53
| Section | Description |
|-|-|
| [Installation](#installation) | How to install the package |
thomwolf's avatar
thomwolf committed
54
| [Model architectures](#model-architectures) | Architectures (with pretrained weights) |
thomwolf's avatar
thomwolf committed
55
56
| [Online demo](#online-demo) | Experimenting with this repo’s text generation capabilities |
| [Quick tour: Usage](#quick-tour) | Tokenizers & models usage: Bert and GPT-2 |
wangfei's avatar
wangfei committed
57
| [Quick tour: TF 2.0 and PyTorch ](#Quick-tour-TF-20-training-and-PyTorch-interoperability) | Train a TF 2.0 model in 10 lines of code, load it in PyTorch |
thomwolf's avatar
thomwolf committed
58
| [Quick tour: Fine-tuning/usage scripts](#quick-tour-of-the-fine-tuningusage-scripts) | Using provided scripts: GLUE, SQuAD and Text generation |
59
| [Migrating from pytorch-transformers to transformers](#Migrating-from-pytorch-transformers-to-transformers) | Migrating your code from pytorch-transformers to transformers |
thomwolf's avatar
thomwolf committed
60
| [Migrating from pytorch-pretrained-bert to pytorch-transformers](#Migrating-from-pytorch-pretrained-bert-to-transformers) | Migrating your code from pytorch-pretrained-bert to transformers |
61
| [Documentation](https://huggingface.co/transformers/) [(v2.1.1)](https://huggingface.co/transformers/v2.1.1) [(v2.0.0)](https://huggingface.co/transformers/v2.0.0) [(v1.2.0)](https://huggingface.co/transformers/v1.2.0) [(v1.1.0)](https://huggingface.co/transformers/v1.1.0) [(v1.0.0)](https://huggingface.co/transformers/v1.0.0) | Full API documentation and more |
thomwolf's avatar
thomwolf committed
62

thomwolf's avatar
thomwolf committed
63
## Installation
VictorSanh's avatar
VictorSanh committed
64

65
This repo is tested on Python 2.7 and 3.5+ (examples are tested only on python 3.5+), PyTorch 1.0.0+ and TensorFlow 2.0.0-rc1
VictorSanh's avatar
VictorSanh committed
66

thomwolf's avatar
thomwolf committed
67
### With pip
thomwolf's avatar
thomwolf committed
68

69
First you need to install one of, or both, TensorFlow 2.0 and PyTorch.
Christopher Goh's avatar
Christopher Goh committed
70
Please refer to [TensorFlow installation page](https://www.tensorflow.org/install/pip#tensorflow-2.0-rc-is-available) and/or [PyTorch installation page](https://pytorch.org/get-started/locally/#start-locally) regarding the specific install command for your platform.
71
72

When TensorFlow 2.0 and/or PyTorch has been installed, 🤗 Transformers can be installed using pip as follows:
thomwolf's avatar
thomwolf committed
73

thomwolf's avatar
thomwolf committed
74
```bash
75
pip install transformers
thomwolf's avatar
thomwolf committed
76
```
VictorSanh's avatar
VictorSanh committed
77

thomwolf's avatar
thomwolf committed
78
### From source
thomwolf's avatar
thomwolf committed
79

80
Here also, you first need to install one of, or both, TensorFlow 2.0 and PyTorch.
Christopher Goh's avatar
Christopher Goh committed
81
Please refer to [TensorFlow installation page](https://www.tensorflow.org/install/pip#tensorflow-2.0-rc-is-available) and/or [PyTorch installation page](https://pytorch.org/get-started/locally/#start-locally) regarding the specific install command for your platform.
82

DenysNahurnyi's avatar
DenysNahurnyi committed
83
When TensorFlow 2.0 and/or PyTorch has been installed, you can install from source by cloning the repository and running:
thomwolf's avatar
thomwolf committed
84

thomwolf's avatar
thomwolf committed
85
86
87
```bash
pip install [--editable] .
```
VictorSanh's avatar
VictorSanh committed
88

thomwolf's avatar
thomwolf committed
89
### Tests
thomwolf's avatar
thomwolf committed
90

DenysNahurnyi's avatar
DenysNahurnyi committed
91
A series of tests are included for the library and the example scripts. Library tests can be found in the [tests folder](https://github.com/huggingface/transformers/tree/master/transformers/tests) and examples tests in the [examples folder](https://github.com/huggingface/transformers/tree/master/examples).
thomwolf's avatar
thomwolf committed
92

thomwolf's avatar
thomwolf committed
93
These tests can be run using `pytest` (install pytest if needed with `pip install pytest`).
thomwolf's avatar
thomwolf committed
94

95
96
Depending on which framework is installed (TensorFlow 2.0 and/or PyTorch), the irrelevant tests will be skipped. Ensure that both frameworks are installed if you want to execute all tests.

thomwolf's avatar
thomwolf committed
97
You can run the tests from the root of the cloned repository with the commands:
thomwolf's avatar
thomwolf committed
98

thomwolf's avatar
thomwolf committed
99
```bash
100
python -m pytest -sv ./transformers/tests/
thomwolf's avatar
thomwolf committed
101
102
python -m pytest -sv ./examples/
```
thomwolf's avatar
thomwolf committed
103

104
105
106
107
### Do you want to run a Transformer model on a mobile device?

