riaz.somc/transformers
1
0
Fork 0
mirror of https://github.com/huggingface/transformers.git synced 2025-07-04 05:10:06 +06:00
Code Issues Actions 26 Packages Projects Releases Wiki Activity

BIG Reorganize examples (#4213)

Browse Source 
* Created using Colaboratory

* [examples] reorganize files

* remove run_tpu_glue.py as superseded by TPU support in Trainer

* Bugfix: int, not tuple

* move files around
This commit is contained in:
Julien Chaumond 2020-05-07 13:48:44 -04:00 committed by GitHub
parent cafa6a9e29
commit 0ae96ff8a7
65 changed files with 1355 additions and 1308 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
.gitignore vendored
View File

@ -132,7 +132,8 @@ proc_data
runs runs
/runs_old /runs_old
/wandb /wandb
examples/runs /examples/runs
/examples/**/*.args
# data # data
/data /data

6
README.md
View File

@ -331,7 +331,7 @@ pip install -r ./examples/requirements.txt
export GLUE_DIR=/path/to/glue export GLUE_DIR=/path/to/glue
export TASK_NAME=MRPC export TASK_NAME=MRPC
python ./examples/run_glue.py \ python ./examples/text-classification/run_glue.py \
--model_name_or_path bert-base-uncased \ --model_name_or_path bert-base-uncased \
--task_name $TASK_NAME \ --task_name $TASK_NAME \
--do_train \ --do_train \
@ -357,7 +357,7 @@ Parallel training is a simple way to use several GPUs (but is slower and less fl
```shell ```shell
export GLUE_DIR=/path/to/glue export GLUE_DIR=/path/to/glue
python ./examples/run_glue.py \ python ./examples/text-classification/run_glue.py \
--model_name_or_path xlnet-large-cased \ --model_name_or_path xlnet-large-cased \
--do_train \ --do_train \
--do_eval \ --do_eval \
@ -382,7 +382,7 @@ On this machine we thus have a batch size of 32, please increase `gradient_accum
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. 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.
```bash ```bash
python -m torch.distributed.launch --nproc_per_node 8 ./examples/run_glue.py \ python -m torch.distributed.launch --nproc_per_node 8 ./examples/text-classification/run_glue.py \
--model_name_or_path bert-large-uncased-whole-word-masking \ --model_name_or_path bert-large-uncased-whole-word-masking \
--task_name MRPC \ --task_name MRPC \
--do_train \ --do_train \