You should check out our [`swift-coreml-transformers`](https://github.com/huggingface/swift-coreml-transformers) repo.

108
It contains a set of tools to convert PyTorch or TensorFlow 2.0 trained Transformer models (currently contains `GPT-2`, `DistilGPT-2`, `BERT`, and `DistilBERT`) to CoreML models that run on iOS devices.
109

110
At some point in the future, you'll be able to seamlessly move from pre-training or fine-tuning models to productizing them in CoreML, or prototype a model or an app in CoreML then research its hyperparameters or architecture from TensorFlow 2.0 and/or PyTorch. Super exciting!
111

thomwolf's avatar
thomwolf committed
112
113
## Model architectures

114
🤗 Transformers currently provides 10 NLU/NLG architectures:
thomwolf's avatar
thomwolf committed
115
116
117
118
119
120
121
122

1. **[BERT](https://github.com/google-research/bert)** (from Google) released with the paper [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805) by Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova.
2. **[GPT](https://github.com/openai/finetune-transformer-lm)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
3. **[GPT-2](https://blog.openai.com/better-language-models/)** (from OpenAI) released with the paper [Language Models are Unsupervised Multitask Learners](https://blog.openai.com/better-language-models/) by Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**.
4. **[Transformer-XL](https://github.com/kimiyoung/transformer-xl)** (from Google/CMU) released with the paper [Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context](https://arxiv.org/abs/1901.02860) by Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov.
5. **[XLNet](https://github.com/zihangdai/xlnet/)** (from Google/CMU) released with the paper [​XLNet: Generalized Autoregressive Pretraining for Language Understanding](https://arxiv.org/abs/1906.08237) by Zhilin Yang*, Zihang Dai*, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, Quoc V. Le.
6. **[XLM](https://github.com/facebookresearch/XLM/)** (from Facebook) released together with the paper [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) by Guillaume Lample and Alexis Conneau.
7. **[RoBERTa](https://github.com/pytorch/fairseq/tree/master/examples/roberta)** (from Facebook), released together with the paper a [Robustly Optimized BERT Pretraining Approach](https://arxiv.org/abs/1907.11692) by Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov.
123
8. **[DistilBERT](https://github.com/huggingface/transformers/tree/master/examples/distillation)** (from HuggingFace), released together with the paper [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://arxiv.org/abs/1910.01108) by Victor Sanh, Lysandre Debut and Thomas Wolf. The same method has been applied to compress GPT2 into [DistilGPT2](https://github.com/huggingface/transformers/tree/master/examples/distillation).
keskarnitish's avatar
keskarnitish committed
124
9. **[CTRL](https://github.com/salesforce/ctrl/)** (from Salesforce) released with the paper [CTRL: A Conditional Transformer Language Model for Controllable Generation](https://arxiv.org/abs/1909.05858) by Nitish Shirish Keskar*, Bryan McCann*, Lav R. Varshney, Caiming Xiong and Richard Socher.
125
126
10. **[T5](https://github.com/google-research/text-to-text-transfer-transformer)** (from Google AI) released with the paper [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://arxiv.org/abs/1910.10683) by Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
11. Want to contribute a new model? We have added a **detailed guide and templates** to guide you in the process of adding a new model. You can find them in the [`templates`](./templates) folder of the repository. Be sure to check the [contributing guidelines](./CONTRIBUTING.md) and contact the maintainers or open an issue to collect feedbacks before starting your PR.
thomwolf's avatar
thomwolf committed
127
128
129

These implementations have been tested on several datasets (see the example scripts) and should match the performances of the original implementations (e.g. ~93 F1 on SQuAD for BERT Whole-Word-Masking, ~88 F1 on RocStories for OpenAI GPT, ~18.3 perplexity on WikiText 103 for Transformer-XL, ~0.916 Peason R coefficient on STS-B for XLNet). You can find more details on the performances in the Examples section of the [documentation](https://huggingface.co/transformers/examples.html).

130
131
132
133
134
135
136
## Online demo

**[Write With Transformer](https://transformer.huggingface.co)**, built by the Hugging Face team at transformer.huggingface.co, is the official demo of this repo’s text generation capabilities.
You can use it to experiment with completions generated by `GPT2Model`, `TransfoXLModel`, and `XLNetModel`.

> “🦄 Write with transformer is to writing what calculators are to calculus.”

Julien Chaumond's avatar
Julien Chaumond committed
137
![write_with_transformer](https://transformer.huggingface.co/front/assets/thumbnail-large.png)
138

thomwolf's avatar
thomwolf committed
139
## Quick tour
thomwolf's avatar
thomwolf committed
140

thomwolf's avatar
thomwolf committed
141
Let's do a very quick overview of the model architectures in 🤗 Transformers. Detailed examples for each model architecture (Bert, GPT, GPT-2, Transformer-XL, XLNet and XLM) can be found in the [full documentation](https://huggingface.co/transformers/).
thomwolf's avatar
thomwolf committed
142
143
144

```python
import torch
145
from transformers import *
thomwolf's avatar
thomwolf committed
146

147
# Transformers has a unified API
thomwolf's avatar
thomwolf committed
148
# for 8 transformer architectures and 30 pretrained weights.
thomwolf's avatar
thomwolf committed
149
#          Model          | Tokenizer          | Pretrained weights shortcut
thomwolf's avatar
thomwolf committed
150
151
152
MODELS = [(BertModel,       BertTokenizer,       'bert-base-uncased'),
          (OpenAIGPTModel,  OpenAIGPTTokenizer,  'openai-gpt'),
          (GPT2Model,       GPT2Tokenizer,       'gpt2'),
keskarnitish's avatar
keskarnitish committed
153
          (CTRLModel,       CTRLTokenizer,       'ctrl'),
thomwolf's avatar
thomwolf committed
154
155
156
157
158
          (TransfoXLModel,  TransfoXLTokenizer,  'transfo-xl-wt103'),
          (XLNetModel,      XLNetTokenizer,      'xlnet-base-cased'),
          (XLMModel,        XLMTokenizer,        'xlm-mlm-enfr-1024'),
          (DistilBertModel, DistilBertTokenizer, 'distilbert-base-uncased'),
          (RobertaModel,    RobertaTokenizer,    'roberta-base')]
thomwolf's avatar
thomwolf committed
159

thomwolf's avatar
thomwolf committed
160
161
# To use TensorFlow 2.0 versions of the models, simply prefix the class names with 'TF', e.g. `TFRobertaModel` is the TF 2.0 counterpart of the PyTorch model `RobertaModel`

thomwolf's avatar
thomwolf committed
162
163
164
165
166
167
168
# Let's encode some text in a sequence of hidden-states using each model:
for model_class, tokenizer_class, pretrained_weights in MODELS:
    # Load pretrained model/tokenizer
    tokenizer = tokenizer_class.from_pretrained(pretrained_weights)
    model = model_class.from_pretrained(pretrained_weights)

    # Encode text
169
    input_ids = torch.tensor([tokenizer.encode("Here is some text to encode", add_special_tokens=True)])  # Add special tokens takes care of adding [CLS], [SEP], <s>... tokens in the right way for each model.
Thomas Wolf's avatar
Thomas Wolf committed
170
171
    with torch.no_grad():
        last_hidden_states = model(input_ids)[0]  # Models outputs are now tuples
thomwolf's avatar
thomwolf committed
172
173
174
175
176
177

# Each architecture is provided with several class for fine-tuning on down-stream tasks, e.g.
BERT_MODEL_CLASSES = [BertModel, BertForPreTraining, BertForMaskedLM, BertForNextSentencePrediction,
                      BertForSequenceClassification, BertForMultipleChoice, BertForTokenClassification,
                      BertForQuestionAnswering]

thomwolf's avatar
thomwolf committed
178
179
180
# All the classes for an architecture can be initiated from pretrained weights for this architecture
# Note that additional weights added for fine-tuning are only initialized
# and need to be trained on the down-stream task
181
182
pretrained_weights = 'bert-base-uncased'
tokenizer = BertTokenizer.from_pretrained(pretrained_weights)
thomwolf's avatar
thomwolf committed
183
184
for model_class in BERT_MODEL_CLASSES:
    # Load pretrained model/tokenizer
185
    model = model_class.from_pretrained(pretrained_weights)
thomwolf's avatar
thomwolf committed
186

Santosh Gupta's avatar
Santosh Gupta committed
187
188
189
190
191
192
    # Models can return full list of hidden-states & attentions weights at each layer
    model = model_class.from_pretrained(pretrained_weights,
                                        output_hidden_states=True,
                                        output_attentions=True)
    input_ids = torch.tensor([tokenizer.encode("Let's see all hidden-states and attentions on this text")])
    all_hidden_states, all_attentions = model(input_ids)[-2:]
thomwolf's avatar
thomwolf committed
193

Santosh Gupta's avatar
Santosh Gupta committed
194
195
196
    # Models are compatible with Torchscript
    model = model_class.from_pretrained(pretrained_weights, torchscript=True)
    traced_model = torch.jit.trace(model, (input_ids,))
thomwolf's avatar
thomwolf committed
197

Santosh Gupta's avatar
Santosh Gupta committed
198
199
200
201
202
    # Simple serialization for models and tokenizers
    model.save_pretrained('./directory/to/save/')  # save
    model = model_class.from_pretrained('./directory/to/save/')  # re-load
    tokenizer.save_pretrained('./directory/to/save/')  # save
    tokenizer = BertTokenizer.from_pretrained('./directory/to/save/')  # re-load
thomwolf's avatar
thomwolf committed
203

Santosh Gupta's avatar
Santosh Gupta committed
204
    # SOTA examples for GLUE, SQUAD, text generation...
thomwolf's avatar
thomwolf committed
205
206
```