1
docs/source/examples.md
View File

@ -1 +0,0 @@
../../examples/README.md

649
docs/source/examples.md Normal file
View File

@ -0,0 +1,649 @@
# Examples
In this section a few examples are put together. All of these examples work for several models, making use of the very
similar API between the different models.
**Important**
To run the latest versions of the examples, you have to install from source and install some specific requirements for the examples.
Execute the following steps in a new virtual environment:
```bash
git clone https://github.com/huggingface/transformers
cd transformers
pip install .
pip install -r ./examples/requirements.txt
```
| Section | Description |
|----------------------------|------------------------------------------------------------------------------------------------------------------------------------------
| [TensorFlow 2.0 models on GLUE](#TensorFlow-2.0-Bert-models-on-GLUE) | Examples running BERT TensorFlow 2.0 model on the GLUE tasks. |
| [Running on TPUs](#running-on-tpus) | Examples on running fine-tuning tasks on Google TPUs to accelerate workloads. |
| [Language Model training](#language-model-training) | Fine-tuning (or training from scratch) the library models for language modeling on a text dataset. Causal language modeling for GPT/GPT-2, masked language modeling for BERT/RoBERTa. |
| [Language Generation](#language-generation) | Conditional text generation using the auto-regressive models of the library: GPT, GPT-2, Transformer-XL and XLNet. |
| [GLUE](#glue) | Examples running BERT/XLM/XLNet/RoBERTa on the 9 GLUE tasks. Examples feature distributed training as well as half-precision. |
| [SQuAD](#squad) | Using BERT/RoBERTa/XLNet/XLM for question answering, examples with distributed training. |
| [Multiple Choice](#multiple-choice) | Examples running BERT/XLNet/RoBERTa on the SWAG/RACE/ARC tasks. |
| [Named Entity Recognition](https://github.com/huggingface/transformers/tree/master/examples/ner) | Using BERT for Named Entity Recognition (NER) on the CoNLL 2003 dataset, examples with distributed training. |
| [XNLI](#xnli) | Examples running BERT/XLM on the XNLI benchmark. |
| [Adversarial evaluation of model performances](#adversarial-evaluation-of-model-performances) | Testing a model with adversarial evaluation of natural language inference on the Heuristic Analysis for NLI Systems (HANS) dataset (McCoy et al., 2019.) |
## TensorFlow 2.0 Bert models on GLUE
Based on the script [`run_tf_glue.py`](https://github.com/huggingface/transformers/blob/master/examples/run_tf_glue.py).
Fine-tuning the library TensorFlow 2.0 Bert model for sequence classification on the MRPC task of the GLUE benchmark: [General Language Understanding Evaluation](https://gluebenchmark.com/).
This script has an option for mixed precision (Automatic Mixed Precision / AMP) to run models on Tensor Cores (NVIDIA Volta/Turing GPUs) and future hardware and an option for XLA, which uses the XLA compiler to reduce model runtime.
Options are toggled using `USE_XLA` or `USE_AMP` variables in the script.
These options and the below benchmark are provided by @tlkh.
Quick benchmarks from the script (no other modifications):
| GPU | Mode | Time (2nd epoch) | Val Acc (3 runs) |
| --------- | -------- | ----------------------- | ----------------------|
| Titan V | FP32 | 41s | 0.8438/0.8281/0.8333 |
| Titan V | AMP | 26s | 0.8281/0.8568/0.8411 |
| V100 | FP32 | 35s | 0.8646/0.8359/0.8464 |
| V100 | AMP | 22s | 0.8646/0.8385/0.8411 |
| 1080 Ti | FP32 | 55s | - |
Mixed precision (AMP) reduces the training time considerably for the same hardware and hyper-parameters (same batch size was used).
## Running on TPUs
You can accelerate your workloads on Google's TPUs. For information on how to setup your TPU environment refer to this
[README](https://github.com/pytorch/xla/blob/master/README.md).
The following are some examples of running the `*_tpu.py` finetuning scripts on TPUs. All steps for data preparation are
identical to your normal GPU + Huggingface setup.
### GLUE
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 `$GLUE_DIR`.
For running your GLUE task on MNLI dataset you can run something like the following:
```
export XRT_TPU_CONFIG="tpu_worker;0;$TPU_IP_ADDRESS:8470"
export GLUE_DIR=/path/to/glue
export TASK_NAME=MNLI
python run_glue_tpu.py \
--model_type bert \
--model_name_or_path bert-base-cased \
--task_name $TASK_NAME \
--do_train \
--do_eval \
--data_dir $GLUE_DIR/$TASK_NAME \
--max_seq_length 128 \
--train_batch_size 32 \
--learning_rate 3e-5 \
--num_train_epochs 3.0 \
--output_dir /tmp/$TASK_NAME \
--overwrite_output_dir \
--logging_steps 50 \
--save_steps 200 \
--num_cores=8 \
--only_log_master
```
## Language model training
Based on the script [`run_language_modeling.py`](https://github.com/huggingface/transformers/blob/master/examples/run_language_modeling.py).
Fine-tuning (or training from scratch) the library models for language modeling on a text dataset for GPT, GPT-2, BERT and RoBERTa (DistilBERT
to be added soon). GPT and GPT-2 are fine-tuned using a causal language modeling (CLM) loss while BERT and RoBERTa
are fine-tuned using a masked language modeling (MLM) loss.
Before running the following example, you should get a file that contains text on which the language model will be
trained or fine-tuned. A good example of such text is the [WikiText-2 dataset](https://blog.einstein.ai/the-wikitext-long-term-dependency-language-modeling-dataset/).
We will refer to two different files: `$TRAIN_FILE`, which contains text for training, and `$TEST_FILE`, which contains
text that will be used for evaluation.
### GPT-2/GPT and causal language modeling
The following example fine-tunes GPT-2 on WikiText-2. We're using the raw WikiText-2 (no tokens were replaced before
the tokenization). The loss here is that of causal language modeling.
```bash
export TRAIN_FILE=/path/to/dataset/wiki.train.raw
export TEST_FILE=/path/to/dataset/wiki.test.raw
python run_language_modeling.py \
--output_dir=output \
--model_type=gpt2 \
--model_name_or_path=gpt2 \
--do_train \
--train_data_file=$TRAIN_FILE \
--do_eval \
--eval_data_file=$TEST_FILE
```
This takes about half an hour to train on a single K80 GPU and about one minute for the evaluation to run. It reaches
a score of ~20 perplexity once fine-tuned on the dataset.
### RoBERTa/BERT and masked language modeling
The following example fine-tunes RoBERTa on WikiText-2. Here too, we're using the raw WikiText-2. The loss is different
as BERT/RoBERTa have a bidirectional mechanism; we're therefore using the same loss that was used during their
pre-training: masked language modeling.
In accordance to the RoBERTa paper, we use dynamic masking rather than static masking. The model may, therefore, converge
slightly slower (over-fitting takes more epochs).
We use the `--mlm` flag so that the script may change its loss function.
```bash
export TRAIN_FILE=/path/to/dataset/wiki.train.raw
export TEST_FILE=/path/to/dataset/wiki.test.raw
python run_language_modeling.py \
--output_dir=output \
--model_type=roberta \
--model_name_or_path=roberta-base \
--do_train \
--train_data_file=$TRAIN_FILE \
--do_eval \
--eval_data_file=$TEST_FILE \
--mlm
```
## Language generation
Based on the script [`run_generation.py`](https://github.com/huggingface/transformers/blob/master/examples/run_generation.py).
Conditional text generation using the auto-regressive models of the library: GPT, GPT-2, Transformer-XL, XLNet, CTRL.
A similar script is used for our official demo [Write With Transfomer](https://transformer.huggingface.co), where you
can try out the different models available in the library.
Example usage:
```bash
python run_generation.py \
--model_type=gpt2 \
--model_name_or_path=gpt2
```
## GLUE
Based on the script [`run_glue.py`](https://github.com/huggingface/transformers/blob/master/examples/text-classification/run_glue.py).
Fine-tuning the library models for sequence classification on the GLUE benchmark: [General Language Understanding
Evaluation](https://gluebenchmark.com/). This script can fine-tune the following models: BERT, XLM, XLNet and RoBERTa.
GLUE is made up of a total of 9 different tasks. We get the following results on the dev set of the benchmark with an
uncased BERT base model (the checkpoint `bert-base-uncased`). All experiments ran single V100 GPUs with a total train
batch sizes between 16 and 64. Some of these tasks have a small dataset and training can lead to high variance in the results
between different runs. We report the median on 5 runs (with different seeds) for each of the metrics.
| Task | Metric | Result |
|-------|------------------------------|-------------|
| CoLA | Matthew's corr | 49.23 |
| SST-2 | Accuracy | 91.97 |
| MRPC | F1/Accuracy | 89.47/85.29 |
| STS-B | Person/Spearman corr. | 83.95/83.70 |
| QQP | Accuracy/F1 | 88.40/84.31 |
| MNLI | Matched acc./Mismatched acc. | 80.61/81.08 |
| QNLI | Accuracy | 87.46 |
| RTE | Accuracy | 61.73 |
| WNLI | Accuracy | 45.07 |
Some of these results are significantly different from the ones reported on the test set
of GLUE benchmark on the website. For QQP and WNLI, please refer to [FAQ #12](https://gluebenchmark.com/faq) on the webite.
Before running any one 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`.
```bash
export GLUE_DIR=/path/to/glue
export TASK_NAME=MRPC
python run_glue.py \
--model_type bert \
--model_name_or_path bert-base-cased \
--task_name $TASK_NAME \
--do_train \
--do_eval \
--data_dir $GLUE_DIR/$TASK_NAME \
--max_seq_length 128 \
--per_gpu_train_batch_size 32 \
--learning_rate 2e-5 \
--num_train_epochs 3.0 \
--output_dir /tmp/$TASK_NAME/
```
where task name can be one of CoLA, SST-2, MRPC, STS-B, QQP, MNLI, QNLI, RTE, WNLI.
The dev set results will be present within the text file `eval_results.txt` in the specified output_dir.
In case of MNLI, since there are two separate dev sets (matched and mismatched), there will be a separate
output folder called `/tmp/MNLI-MM/` in addition to `/tmp/MNLI/`.
The code has not been tested with half-precision training with apex on any GLUE task apart from MRPC, MNLI,
CoLA, SST-2. The following section provides details on how to run half-precision training with MRPC. With that being
said, there shouldn’t be any issues in running half-precision training with the remaining GLUE tasks as well,
since the data processor for each task inherits from the base class DataProcessor.
### MRPC
#### Fine-tuning example
The following examples fine-tune BERT on the Microsoft Research Paraphrase Corpus (MRPC) corpus and runs in less
than 10 minutes on a single K-80 and in 27 seconds (!) on single tesla V100 16GB with apex installed.
Before running any one of these GLUE tasks you should download the