thomwolf's avatar
thomwolf committed
207
208
209
210
211
212
213
## Quick tour TF 2.0 training and PyTorch interoperability

Let's do a quick example of how a TensorFlow 2.0 model can be trained in 12 lines of code with 🤗 Transformers and then loaded in PyTorch for fast inspection/tests.

```python
import tensorflow as tf
import tensorflow_datasets
thomwolf's avatar
thomwolf committed
214
from transformers import *
thomwolf's avatar
thomwolf committed
215
216
217
218
219
220
221

# Load dataset, tokenizer, model from pretrained model/vocabulary
tokenizer = BertTokenizer.from_pretrained('bert-base-cased')
model = TFBertForSequenceClassification.from_pretrained('bert-base-cased')
data = tensorflow_datasets.load('glue/mrpc')

# Prepare dataset for GLUE as a tf.data.Dataset instance
thomwolf's avatar
thomwolf committed
222
223
train_dataset = glue_convert_examples_to_features(data['train'], tokenizer, max_length=128, task='mrpc')
valid_dataset = glue_convert_examples_to_features(data['validation'], tokenizer, max_length=128, task='mrpc')
thomwolf's avatar
thomwolf committed
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
train_dataset = train_dataset.shuffle(100).batch(32).repeat(2)
valid_dataset = valid_dataset.batch(64)

# Prepare training: Compile tf.keras model with optimizer, loss and learning rate schedule 
optimizer = tf.keras.optimizers.Adam(learning_rate=3e-5, epsilon=1e-08, clipnorm=1.0)
loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)
metric = tf.keras.metrics.SparseCategoricalAccuracy('accuracy')
model.compile(optimizer=optimizer, loss=loss, metrics=[metric])

# Train and evaluate using tf.keras.Model.fit()
history = model.fit(train_dataset, epochs=2, steps_per_epoch=115,
                    validation_data=valid_dataset, validation_steps=7)

# Load the TensorFlow model in PyTorch for inspection
model.save_pretrained('./save/')
pytorch_model = BertForSequenceClassification.from_pretrained('./save/', from_tf=True)

# Quickly test a few predictions - MRPC is a paraphrasing task, let's see if our model learned the task
sentence_0 = "This research was consistent with his findings."
sentence_1 = "His findings were compatible with this research."
sentence_2 = "His findings were not compatible with this research."
inputs_1 = tokenizer.encode_plus(sentence_0, sentence_1, add_special_tokens=True, return_tensors='pt')
inputs_2 = tokenizer.encode_plus(sentence_0, sentence_2, add_special_tokens=True, return_tensors='pt')

248
249
250
pred_1 = pytorch_model(inputs_1['input_ids'], token_type_ids=inputs_1['token_type_ids'])[0].argmax().item()
pred_2 = pytorch_model(inputs_2['input_ids'], token_type_ids=inputs_2['token_type_ids'])[0].argmax().item()

thomwolf's avatar
thomwolf committed
251
252
253
254
print("sentence_1 is", "a paraphrase" if pred_1 else "not a paraphrase", "of sentence_0")
print("sentence_2 is", "a paraphrase" if pred_2 else "not a paraphrase", "of sentence_0")
```

thomwolf's avatar
thomwolf committed
255
## Quick tour of the fine-tuning/usage scripts
thomwolf's avatar
thomwolf committed
256

thomwolf's avatar
thomwolf committed
257
The library comprises several example scripts with SOTA performances for NLU and NLG tasks:
thomwolf's avatar
thomwolf committed
258