[GLUE data](https://gluebenchmark.com/tasks) by running
[this script](https://gist.github.com/W4ngatang/60c2bdb54d156a41194446737ce03e2e)
and unpack it to some directory `$GLUE_DIR`.
```bash
export GLUE_DIR=/path/to/glue
python run_glue.py \
--model_name_or_path bert-base-cased \
--task_name MRPC \
--do_train \
--do_eval \
--data_dir $GLUE_DIR/MRPC/ \
--max_seq_length 128 \
--per_gpu_train_batch_size 32 \
--learning_rate 2e-5 \
--num_train_epochs 3.0 \
--output_dir /tmp/mrpc_output/
```
Our test ran on a few seeds with [the original implementation hyper-
parameters](https://github.com/google-research/bert#sentence-and-sentence-pair-classification-tasks) gave evaluation
results between 84% and 88%.
#### Using Apex and mixed-precision
Using Apex and 16 bit precision, the fine-tuning on MRPC only takes 27 seconds. First install
[apex](https://github.com/NVIDIA/apex), then run the following example:
```bash
export GLUE_DIR=/path/to/glue
python run_glue.py \
--model_name_or_path bert-base-cased \
--task_name MRPC \
--do_train \
--do_eval \
--data_dir $GLUE_DIR/MRPC/ \
--max_seq_length 128 \
--per_gpu_train_batch_size 32 \
--learning_rate 2e-5 \
--num_train_epochs 3.0 \
--output_dir /tmp/mrpc_output/ \
--fp16
```
#### Distributed training
Here is an example using distributed training on 8 V100 GPUs. The model used is the BERT whole-word-masking and it
reaches F1 > 92 on MRPC.
```bash
export GLUE_DIR=/path/to/glue
python -m torch.distributed.launch \
--nproc_per_node 8 run_glue.py \
--model_name_or_path bert-base-cased \
--task_name MRPC \
--do_train \
--do_eval \
--data_dir $GLUE_DIR/MRPC/ \
--max_seq_length 128 \
--per_gpu_train_batch_size 8 \
--learning_rate 2e-5 \
--num_train_epochs 3.0 \
--output_dir /tmp/mrpc_output/
```
Training with these hyper-parameters gave us the following results:
```bash
acc = 0.8823529411764706
acc_and_f1 = 0.901702786377709
eval_loss = 0.3418912578906332
f1 = 0.9210526315789473
global_step = 174
loss = 0.07231863956341798
```
### MNLI
The following example uses the BERT-large, uncased, whole-word-masking model and fine-tunes it on the MNLI task.
```bash
export GLUE_DIR=/path/to/glue
python -m torch.distributed.launch \
--nproc_per_node 8 run_glue.py \
--model_name_or_path bert-base-cased \
--task_name mnli \
--do_train \
--do_eval \
--data_dir $GLUE_DIR/MNLI/ \
--max_seq_length 128 \
--per_gpu_train_batch_size 8 \
--learning_rate 2e-5 \
--num_train_epochs 3.0 \
--output_dir output_dir \
```
The results are the following:
```bash
***** Eval results *****
acc = 0.8679706601466992
eval_loss = 0.4911287787382479
global_step = 18408
loss = 0.04755385363816904
***** Eval results *****
acc = 0.8747965825874695
eval_loss = 0.45516540421714036
global_step = 18408
loss = 0.04755385363816904
```
## Multiple Choice
Based on the script [`run_multiple_choice.py`]().
#### Fine-tuning on SWAG
Download [swag](https://github.com/rowanz/swagaf/tree/master/data) data
```bash
#training on 4 tesla V100(16GB) GPUS
export SWAG_DIR=/path/to/swag_data_dir
python ./examples/run_multiple_choice.py \
--task_name swag \
--model_name_or_path roberta-base \
--do_train \
--do_eval \
--data_dir $SWAG_DIR \
--learning_rate 5e-5 \
--num_train_epochs 3 \
--max_seq_length 80 \
--output_dir models_bert/swag_base \
--per_gpu_eval_batch_size=16 \
--per_gpu_train_batch_size=16 \
--gradient_accumulation_steps 2 \
--overwrite_output
```
Training with the defined hyper-parameters yields the following results:
```
***** Eval results *****
eval_acc = 0.8338998300509847
eval_loss = 0.44457291918821606
```
## SQuAD
Based on the script [`run_squad.py`](https://github.com/huggingface/transformers/blob/master/examples/run_squad.py).
#### Fine-tuning BERT on SQuAD1.0
This example code fine-tunes BERT on the SQuAD1.0 dataset. It runs in 24 min (with BERT-base) or 68 min (with BERT-large)
on a single tesla V100 16GB. 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)
* [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)
And for SQuAD2.0, you need to download:
- [train-v2.0.json](https://rajpurkar.github.io/SQuAD-explorer/dataset/train-v2.0.json)
- [dev-v2.0.json](https://rajpurkar.github.io/SQuAD-explorer/dataset/dev-v2.0.json)
- [evaluate-v2.0.py](https://worksheets.codalab.org/rest/bundles/0x6b567e1cf2e041ec80d7098f031c5c9e/contents/blob/)
```bash
export SQUAD_DIR=/path/to/SQUAD
python run_squad.py \
--model_type bert \
--model_name_or_path bert-base-uncased \
--do_train \
--do_eval \
--train_file $SQUAD_DIR/train-v1.1.json \
--predict_file $SQUAD_DIR/dev-v1.1.json \
--per_gpu_train_batch_size 12 \
--learning_rate 3e-5 \
--num_train_epochs 2.0 \
--max_seq_length 384 \
--doc_stride 128 \
--output_dir /tmp/debug_squad/
```
Training with the previously defined hyper-parameters yields the following results:
```bash
f1 = 88.52
exact_match = 81.22
```
#### Distributed training
Here is an example using distributed training on 8 V100 GPUs and Bert Whole Word Masking uncased model to reach a F1 > 93 on SQuAD1.1:
```bash
python -m torch.distributed.launch --nproc_per_node=8 ./examples/run_squad.py \
--model_type bert \
--model_name_or_path bert-large-uncased-whole-word-masking \
--do_train \
--do_eval \
--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 ./examples/models/wwm_uncased_finetuned_squad/ \
--per_gpu_eval_batch_size=3 \
--per_gpu_train_batch_size=3 \
```
Training with the previously defined hyper-parameters yields the following results:
```bash
f1 = 93.15
exact_match = 86.91
```
This fine-tuned model is available as a checkpoint under the reference
`bert-large-uncased-whole-word-masking-finetuned-squad`.
#### Fine-tuning XLNet on SQuAD
This example code fine-tunes XLNet on both SQuAD1.0 and SQuAD2.0 dataset. See above to download the data for SQuAD .
##### Command for SQuAD1.0:
```bash
export SQUAD_DIR=/path/to/SQUAD
python run_squad.py \
--model_type xlnet \
--model_name_or_path xlnet-large-cased \
--do_train \
--do_eval \
--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 ./wwm_cased_finetuned_squad/ \
--per_gpu_eval_batch_size=4 \
--per_gpu_train_batch_size=4 \
--save_steps 5000
```
##### Command for SQuAD2.0:
```bash
export SQUAD_DIR=/path/to/SQUAD
python run_squad.py \
--model_type xlnet \
--model_name_or_path xlnet-large-cased \
--do_train \
--do_eval \
--version_2_with_negative \
--train_file $SQUAD_DIR/train-v2.0.json \
--predict_file $SQUAD_DIR/dev-v2.0.json \
--learning_rate 3e-5 \
--num_train_epochs 4 \
--max_seq_length 384 \
--doc_stride 128 \
--output_dir ./wwm_cased_finetuned_squad/ \
--per_gpu_eval_batch_size=2 \
--per_gpu_train_batch_size=2 \
--save_steps 5000
```
Larger batch size may improve the performance while costing more memory.
##### Results for SQuAD1.0 with the previously defined hyper-parameters:
```python
{
"exact": 85.45884578997162,
"f1": 92.5974600601065,
"total": 10570,
"HasAns_exact": 85.45884578997162,
"HasAns_f1": 92.59746006010651,
"HasAns_total": 10570
}
```
##### Results for SQuAD2.0 with the previously defined hyper-parameters:
```python
{
"exact": 80.4177545691906,
"f1": 84.07154997729623,
"total": 11873,
"HasAns_exact": 76.73751686909581,
"HasAns_f1": 84.05558584352873,
"HasAns_total": 5928,
"NoAns_exact": 84.0874684608915,
"NoAns_f1": 84.0874684608915,
"NoAns_total": 5945
}
```
## XNLI
Based on the script [`run_xnli.py`](https://github.com/huggingface/transformers/blob/master/examples/run_xnli.py).
[XNLI](https://www.nyu.edu/projects/bowman/xnli/) is crowd-sourced dataset based on [MultiNLI](http://www.nyu.edu/projects/bowman/multinli/). It is an evaluation benchmark for cross-lingual text representations. Pairs of text are labeled with textual entailment annotations for 15 different languages (including both high-resource language such as English and low-resource languages such as Swahili).
#### Fine-tuning on XNLI
This example code fine-tunes mBERT (multi-lingual BERT) on the XNLI dataset. It runs in 106 mins
on a single tesla V100 16GB. The data for XNLI can be downloaded with the following links and should be both saved (and un-zipped) in a
`$XNLI_DIR` directory.
* [XNLI 1.0](https://www.nyu.edu/projects/bowman/xnli/XNLI-1.0.zip)
* [XNLI-MT 1.0](https://www.nyu.edu/projects/bowman/xnli/XNLI-MT-1.0.zip)
```bash
export XNLI_DIR=/path/to/XNLI
python run_xnli.py \
--model_type bert \
--model_name_or_path bert-base-multilingual-cased \
--language de \
--train_language en \
--do_train \
--do_eval \
--data_dir $XNLI_DIR \
--per_gpu_train_batch_size 32 \
--learning_rate 5e-5 \
--num_train_epochs 2.0 \
--max_seq_length 128 \
--output_dir /tmp/debug_xnli/ \
--save_steps -1
```
Training with the previously defined hyper-parameters yields the following results on the **test** set:
```bash
acc = 0.7093812375249501
```
## MM-IMDb
Based on the script [`run_mmimdb.py`](https://github.com/huggingface/transformers/blob/master/examples/contrib/mm-imdb/run_mmimdb.py).
[MM-IMDb](http://lisi1.unal.edu.co/mmimdb/) is a Multimodal dataset with around 26,000 movies including images, plots and other metadata.
### Training on MM-IMDb
```
python run_mmimdb.py \
--data_dir /path/to/mmimdb/dataset/ \
--model_type bert \
--model_name_or_path bert-base-uncased \
--output_dir /path/to/save/dir/ \
--do_train \
--do_eval \
--max_seq_len 512 \
--gradient_accumulation_steps 20 \
--num_image_embeds 3 \
--num_train_epochs 100 \
--patience 5
```
## Adversarial evaluation of model performances
Here is an example on evaluating a model using adversarial evaluation of natural language inference with the Heuristic Analysis for NLI Systems (HANS) dataset [McCoy et al., 2019](https://arxiv.org/abs/1902.01007). The example was gracefully provided by [Nafise Sadat Moosavi](https://github.com/ns-moosavi).
The HANS dataset can be downloaded from [this location](https://github.com/tommccoy1/hans).
This is an example of using test_hans.py:
```bash
export HANS_DIR=path-to-hans
export MODEL_TYPE=type-of-the-model-e.g.-bert-roberta-xlnet-etc
export MODEL_PATH=path-to-the-model-directory-that-is-trained-on-NLI-e.g.-by-using-run_glue.py
python examples/hans/test_hans.py \
--task_name hans \
--model_type $MODEL_TYPE \
--do_eval \
--data_dir $HANS_DIR \
--model_name_or_path $MODEL_PATH \
--max_seq_length 128 \
--output_dir $MODEL_PATH \
```
This will create the hans_predictions.txt file in MODEL_PATH, which can then be evaluated using hans/evaluate_heur_output.py from the HANS dataset.
The results of the BERT-base model that is trained on MNLI using batch size 8 and the random seed 42 on the HANS dataset is as follows:
```bash
Heuristic entailed results:
lexical_overlap: 0.9702
subsequence: 0.9942
constituent: 0.9962
Heuristic non-entailed results:
lexical_overlap: 0.199
subsequence: 0.0396
constituent: 0.118
```