thomwolf's avatar
thomwolf committed
259
260
- `run_glue.py`: an example fine-tuning Bert, XLNet and XLM on nine different GLUE tasks (*sequence-level classification*)
- `run_squad.py`: an example fine-tuning Bert, XLNet and XLM on the question answering dataset SQuAD 2.0 (*token-level classification*)
keskarnitish's avatar
keskarnitish committed
261
- `run_generation.py`: an example using GPT, GPT-2, CTRL, Transformer-XL and XLNet for conditional language generation
thomwolf's avatar
thomwolf committed
262
- other model-specific examples (see the documentation).
thomwolf's avatar
thomwolf committed
263

thomwolf's avatar
thomwolf committed
264
Here are three quick usage examples for these scripts:
thomwolf's avatar
thomwolf committed
265

thomwolf's avatar
thomwolf committed
266
### `run_glue.py`: Fine-tuning on GLUE tasks for sequence classification
thomwolf's avatar
thomwolf committed
267

thomwolf's avatar
thomwolf committed
268
The [General Language Understanding Evaluation (GLUE) benchmark](https://gluebenchmark.com/) is a collection of nine sentence- or sentence-pair language understanding tasks for evaluating and analyzing natural language understanding systems.
thomwolf's avatar
thomwolf committed
269

thomwolf's avatar
thomwolf committed
270
271
272
273
Before running anyone of these GLUE tasks you should download the
[GLUE data](https://gluebenchmark.com/tasks) by running
[this script](https://gist.github.com/W4ngatang/60c2bdb54d156a41194446737ce03e2e)
and unpack it to some directory `$GLUE_DIR`.
thomwolf's avatar
thomwolf committed
274

275
276
277
278
279
280
You should also install the additional packages required by the examples:

```shell
pip install -r ./examples/requirements.txt
```

thomwolf's avatar
thomwolf committed
281
282
283
```shell
export GLUE_DIR=/path/to/glue
export TASK_NAME=MRPC
thomwolf's avatar
thomwolf committed
284

285
286
287
288
289
290
291
292
293
294
295
296
297
298
python ./examples/run_glue.py \
    --model_type bert \
    --model_name_or_path bert-base-uncased \
    --task_name $TASK_NAME \
    --do_train \
    --do_eval \
    --do_lower_case \
    --data_dir $GLUE_DIR/$TASK_NAME \
    --max_seq_length 128 \
    --per_gpu_eval_batch_size=8   \
    --per_gpu_train_batch_size=8   \
    --learning_rate 2e-5 \
    --num_train_epochs 3.0 \
    --output_dir /tmp/$TASK_NAME/
thomwolf's avatar
thomwolf committed
299
300
```

thomwolf's avatar
thomwolf committed
301
where task name can be one of CoLA, SST-2, MRPC, STS-B, QQP, MNLI, QNLI, RTE, WNLI.
thomwolf's avatar
thomwolf committed
302

thomwolf's avatar
thomwolf committed
303
The dev set results will be present within the text file 'eval_results.txt' in the specified output_dir. In case of MNLI, since there are two separate dev sets, matched and mismatched, there will be a separate output folder called '/tmp/MNLI-MM/' in addition to '/tmp/MNLI/'.
thomwolf's avatar
thomwolf committed
304

thomwolf's avatar
thomwolf committed
305
#### Fine-tuning XLNet model on the STS-B regression task
thomwolf's avatar
thomwolf committed
306

thomwolf's avatar
thomwolf committed
307
This example code fine-tunes XLNet on the STS-B corpus using parallel training on a server with 4 V100 GPUs.
308
Parallel training is a simple way to use several GPUs (but is slower and less flexible than distributed training, see below).
thomwolf's avatar
thomwolf committed
309

thomwolf's avatar
thomwolf committed
310
311
```shell
export GLUE_DIR=/path/to/glue
thomwolf's avatar
thomwolf committed
312

thomwolf's avatar
thomwolf committed
313
314
315
316
python ./examples/run_glue.py \
    --model_type xlnet \
    --model_name_or_path xlnet-large-cased \
    --do_train  \
317
    --do_eval   \
thomwolf's avatar
thomwolf committed
318
319
320
321
322
323
324
325
326
327
328
329
    --task_name=sts-b     \
    --data_dir=${GLUE_DIR}/STS-B  \
    --output_dir=./proc_data/sts-b-110   \
    --max_seq_length=128   \
    --per_gpu_eval_batch_size=8   \
    --per_gpu_train_batch_size=8   \
    --gradient_accumulation_steps=1 \
    --max_steps=1200  \
    --model_name=xlnet-large-cased   \
    --overwrite_output_dir   \
    --overwrite_cache \
    --warmup_steps=120
thomwolf's avatar
thomwolf committed
330
331
```

Anthony MOI's avatar
Anthony MOI committed
332
On this machine we thus have a batch size of 32, please increase `gradient_accumulation_steps` to reach the same batch size if you have a smaller machine. These hyper-parameters should result in a Pearson correlation coefficient of `+0.917` on the development set.
thomwolf's avatar
thomwolf committed
333

thomwolf's avatar
thomwolf committed
334
#### Fine-tuning Bert model on the MRPC classification task
thomwolf's avatar
thomwolf committed
335

thomwolf's avatar
thomwolf committed
336
This example code fine-tunes the Bert Whole Word Masking model on the Microsoft Research Paraphrase Corpus (MRPC) corpus using distributed training on 8 V100 GPUs to reach a F1 > 92.
thomwolf's avatar
thomwolf committed
337

thomwolf's avatar
thomwolf committed
338
```bash
339
python -m torch.distributed.launch --nproc_per_node 8 ./examples/run_glue.py   \
thomwolf's avatar
thomwolf committed
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
    --model_type bert \
    --model_name_or_path bert-large-uncased-whole-word-masking \
    --task_name MRPC \
    --do_train   \
    --do_eval   \
    --do_lower_case   \
    --data_dir $GLUE_DIR/MRPC/   \
    --max_seq_length 128   \
    --per_gpu_eval_batch_size=8   \
    --per_gpu_train_batch_size=8   \
    --learning_rate 2e-5   \
    --num_train_epochs 3.0  \
    --output_dir /tmp/mrpc_output/ \
    --overwrite_output_dir   \
    --overwrite_cache \
thomwolf's avatar
thomwolf committed
355
356
```

thomwolf's avatar
thomwolf committed
357
Training with these hyper-parameters gave us the following results:
thomwolf's avatar
thomwolf committed
358

thomwolf's avatar
thomwolf committed
359
360
361
362
363
364
365
```bash
  acc = 0.8823529411764706
  acc_and_f1 = 0.901702786377709
  eval_loss = 0.3418912578906332
  f1 = 0.9210526315789473
  global_step = 174
  loss = 0.07231863956341798
thomwolf's avatar
thomwolf committed
366
367
```

thomwolf's avatar
thomwolf committed
368
### `run_squad.py`: Fine-tuning on SQuAD for question-answering
thomwolf's avatar
thomwolf committed
369

thomwolf's avatar
thomwolf committed
370
This example code fine-tunes BERT on the SQuAD dataset using distributed training on 8 V100 GPUs and Bert Whole Word Masking uncased model to reach a F1 > 93 on SQuAD:
thomwolf's avatar
thomwolf committed
371

thomwolf's avatar
thomwolf committed
372
```bash
373
python -m torch.distributed.launch --nproc_per_node=8 ./examples/run_squad.py \
thomwolf's avatar
thomwolf committed
374
375
376
    --model_type bert \
    --model_name_or_path bert-large-uncased-whole-word-masking \
    --do_train \
thomwolf's avatar
thomwolf committed
377
    --do_eval \
thomwolf's avatar
thomwolf committed
378
379
380
381
382
383
384
385
386
387
    --do_lower_case \
    --train_file $SQUAD_DIR/train-v1.1.json \
    --predict_file $SQUAD_DIR/dev-v1.1.json \
    --learning_rate 3e-5 \
    --num_train_epochs 2 \
    --max_seq_length 384 \
    --doc_stride 128 \
    --output_dir ../models/wwm_uncased_finetuned_squad/ \
    --per_gpu_eval_batch_size=3   \
    --per_gpu_train_batch_size=3   \
thomwolf's avatar
thomwolf committed
388
389
```

thomwolf's avatar
thomwolf committed
390
Training with these hyper-parameters gave us the following results:
thomwolf's avatar
thomwolf committed
391

thomwolf's avatar
thomwolf committed
392
393
394
```bash
python $SQUAD_DIR/evaluate-v1.1.py $SQUAD_DIR/dev-v1.1.json ../models/wwm_uncased_finetuned_squad/predictions.json
{"exact_match": 86.91579943235573, "f1": 93.1532499015869}
thomwolf's avatar
thomwolf committed
395
396
```

thomwolf's avatar
thomwolf committed
397
This is the model provided as `bert-large-uncased-whole-word-masking-finetuned-squad`.
398

keskarnitish's avatar
keskarnitish committed
399
### `run_generation.py`: Text generation with GPT, GPT-2, CTRL, Transformer-XL and XLNet
400

thomwolf's avatar
thomwolf committed
401
A conditional generation script is also included to generate text from a prompt.
DenysNahurnyi's avatar
DenysNahurnyi committed
402
The generation script includes the [tricks](https://github.com/rusiaaman/XLNet-gen#methodology) proposed by Aman Rusia to get high-quality generation with memory models like Transformer-XL and XLNet (include a predefined text to make short inputs longer).
403

thomwolf's avatar
thomwolf committed
404
Here is how to run the script with the small version of OpenAI GPT-2 model:
405