2
docs/source/main_classes/processors.rst
View File

@ -54,7 +54,7 @@ Additionally, the following method can be used to load values from a data file
Example usage Example usage
^^^^^^^^^^^^^^^^^^^^^^^^^ ^^^^^^^^^^^^^^^^^^^^^^^^^
An example using these processors is given in the `run_glue.py <https://github.com/huggingface/pytorch-transformers/blob/master/examples/run_glue.py>`__ script. An example using these processors is given in the `run_glue.py <https://github.com/huggingface/pytorch-transformers/blob/master/examples/text-classification/run_glue.py>`__ script.
XNLI XNLI

2
docs/source/usage.rst
View File

@ -44,7 +44,7 @@ Sequence Classification
Sequence classification is the task of classifying sequences according to a given number of classes. An example Sequence classification is the task of classifying sequences according to a given number of classes. An example
of sequence classification is the GLUE dataset, which is entirely based on that task. If you would like to fine-tune of sequence classification is the GLUE dataset, which is entirely based on that task. If you would like to fine-tune
a model on a GLUE sequence classification task, you may leverage the a model on a GLUE sequence classification task, you may leverage the
`run_glue.py <https://github.com/huggingface/transformers/tree/master/examples/run_glue.py>`_ or `run_glue.py <https://github.com/huggingface/transformers/tree/master/examples/text-classification/run_glue.py>`_ or
`run_tf_glue.py <https://github.com/huggingface/transformers/tree/master/examples/run_tf_glue.py>`_ scripts. `run_tf_glue.py <https://github.com/huggingface/transformers/tree/master/examples/run_tf_glue.py>`_ scripts.
Here is an example using the pipelines do to sentiment analysis: identifying if a sequence is positive or negative. Here is an example using the pipelines do to sentiment analysis: identifying if a sequence is positive or negative.

622
examples/README.md
View File

@ -15,7 +15,7 @@ pip install -r ./examples/requirements.txt
``` ```
| Section | Description | | Section | Description |
|----------------------------|------------------------------------------------------------------------------------------------------------------------------------------ |----------------------------|-----------------------------------------------------
| [TensorFlow 2.0 models on GLUE](#TensorFlow-2.0-Bert-models-on-GLUE) | Examples running BERT TensorFlow 2.0 model on the GLUE tasks. | | [TensorFlow 2.0 models on GLUE](#TensorFlow-2.0-Bert-models-on-GLUE) | Examples running BERT TensorFlow 2.0 model on the GLUE tasks. |
| [Running on TPUs](#running-on-tpus) | Examples on running fine-tuning tasks on Google TPUs to accelerate workloads. | | [Running on TPUs](#running-on-tpus) | Examples on running fine-tuning tasks on Google TPUs to accelerate workloads. |
| [Language Model training](#language-model-training) | Fine-tuning (or training from scratch) the library models for language modeling on a text dataset. Causal language modeling for GPT/GPT-2, masked language modeling for BERT/RoBERTa. | | [Language Model training](#language-model-training) | Fine-tuning (or training from scratch) the library models for language modeling on a text dataset. Causal language modeling for GPT/GPT-2, masked language modeling for BERT/RoBERTa. |
@ -27,623 +27,3 @@ pip install -r ./examples/requirements.txt
| [XNLI](#xnli) | Examples running BERT/XLM on the XNLI benchmark. | | [XNLI](#xnli) | Examples running BERT/XLM on the XNLI benchmark. |
| [Adversarial evaluation of model performances](#adversarial-evaluation-of-model-performances) | Testing a model with adversarial evaluation of natural language inference on the Heuristic Analysis for NLI Systems (HANS) dataset (McCoy et al., 2019.) | | [Adversarial evaluation of model performances](#adversarial-evaluation-of-model-performances) | Testing a model with adversarial evaluation of natural language inference on the Heuristic Analysis for NLI Systems (HANS) dataset (McCoy et al., 2019.) |
## TensorFlow 2.0 Bert models on GLUE
Based on the script [`run_tf_glue.py`](https://github.com/huggingface/transformers/blob/master/examples/run_tf_glue.py).
Fine-tuning the library TensorFlow 2.0 Bert model for sequence classification on the MRPC task of the GLUE benchmark: [General Language Understanding Evaluation](https://gluebenchmark.com/).
This script has an option for mixed precision (Automatic Mixed Precision / AMP) to run models on Tensor Cores (NVIDIA Volta/Turing GPUs) and future hardware and an option for XLA, which uses the XLA compiler to reduce model runtime.
Options are toggled using `USE_XLA` or `USE_AMP` variables in the script.
These options and the below benchmark are provided by @tlkh.
Quick benchmarks from the script (no other modifications):
| GPU | Mode | Time (2nd epoch) | Val Acc (3 runs) |
| --------- | -------- | ----------------------- | ----------------------|
| Titan V | FP32 | 41s | 0.8438/0.8281/0.8333 |
| Titan V | AMP | 26s | 0.8281/0.8568/0.8411 |
| V100 | FP32 | 35s | 0.8646/0.8359/0.8464 |
| V100 | AMP | 22s | 0.8646/0.8385/0.8411 |
| 1080 Ti | FP32 | 55s | - |
Mixed precision (AMP) reduces the training time considerably for the same hardware and hyper-parameters (same batch size was used).
## Running on TPUs
You can accelerate your workloads on Google's TPUs. For information on how to setup your TPU environment refer to this
[README](https://github.com/pytorch/xla/blob/master/README.md).
The following are some examples of running the `*_tpu.py` finetuning scripts on TPUs. All steps for data preparation are
identical to your normal GPU + Huggingface setup.
### GLUE
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 `$GLUE_DIR`.
For running your GLUE task on MNLI dataset you can run something like the following:
```
export XRT_TPU_CONFIG="tpu_worker;0;$TPU_IP_ADDRESS:8470"
export GLUE_DIR=/path/to/glue
export TASK_NAME=MNLI
python run_glue_tpu.py \
--model_type bert \
--model_name_or_path bert-base-cased \
--task_name $TASK_NAME \
--do_train \
--do_eval \
--data_dir $GLUE_DIR/$TASK_NAME \
--max_seq_length 128 \
--train_batch_size 32 \
--learning_rate 3e-5 \
--num_train_epochs 3.0 \
--output_dir /tmp/$TASK_NAME \
--overwrite_output_dir \
--logging_steps 50 \
--save_steps 200 \
--num_cores=8 \
--only_log_master
```
## Language model training
Based on the script [`run_language_modeling.py`](https://github.com/huggingface/transformers/blob/master/examples/run_language_modeling.py).
Fine-tuning (or training from scratch) the library models for language modeling on a text dataset for GPT, GPT-2, BERT and RoBERTa (DistilBERT
to be added soon). GPT and GPT-2 are fine-tuned using a causal language modeling (CLM) loss while BERT and RoBERTa
are fine-tuned using a masked language modeling (MLM) loss.
Before running the following example, you should get a file that contains text on which the language model will be
trained or fine-tuned. A good example of such text is the [WikiText-2 dataset](https://blog.einstein.ai/the-wikitext-long-term-dependency-language-modeling-dataset/).
We will refer to two different files: `$TRAIN_FILE`, which contains text for training, and `$TEST_FILE`, which contains
text that will be used for evaluation.
### GPT-2/GPT and causal language modeling
The following example fine-tunes GPT-2 on WikiText-2. We're using the raw WikiText-2 (no tokens were replaced before
the tokenization). The loss here is that of causal language modeling.
```bash
export TRAIN_FILE=/path/to/dataset/wiki.train.raw
export TEST_FILE=/path/to/dataset/wiki.test.raw
python run_language_modeling.py \
--output_dir=output \
--model_type=gpt2 \
--model_name_or_path=gpt2 \
--do_train \
--train_data_file=$TRAIN_FILE \
--do_eval \
--eval_data_file=$TEST_FILE
```
This takes about half an hour to train on a single K80 GPU and about one minute for the evaluation to run. It reaches
a score of ~20 perplexity once fine-tuned on the dataset.
### RoBERTa/BERT and masked language modeling
The following example fine-tunes RoBERTa on WikiText-2. Here too, we're using the raw WikiText-2. The loss is different
as BERT/RoBERTa have a bidirectional mechanism; we're therefore using the same loss that was used during their
pre-training: masked language modeling.
In accordance to the RoBERTa paper, we use dynamic masking rather than static masking. The model may, therefore, converge
slightly slower (over-fitting takes more epochs).
We use the `--mlm` flag so that the script may change its loss function.
```bash
export TRAIN_FILE=/path/to/dataset/wiki.train.raw
export TEST_FILE=/path/to/dataset/wiki.test.raw
python run_language_modeling.py \
--output_dir=output \
--model_type=roberta \
--model_name_or_path=roberta-base \
--do_train \
--train_data_file=$TRAIN_FILE \
--do_eval \
--eval_data_file=$TEST_FILE \
--mlm
```
## Language generation
Based on the script [`run_generation.py`](https://github.com/huggingface/transformers/blob/master/examples/run_generation.py).
Conditional text generation using the auto-regressive models of the library: GPT, GPT-2, Transformer-XL, XLNet, CTRL.
A similar script is used for our official demo [Write With Transfomer](https://transformer.huggingface.co), where you
can try out the different models available in the library.
Example usage:
```bash
python run_generation.py \
--model_type=gpt2 \
--model_name_or_path=gpt2
```
## GLUE
Based on the script [`run_glue.py`](https://github.com/huggingface/transformers/blob/master/examples/run_glue.py).
Fine-tuning the library models for sequence classification on the GLUE benchmark: [General Language Understanding
Evaluation](https://gluebenchmark.com/). This script can fine-tune the following models: BERT, XLM, XLNet and RoBERTa.
GLUE is made up of a total of 9 different tasks. We get the following results on the dev set of the benchmark with an
uncased BERT base model (the checkpoint `bert-base-uncased`). All experiments ran single V100 GPUs with a total train
batch sizes between 16 and 64. Some of these tasks have a small dataset and training can lead to high variance in the results
between different runs. We report the median on 5 runs (with different seeds) for each of the metrics.
| Task | Metric | Result |
|-------|------------------------------|-------------|
| CoLA | Matthew's corr | 49.23 |
| SST-2 | Accuracy | 91.97 |
| MRPC | F1/Accuracy | 89.47/85.29 |
| STS-B | Person/Spearman corr. | 83.95/83.70 |
| QQP | Accuracy/F1 | 88.40/84.31 |
| MNLI | Matched acc./Mismatched acc. | 80.61/81.08 |
| QNLI | Accuracy | 87.46 |
| RTE | Accuracy | 61.73 |
| WNLI | Accuracy | 45.07 |
Some of these results are significantly different from the ones reported on the test set
of GLUE benchmark on the website. For QQP and WNLI, please refer to [FAQ #12](https://gluebenchmark.com/faq) on the webite.