thomwolf's avatar
thomwolf committed
406
```shell
thomwolf's avatar
thomwolf committed
407
python ./examples/run_generation.py \
thomwolf's avatar
thomwolf committed
408
409
410
    --model_type=gpt2 \
    --length=20 \
    --model_name_or_path=gpt2 \
411
412
```

keskarnitish's avatar
keskarnitish committed
413
414
415
416
417
and from the Salesforce CTRL model: 
```shell
python ./examples/run_generation.py \
    --model_type=ctrl \
    --length=20 \
418
    --model_name_or_path=ctrl \
keskarnitish's avatar
keskarnitish committed
419
420
421
422
    --temperature=0 \
    --repetition_penalty=1.2 \
```

thomwolf's avatar
thomwolf committed
423
424
425
426
427
428
429
430
431
432
433
434
435
## Migrating from pytorch-transformers to transformers

Here is a quick summary of what you should take care of when migrating from `pytorch-transformers` to `transformers`.

### Positional order of some models' keywords inputs (`attention_mask`, `token_type_ids`...) changed

To be able to use Torchscript (see #1010, #1204 and #1195) the specific order of some models **keywords inputs** (`attention_mask`, `token_type_ids`...) has been changed.

If you used to call the models with keyword names for keyword arguments, e.g. `model(inputs_ids, attention_mask=attention_mask, token_type_ids=token_type_ids)`, this should not cause any change.

If you used to call the models with positional inputs for keyword arguments, e.g. `model(inputs_ids, attention_mask, token_type_ids)`, you may have to double check the exact order of input arguments.


436
## Migrating from pytorch-pretrained-bert to transformers
thomwolf's avatar
thomwolf committed
437

thomwolf's avatar
thomwolf committed
438
Here is a quick summary of what you should take care of when migrating from `pytorch-pretrained-bert` to `transformers`.
thomwolf's avatar
thomwolf committed
439
440
441

### Models always output `tuples`

442
The main breaking change when migrating from `pytorch-pretrained-bert` to `transformers` is that every model's forward method always outputs a `tuple` with various elements depending on the model and the configuration parameters.
thomwolf's avatar
thomwolf committed
443

DenysNahurnyi's avatar
DenysNahurnyi committed
444
The exact content of the tuples for each model is detailed in the models' docstrings and the [documentation](https://huggingface.co/transformers/).
thomwolf's avatar
thomwolf committed
445
446
447

In pretty much every case, you will be fine by taking the first element of the output as the output you previously used in `pytorch-pretrained-bert`.

448
Here is a `pytorch-pretrained-bert` to `transformers` conversion example for a `BertForSequenceClassification` classification model:
thomwolf's avatar
thomwolf committed
449
450
451
452
453
454
455
456

```python
# Let's load our model
model = BertForSequenceClassification.from_pretrained('bert-base-uncased')

# If you used to have this line in pytorch-pretrained-bert:
loss = model(input_ids, labels=labels)

457
# Now just use this line in transformers to extract the loss from the output tuple:
thomwolf's avatar
thomwolf committed
458
459
460
outputs = model(input_ids, labels=labels)
loss = outputs[0]

461
# In transformers you can also have access to the logits:
thomwolf's avatar
thomwolf committed
462
463
loss, logits = outputs[:2]

464
# And even the attention weights if you configure the model to output them (and other outputs too, see the docstrings and documentation)
thomwolf's avatar
thomwolf committed
465
466
467
468
469
model = BertForSequenceClassification.from_pretrained('bert-base-uncased', output_attentions=True)
outputs = model(input_ids, labels=labels)
loss, logits, attentions = outputs
```

470
471
472
473
### Using hidden states

By enabling the configuration option `output_hidden_states`, it was possible to retrieve the last hidden states of the encoder. In `pytorch-transformers` as well as `transformers` the return value has changed slightly: `all_hidden_states` now also includes the hidden state of the embeddings in addition to those of the encoding layers. This allows users to easily access the embeddings final state.

thomwolf's avatar
thomwolf committed
474
475
### Serialization

DenysNahurnyi's avatar
DenysNahurnyi committed
476
Breaking change in the `from_pretrained()` method:
477

Christopher Goh's avatar
Christopher Goh committed
478
1. Models are now set in evaluation mode by default when instantiated with the `from_pretrained()` method. To train them, don't forget to set them back in training mode (`model.train()`) to activate the dropout modules.
479