Before running any one 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`.
```bash
export GLUE_DIR=/path/to/glue
export TASK_NAME=MRPC
python run_glue.py \
--model_type bert \
--model_name_or_path bert-base-cased \
--task_name $TASK_NAME \
--do_train \
--do_eval \
--data_dir $GLUE_DIR/$TASK_NAME \
--max_seq_length 128 \
--per_gpu_train_batch_size 32 \
--learning_rate 2e-5 \
--num_train_epochs 3.0 \
--output_dir /tmp/$TASK_NAME/
```
where task name can be one of CoLA, SST-2, MRPC, STS-B, QQP, MNLI, QNLI, RTE, WNLI.
The dev set results will be present within the text file `eval_results.txt` in the specified output_dir.
In case of MNLI, since there are two separate dev sets (matched and mismatched), there will be a separate
output folder called `/tmp/MNLI-MM/` in addition to `/tmp/MNLI/`.
The code has not been tested with half-precision training with apex on any GLUE task apart from MRPC, MNLI,
CoLA, SST-2. The following section provides details on how to run half-precision training with MRPC. With that being
said, there shouldn’t be any issues in running half-precision training with the remaining GLUE tasks as well,
since the data processor for each task inherits from the base class DataProcessor.
### MRPC
#### Fine-tuning example
The following examples fine-tune BERT on the Microsoft Research Paraphrase Corpus (MRPC) corpus and runs in less
than 10 minutes on a single K-80 and in 27 seconds (!) on single tesla V100 16GB with apex installed.
Before running any one of these GLUE tasks you should download the
[GLUE data](https://gluebenchmark.com/tasks) by running
[this script](https://gist.github.com/W4ngatang/60c2bdb54d156a41194446737ce03e2e)
and unpack it to some directory `$GLUE_DIR`.
```bash
export GLUE_DIR=/path/to/glue
python run_glue.py \
--model_name_or_path bert-base-cased \
--task_name MRPC \
--do_train \
--do_eval \
--data_dir $GLUE_DIR/MRPC/ \
--max_seq_length 128 \
--per_gpu_train_batch_size 32 \
--learning_rate 2e-5 \
--num_train_epochs 3.0 \
--output_dir /tmp/mrpc_output/
```
Our test ran on a few seeds with [the original implementation hyper-
parameters](https://github.com/google-research/bert#sentence-and-sentence-pair-classification-tasks) gave evaluation
results between 84% and 88%.
#### Using Apex and mixed-precision
Using Apex and 16 bit precision, the fine-tuning on MRPC only takes 27 seconds. First install
[apex](https://github.com/NVIDIA/apex), then run the following example:
```bash
export GLUE_DIR=/path/to/glue
python run_glue.py \
--model_name_or_path bert-base-cased \
--task_name MRPC \
--do_train \
--do_eval \
--data_dir $GLUE_DIR/MRPC/ \
--max_seq_length 128 \
--per_gpu_train_batch_size 32 \
--learning_rate 2e-5 \
--num_train_epochs 3.0 \
--output_dir /tmp/mrpc_output/ \
--fp16
```
#### Distributed training
Here is an example using distributed training on 8 V100 GPUs. The model used is the BERT whole-word-masking and it
reaches F1 > 92 on MRPC.
```bash
export GLUE_DIR=/path/to/glue
python -m torch.distributed.launch \
--nproc_per_node 8 run_glue.py \
--model_name_or_path bert-base-cased \
--task_name MRPC \
--do_train \
--do_eval \
--data_dir $GLUE_DIR/MRPC/ \
--max_seq_length 128 \
--per_gpu_train_batch_size 8 \
--learning_rate 2e-5 \
--num_train_epochs 3.0 \
--output_dir /tmp/mrpc_output/
```
Training with these hyper-parameters gave us the following results:
```bash
acc = 0.8823529411764706
acc_and_f1 = 0.901702786377709
eval_loss = 0.3418912578906332
f1 = 0.9210526315789473
global_step = 174
loss = 0.07231863956341798
```
### MNLI
The following example uses the BERT-large, uncased, whole-word-masking model and fine-tunes it on the MNLI task.
```bash
export GLUE_DIR=/path/to/glue
python -m torch.distributed.launch \
--nproc_per_node 8 run_glue.py \
--model_name_or_path bert-base-cased \
--task_name mnli \
--do_train \
--do_eval \
--data_dir $GLUE_DIR/MNLI/ \
--max_seq_length 128 \
--per_gpu_train_batch_size 8 \
--learning_rate 2e-5 \
--num_train_epochs 3.0 \
--output_dir output_dir \
```
The results are the following:
```bash
***** Eval results *****
acc = 0.8679706601466992
eval_loss = 0.4911287787382479
global_step = 18408
loss = 0.04755385363816904
***** Eval results *****
acc = 0.8747965825874695
eval_loss = 0.45516540421714036
global_step = 18408
loss = 0.04755385363816904
```
## Multiple Choice
Based on the script [`run_multiple_choice.py`]().
#### Fine-tuning on SWAG
Download [swag](https://github.com/rowanz/swagaf/tree/master/data) data
```bash
#training on 4 tesla V100(16GB) GPUS
export SWAG_DIR=/path/to/swag_data_dir
python ./examples/run_multiple_choice.py \
--task_name swag \
--model_name_or_path roberta-base \
--do_train \
--do_eval \
--data_dir $SWAG_DIR \
--learning_rate 5e-5 \
--num_train_epochs 3 \
--max_seq_length 80 \
--output_dir models_bert/swag_base \
--per_gpu_eval_batch_size=16 \
--per_gpu_train_batch_size=16 \
--gradient_accumulation_steps 2 \
--overwrite_output
```
Training with the defined hyper-parameters yields the following results:
```
***** Eval results *****
eval_acc = 0.8338998300509847
eval_loss = 0.44457291918821606
```
## SQuAD
Based on the script [`run_squad.py`](https://github.com/huggingface/transformers/blob/master/examples/run_squad.py).
#### Fine-tuning BERT on SQuAD1.0
This example code fine-tunes BERT on the SQuAD1.0 dataset. It runs in 24 min (with BERT-base) or 68 min (with BERT-large)
on a single tesla V100 16GB. 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)
* [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)
And for SQuAD2.0, you need to download:
- [train-v2.0.json](https://rajpurkar.github.io/SQuAD-explorer/dataset/train-v2.0.json)
- [dev-v2.0.json](https://rajpurkar.github.io/SQuAD-explorer/dataset/dev-v2.0.json)
- [evaluate-v2.0.py](https://worksheets.codalab.org/rest/bundles/0x6b567e1cf2e041ec80d7098f031c5c9e/contents/blob/)
```bash
export SQUAD_DIR=/path/to/SQUAD
python run_squad.py \
--model_type bert \
--model_name_or_path bert-base-uncased \
--do_train \
--do_eval \
--train_file $SQUAD_DIR/train-v1.1.json \
--predict_file $SQUAD_DIR/dev-v1.1.json \
--per_gpu_train_batch_size 12 \
--learning_rate 3e-5 \
--num_train_epochs 2.0 \
--max_seq_length 384 \
--doc_stride 128 \
--output_dir /tmp/debug_squad/
```
Training with the previously defined hyper-parameters yields the following results:
```bash
f1 = 88.52
exact_match = 81.22
```
#### Distributed training
Here is an example using distributed training on 8 V100 GPUs and Bert Whole Word Masking uncased model to reach a F1 > 93 on SQuAD1.1:
```bash
python -m torch.distributed.launch --nproc_per_node=8 ./examples/run_squad.py \
--model_type bert \
--model_name_or_path bert-large-uncased-whole-word-masking \
--do_train \
--do_eval \
--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 ./examples/models/wwm_uncased_finetuned_squad/ \
--per_gpu_eval_batch_size=3 \
--per_gpu_train_batch_size=3 \
```
Training with the previously defined hyper-parameters yields the following results:
```bash
f1 = 93.15
exact_match = 86.91
```
This fine-tuned model is available as a checkpoint under the reference
`bert-large-uncased-whole-word-masking-finetuned-squad`.
#### Fine-tuning XLNet on SQuAD
This example code fine-tunes XLNet on both SQuAD1.0 and SQuAD2.0 dataset. See above to download the data for SQuAD .
##### Command for SQuAD1.0:
```bash
export SQUAD_DIR=/path/to/SQUAD
python run_squad.py \
--model_type xlnet \
--model_name_or_path xlnet-large-cased \
--do_train \
--do_eval \
--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 ./wwm_cased_finetuned_squad/ \
--per_gpu_eval_batch_size=4 \
--per_gpu_train_batch_size=4 \
--save_steps 5000
```
##### Command for SQuAD2.0:
```bash
export SQUAD_DIR=/path/to/SQUAD
python run_squad.py \
--model_type xlnet \
--model_name_or_path xlnet-large-cased \
--do_train \
--do_eval \
--version_2_with_negative \
--train_file $SQUAD_DIR/train-v2.0.json \
--predict_file $SQUAD_DIR/dev-v2.0.json \
--learning_rate 3e-5 \
--num_train_epochs 4 \
--max_seq_length 384 \
--doc_stride 128 \
--output_dir ./wwm_cased_finetuned_squad/ \
--per_gpu_eval_batch_size=2 \
--per_gpu_train_batch_size=2 \
--save_steps 5000
```
Larger batch size may improve the performance while costing more memory.
##### Results for SQuAD1.0 with the previously defined hyper-parameters:
```python
{
"exact": 85.45884578997162,
"f1": 92.5974600601065,
"total": 10570,
"HasAns_exact": 85.45884578997162,
"HasAns_f1": 92.59746006010651,
"HasAns_total": 10570
}
```
##### Results for SQuAD2.0 with the previously defined hyper-parameters:
```python
{
"exact": 80.4177545691906,
"f1": 84.07154997729623,
"total": 11873,
"HasAns_exact": 76.73751686909581,
"HasAns_f1": 84.05558584352873,
"HasAns_total": 5928,
"NoAns_exact": 84.0874684608915,
"NoAns_f1": 84.0874684608915,
"NoAns_total": 5945
}
```
## XNLI
Based on the script [`run_xnli.py`](https://github.com/huggingface/transformers/blob/master/examples/run_xnli.py).