Christopher Goh's avatar
Christopher Goh committed
480
2. The additional `*input` and `**kwargs` arguments supplied to the `from_pretrained()` method used to be directly passed to the underlying model's class `__init__()` method. They are now used to update the model configuration attribute instead, which can break derived model classes built based on the previous `BertForSequenceClassification` examples. We are working on a way to mitigate this breaking change in [#866](https://github.com/huggingface/transformers/pull/866) by forwarding the the model's `__init__()` method (i) the provided positional arguments and (ii) the keyword arguments which do not match any configuration class attributes.
481

thomwolf's avatar
typos  
thomwolf committed
482
Also, while not a breaking change, the serialization methods have been standardized and you probably should switch to the new method `save_pretrained(save_directory)` if you were using any other serialization method before.
thomwolf's avatar
thomwolf committed
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508

Here is an example:

```python
### Let's load a model and tokenizer
model = BertForSequenceClassification.from_pretrained('bert-base-uncased')
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')

### Do some stuff to our model and tokenizer
# Ex: add new tokens to the vocabulary and embeddings of our model
tokenizer.add_tokens(['[SPECIAL_TOKEN_1]', '[SPECIAL_TOKEN_2]'])
model.resize_token_embeddings(len(tokenizer))
# Train our model
train(model)

### Now let's save our model and tokenizer to a directory
model.save_pretrained('./my_saved_model_directory/')
tokenizer.save_pretrained('./my_saved_model_directory/')

### Reload the model and the tokenizer
model = BertForSequenceClassification.from_pretrained('./my_saved_model_directory/')
tokenizer = BertTokenizer.from_pretrained('./my_saved_model_directory/')
```

### Optimizers: BertAdam & OpenAIAdam are now AdamW, schedules are standard PyTorch schedules

509
510
511
512
513
514
515
The two optimizers previously included, `BertAdam` and `OpenAIAdam`, have been replaced by a single `AdamW` optimizer which has a few differences:

- it only implements weights decay correction,
- schedules are now externals (see below),
- gradient clipping is now also external (see below).

The new optimizer `AdamW` matches PyTorch `Adam` optimizer API and let you use standard PyTorch or apex methods for the schedule and clipping.
thomwolf's avatar
thomwolf committed
516
517
518
519
520
521
522
523

The schedules are now standard [PyTorch learning rate schedulers](https://pytorch.org/docs/stable/optim.html#how-to-adjust-learning-rate) and not part of the optimizer anymore.

Here is a conversion examples from `BertAdam` with a linear warmup and decay schedule to `AdamW` and the same schedule:

```python
# Parameters:
lr = 1e-3
524
max_grad_norm = 1.0
thomwolf's avatar
thomwolf committed
525
526
527
528
529
530
531
532
533
534
535
536
num_total_steps = 1000
num_warmup_steps = 100
warmup_proportion = float(num_warmup_steps) / float(num_total_steps)  # 0.1

### Previously BertAdam optimizer was instantiated like this:
optimizer = BertAdam(model.parameters(), lr=lr, schedule='warmup_linear', warmup=warmup_proportion, t_total=num_total_steps)
### and used like this:
for batch in train_data:
    loss = model(batch)
    loss.backward()
    optimizer.step()

537
### In Transformers, optimizer and schedules are splitted and instantiated like this:
thomwolf's avatar
thomwolf committed
538
539
540
541
optimizer = AdamW(model.parameters(), lr=lr, correct_bias=False)  # To reproduce BertAdam specific behavior set correct_bias=False
scheduler = WarmupLinearSchedule(optimizer, warmup_steps=num_warmup_steps, t_total=num_total_steps)  # PyTorch scheduler
### and used like this:
for batch in train_data:
542
    model.train()
thomwolf's avatar
thomwolf committed
543
544
    loss = model(batch)
    loss.backward()
545
    torch.nn.utils.clip_grad_norm_(model.parameters(), max_grad_norm)  # Gradient clipping is not in AdamW anymore (so you can use amp without issue)
thomwolf's avatar
thomwolf committed
546
    optimizer.step()
thomwolf's avatar
thomwolf committed
547
    scheduler.step()
thomwolf's avatar
thomwolf committed
548
    optimizer.zero_grad()
thomwolf's avatar
thomwolf committed
549
550
```

thomwolf's avatar
thomwolf committed
551
## Citation
thomwolf's avatar
thomwolf committed
552

thomwolf's avatar
thomwolf committed
553
554
We now have a paper you can cite for the 🤗 Transformers library:
```
thomwolf's avatar
thomwolf committed
555
556
557
558
559
560
@article{Wolf2019HuggingFacesTS,
  title={HuggingFace's Transformers: State-of-the-art Natural Language Processing},
  author={Thomas Wolf and Lysandre Debut and Victor Sanh and Julien Chaumond and Clement Delangue and Anthony Moi and Pierric Cistac and Tim Rault and R'emi Louf and Morgan Funtowicz and Jamie Brew},
  journal={ArXiv},
  year={2019},
  volume={abs/1910.03771}
thomwolf's avatar
thomwolf committed
561
562
}
```