[XNLI](https://www.nyu.edu/projects/bowman/xnli/) is crowd-sourced dataset based on [MultiNLI](http://www.nyu.edu/projects/bowman/multinli/). It is an evaluation benchmark for cross-lingual text representations. Pairs of text are labeled with textual entailment annotations for 15 different languages (including both high-resource language such as English and low-resource languages such as Swahili).
#### Fine-tuning on XNLI
This example code fine-tunes mBERT (multi-lingual BERT) on the XNLI dataset. It runs in 106 mins
on a single tesla V100 16GB. The data for XNLI can be downloaded with the following links and should be both saved (and un-zipped) in a
`$XNLI_DIR` directory.
* [XNLI 1.0](https://www.nyu.edu/projects/bowman/xnli/XNLI-1.0.zip)
* [XNLI-MT 1.0](https://www.nyu.edu/projects/bowman/xnli/XNLI-MT-1.0.zip)
```bash
export XNLI_DIR=/path/to/XNLI
python run_xnli.py \
--model_type bert \
--model_name_or_path bert-base-multilingual-cased \
--language de \
--train_language en \
--do_train \
--do_eval \
--data_dir $XNLI_DIR \
--per_gpu_train_batch_size 32 \
--learning_rate 5e-5 \
--num_train_epochs 2.0 \
--max_seq_length 128 \
--output_dir /tmp/debug_xnli/ \
--save_steps -1
```
Training with the previously defined hyper-parameters yields the following results on the **test** set:
```bash
acc = 0.7093812375249501
```
## MM-IMDb
Based on the script [`run_mmimdb.py`](https://github.com/huggingface/transformers/blob/master/examples/mm-imdb/run_mmimdb.py).
[MM-IMDb](http://lisi1.unal.edu.co/mmimdb/) is a Multimodal dataset with around 26,000 movies including images, plots and other metadata.
### Training on MM-IMDb
```
python run_mmimdb.py \
--data_dir /path/to/mmimdb/dataset/ \
--model_type bert \
--model_name_or_path bert-base-uncased \
--output_dir /path/to/save/dir/ \
--do_train \
--do_eval \
--max_seq_len 512 \
--gradient_accumulation_steps 20 \
--num_image_embeds 3 \
--num_train_epochs 100 \
--patience 5
```
## Adversarial evaluation of model performances
Here is an example on evaluating a model using adversarial evaluation of natural language inference with the Heuristic Analysis for NLI Systems (HANS) dataset [McCoy et al., 2019](https://arxiv.org/abs/1902.01007). The example was gracefully provided by [Nafise Sadat Moosavi](https://github.com/ns-moosavi).
The HANS dataset can be downloaded from [this location](https://github.com/tommccoy1/hans).
This is an example of using test_hans.py:
```bash
export HANS_DIR=path-to-hans
export MODEL_TYPE=type-of-the-model-e.g.-bert-roberta-xlnet-etc
export MODEL_PATH=path-to-the-model-directory-that-is-trained-on-NLI-e.g.-by-using-run_glue.py
python examples/hans/test_hans.py \
--task_name hans \
--model_type $MODEL_TYPE \
--do_eval \
--data_dir $HANS_DIR \
--model_name_or_path $MODEL_PATH \
--max_seq_length 128 \
--output_dir $MODEL_PATH \
```
This will create the hans_predictions.txt file in MODEL_PATH, which can then be evaluated using hans/evaluate_heur_output.py from the HANS dataset.
The results of the BERT-base model that is trained on MNLI using batch size 8 and the random seed 42 on the HANS dataset is as follows:
```bash
Heuristic entailed results:
lexical_overlap: 0.9702
subsequence: 0.9942
constituent: 0.9962
Heuristic non-entailed results:
lexical_overlap: 0.199
subsequence: 0.0396
constituent: 0.118
```

38
examples/adversarial/README.md Normal file
View File

@ -0,0 +1,38 @@
## Adversarial evaluation of model performances
Here is an example on evaluating a model using adversarial evaluation of natural language inference with the Heuristic Analysis for NLI Systems (HANS) dataset [McCoy et al., 2019](https://arxiv.org/abs/1902.01007). The example was gracefully provided by [Nafise Sadat Moosavi](https://github.com/ns-moosavi).
The HANS dataset can be downloaded from [this location](https://github.com/tommccoy1/hans).
This is an example of using test_hans.py:
```bash
export HANS_DIR=path-to-hans
export MODEL_TYPE=type-of-the-model-e.g.-bert-roberta-xlnet-etc
export MODEL_PATH=path-to-the-model-directory-that-is-trained-on-NLI-e.g.-by-using-run_glue.py
python examples/hans/test_hans.py \
--task_name hans \
--model_type $MODEL_TYPE \
--do_eval \
--data_dir $HANS_DIR \
--model_name_or_path $MODEL_PATH \
--max_seq_length 128 \
--output_dir $MODEL_PATH \
```
This will create the hans_predictions.txt file in MODEL_PATH, which can then be evaluated using hans/evaluate_heur_output.py from the HANS dataset.
The results of the BERT-base model that is trained on MNLI using batch size 8 and the random seed 42 on the HANS dataset is as follows:
```bash
Heuristic entailed results:
lexical_overlap: 0.9702
subsequence: 0.9942
constituent: 0.9962
Heuristic non-entailed results:
lexical_overlap: 0.199
subsequence: 0.0396
constituent: 0.118
```

0
examples/hans/hans_processors.py → examples/adversarial/hans_processors.py
View File

0
examples/hans/test_hans.py → examples/adversarial/test_hans.py
View File

0
examples/hans/utils_hans.py → examples/adversarial/utils_hans.py
View File

0
examples/run_bertology.py → examples/bertology/run_bertology.py
View File

23
examples/contrib/mm-imdb/README.md Normal file
View File

@ -0,0 +1,23 @@
## MM-IMDb
Based on the script [`run_mmimdb.py`](https://github.com/huggingface/transformers/blob/master/examples/contrib/mm-imdb/run_mmimdb.py).
[MM-IMDb](http://lisi1.unal.edu.co/mmimdb/) is a Multimodal dataset with around 26,000 movies including images, plots and other metadata.
### Training on MM-IMDb
```
python run_mmimdb.py \
--data_dir /path/to/mmimdb/dataset/ \
--model_type bert \
--model_name_or_path bert-base-uncased \
--output_dir /path/to/save/dir/ \
--do_train \
--do_eval \
--max_seq_len 512 \
--gradient_accumulation_steps 20 \
--num_image_embeds 3 \
--num_train_epochs 100 \
--patience 5
```

0
examples/mm-imdb/run_mmimdb.py → examples/contrib/mm-imdb/run_mmimdb.py
View File

0
examples/mm-imdb/utils_mmimdb.py → examples/contrib/mm-imdb/utils_mmimdb.py
View File

9
examples/glue/README.md
View File

@ -1,9 +0,0 @@
# GLUE Benchmark
Based on the script [`run_glue.py`](https://github.com/huggingface/transformers/blob/master/examples/run_glue.py).
#### Run PyTorch version using PyTorch-Lightning
Run `bash run_pl.sh` from the `glue` directory. This will also install `pytorch-lightning` and the requirements in `examples/requirements.txt`. It is a shell pipeline that will automatically download, pre-process the data and run the specified models. Logs are saved in `lightning_logs` directory.
Pass `--n_gpu` flag to change the number of GPUs. Default uses 1. At the end, the expected results are: `TEST RESULTS {'val_loss': tensor(0.0707), 'precision': 0.852427800698191, 'recall': 0.869537067011978, 'f1': 0.8608974358974358}`

63
examples/language-modeling/README.md Normal file
View File

@ -0,0 +1,63 @@
## Language model training
Based on the script [`run_language_modeling.py`](https://github.com/huggingface/transformers/blob/master/examples/run_language_modeling.py).
Fine-tuning (or training from scratch) the library models for language modeling on a text dataset for GPT, GPT-2, BERT and RoBERTa (DistilBERT
to be added soon). GPT and GPT-2 are fine-tuned using a causal language modeling (CLM) loss while BERT and RoBERTa
are fine-tuned using a masked language modeling (MLM) loss.
Before running the following example, you should get a file that contains text on which the language model will be
trained or fine-tuned. A good example of such text is the [WikiText-2 dataset](https://blog.einstein.ai/the-wikitext-long-term-dependency-language-modeling-dataset/).
We will refer to two different files: `$TRAIN_FILE`, which contains text for training, and `$TEST_FILE`, which contains
text that will be used for evaluation.
### GPT-2/GPT and causal language modeling
The following example fine-tunes GPT-2 on WikiText-2. We're using the raw WikiText-2 (no tokens were replaced before
the tokenization). The loss here is that of causal language modeling.
```bash
export TRAIN_FILE=/path/to/dataset/wiki.train.raw
export TEST_FILE=/path/to/dataset/wiki.test.raw
python run_language_modeling.py \
--output_dir=output \
--model_type=gpt2 \
--model_name_or_path=gpt2 \
--do_train \
--train_data_file=$TRAIN_FILE \
--do_eval \
--eval_data_file=$TEST_FILE
```
This takes about half an hour to train on a single K80 GPU and about one minute for the evaluation to run. It reaches
a score of ~20 perplexity once fine-tuned on the dataset.
### RoBERTa/BERT and masked language modeling
The following example fine-tunes RoBERTa on WikiText-2. Here too, we're using the raw WikiText-2. The loss is different
as BERT/RoBERTa have a bidirectional mechanism; we're therefore using the same loss that was used during their
pre-training: masked language modeling.
In accordance to the RoBERTa paper, we use dynamic masking rather than static masking. The model may, therefore, converge
slightly slower (over-fitting takes more epochs).
We use the `--mlm` flag so that the script may change its loss function.
```bash
export TRAIN_FILE=/path/to/dataset/wiki.train.raw
export TEST_FILE=/path/to/dataset/wiki.test.raw
python run_language_modeling.py \
--output_dir=output \
--model_type=roberta \
--model_name_or_path=roberta-base \
--do_train \
--train_data_file=$TRAIN_FILE \
--do_eval \
--eval_data_file=$TEST_FILE \
--mlm
```

0
examples/run_language_modeling.py → examples/language-modeling/run_language_modeling.py
View File

0
examples/transformer_base.py → examples/lightning_base.py
View File

31
examples/multiple-choice/README.md Normal file
View File

@ -0,0 +1,31 @@
## Multiple Choice
Based on the script [`run_multiple_choice.py`]().
#### Fine-tuning on SWAG
Download [swag](https://github.com/rowanz/swagaf/tree/master/data) data
```bash
#training on 4 tesla V100(16GB) GPUS
export SWAG_DIR=/path/to/swag_data_dir
python ./examples/run_multiple_choice.py \
--task_name swag \
--model_name_or_path roberta-base \
--do_train \
--do_eval \
--data_dir $SWAG_DIR \
--learning_rate 5e-5 \
--num_train_epochs 3 \
--max_seq_length 80 \
--output_dir models_bert/swag_base \
--per_gpu_eval_batch_size=16 \
--per_gpu_train_batch_size=16 \
--gradient_accumulation_steps 2 \
--overwrite_output
```
Training with the defined hyper-parameters yields the following results:
```
***** Eval results *****
eval_acc = 0.8338998300509847
eval_loss = 0.44457291918821606
```

0
examples/run_multiple_choice.py → examples/multiple-choice/run_multiple_choice.py
View File

0
examples/utils_multiple_choice.py → examples/multiple-choice/utils_multiple_choice.py
View File

159
examples/question-answering/README.md Normal file
View File

@ -0,0 +1,159 @@
## SQuAD
Based on the script [`run_squad.py`](https://github.com/huggingface/transformers/blob/master/examples/run_squad.py).
#### Fine-tuning BERT on SQuAD1.0
This example code fine-tunes BERT on the SQuAD1.0 dataset. It runs in 24 min (with BERT-base) or 68 min (with BERT-large)
on a single tesla V100 16GB. 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)
* [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)
And for SQuAD2.0, you need to download:
- [train-v2.0.json](https://rajpurkar.github.io/SQuAD-explorer/dataset/train-v2.0.json)
- [dev-v2.0.json](https://rajpurkar.github.io/SQuAD-explorer/dataset/dev-v2.0.json)
- [evaluate-v2.0.py](https://worksheets.codalab.org/rest/bundles/0x6b567e1cf2e041ec80d7098f031c5c9e/contents/blob/)
```bash
export SQUAD_DIR=/path/to/SQUAD
python run_squad.py \
--model_type bert \
--model_name_or_path bert-base-uncased \
--do_train \
--do_eval \
--train_file $SQUAD_DIR/train-v1.1.json \
--predict_file $SQUAD_DIR/dev-v1.1.json \
--per_gpu_train_batch_size 12 \
--learning_rate 3e-5 \
--num_train_epochs 2.0 \
--max_seq_length 384 \
--doc_stride 128 \
--output_dir /tmp/debug_squad/
```
Training with the previously defined hyper-parameters yields the following results:
```bash
f1 = 88.52
exact_match = 81.22
```
#### Distributed training
Here is an example using distributed training on 8 V100 GPUs and Bert Whole Word Masking uncased model to reach a F1 > 93 on SQuAD1.1:
```bash
python -m torch.distributed.launch --nproc_per_node=8 ./examples/run_squad.py \
--model_type bert \
--model_name_or_path bert-large-uncased-whole-word-masking \
--do_train \
--do_eval \
--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 ./examples/models/wwm_uncased_finetuned_squad/ \
--per_gpu_eval_batch_size=3 \
--per_gpu_train_batch_size=3 \
```
Training with the previously defined hyper-parameters yields the following results:
```bash
f1 = 93.15
exact_match = 86.91
```
This fine-tuned model is available as a checkpoint under the reference
`bert-large-uncased-whole-word-masking-finetuned-squad`.
#### Fine-tuning XLNet on SQuAD
This example code fine-tunes XLNet on both SQuAD1.0 and SQuAD2.0 dataset. See above to download the data for SQuAD .
##### Command for SQuAD1.0:
```bash
export SQUAD_DIR=/path/to/SQUAD
python run_squad.py \
--model_type xlnet \
--model_name_or_path xlnet-large-cased \
--do_train \
--do_eval \
--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 ./wwm_cased_finetuned_squad/ \
--per_gpu_eval_batch_size=4 \
--per_gpu_train_batch_size=4 \
--save_steps 5000
```
##### Command for SQuAD2.0:
```bash
export SQUAD_DIR=/path/to/SQUAD
python run_squad.py \
--model_type xlnet \
--model_name_or_path xlnet-large-cased \
--do_train \
--do_eval \
--version_2_with_negative \
--train_file $SQUAD_DIR/train-v2.0.json \
--predict_file $SQUAD_DIR/dev-v2.0.json \
--learning_rate 3e-5 \
--num_train_epochs 4 \
--max_seq_length 384 \
--doc_stride 128 \
--output_dir ./wwm_cased_finetuned_squad/ \
--per_gpu_eval_batch_size=2 \
--per_gpu_train_batch_size=2 \
--save_steps 5000
```
Larger batch size may improve the performance while costing more memory.
##### Results for SQuAD1.0 with the previously defined hyper-parameters:
```python
{
"exact": 85.45884578997162,
"f1": 92.5974600601065,
"total": 10570,
"HasAns_exact": 85.45884578997162,
"HasAns_f1": 92.59746006010651,
"HasAns_total": 10570
}
```
##### Results for SQuAD2.0 with the previously defined hyper-parameters:
```python
{
"exact": 80.4177545691906,
"f1": 84.07154997729623,
"total": 11873,
"HasAns_exact": 76.73751686909581,
"HasAns_f1": 84.05558584352873,
"HasAns_total": 5928,
"NoAns_exact": 84.0874684608915,
"NoAns_f1": 84.0874684608915,
"NoAns_total": 5945
}
```

0
examples/run_squad.py → examples/question-answering/run_squad.py
View File

1
examples/requirements.txt
View File

@ -1,4 +1,3 @@
tensorboardX
tensorboard tensorboard
scikit-learn scikit-learn
seqeval seqeval

611
examples/run_tpu_glue.py
View File

@ -1,611 +0,0 @@
# coding=utf-8
# Copyright 2019 The Google AI Language Team Authors and The HuggingFace Inc. team.
# Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
""" Finetuning the library models for sequence classification on GLUE (Bert, DistilBert, XLNet, RoBERTa)."""
from __future__ import absolute_import, division, print_function
import argparse
import glob
import logging
import os
import random
import numpy as np
import torch
import torch_xla.core.xla_model as xm
import torch_xla.debug.metrics as met
import torch_xla.distributed.parallel_loader as pl
import torch_xla.distributed.xla_multiprocessing as xmp
from torch.utils.data import DataLoader, RandomSampler, TensorDataset
from torch.utils.data.distributed import DistributedSampler
from tqdm import tqdm, trange
from transformers import (
WEIGHTS_NAME,
AdamW,
BertConfig,
BertForSequenceClassification,
BertTokenizer,
DistilBertConfig,
DistilBertForSequenceClassification,
DistilBertTokenizer,
RobertaConfig,
RobertaForSequenceClassification,
RobertaTokenizer,
XLMConfig,
XLMForSequenceClassification,
XLMTokenizer,
XLNetConfig,
XLNetForSequenceClassification,
XLNetTokenizer,
get_linear_schedule_with_warmup,
)
from transformers import glue_compute_metrics as compute_metrics
from transformers import glue_convert_examples_to_features as convert_examples_to_features
from transformers import glue_output_modes as output_modes
from transformers import glue_processors as processors
try:
# Only tensorboardX supports writing directly to gs://
from tensorboardX import SummaryWriter
except ImportError:
from torch.utils.tensorboard import SummaryWriter
logger = logging.getLogger(__name__)
ALL_MODELS = sum(
(
tuple(conf.pretrained_config_archive_map.keys())
for conf in (BertConfig, XLNetConfig, XLMConfig, RobertaConfig, DistilBertConfig)
),
(),
)
MODEL_CLASSES = {
"bert": (BertConfig, BertForSequenceClassification, BertTokenizer),
"xlnet": (XLNetConfig, XLNetForSequenceClassification, XLNetTokenizer),
"xlm": (XLMConfig, XLMForSequenceClassification, XLMTokenizer),
"roberta": (RobertaConfig, RobertaForSequenceClassification, RobertaTokenizer),
"distilbert": (DistilBertConfig, DistilBertForSequenceClassification, DistilBertTokenizer),
}
def set_seed(seed):
random.seed(seed)
np.random.seed(seed)
torch.manual_seed(seed)
def get_sampler(dataset):
if xm.xrt_world_size() <= 1:
return RandomSampler(dataset)
return DistributedSampler(dataset, num_replicas=xm.xrt_world_size(), rank=xm.get_ordinal())
def train(args, train_dataset, model, tokenizer, disable_logging=False):
""" Train the model """
if xm.is_master_ordinal():
# Only master writes to Tensorboard
tb_writer = SummaryWriter(args.tensorboard_logdir)
train_sampler = get_sampler(train_dataset)
dataloader = DataLoader(train_dataset, sampler=train_sampler, batch_size=args.train_batch_size)
if args.max_steps > 0:
t_total = args.max_steps
args.num_train_epochs = args.max_steps // (len(dataloader) // args.gradient_accumulation_steps) + 1
else:
t_total = len(dataloader) // args.gradient_accumulation_steps * args.num_train_epochs
# Prepare optimizer and schedule (linear warmup and decay)
no_decay = ["bias", "LayerNorm.weight"]
optimizer_grouped_parameters = [
{
"params": [p for n, p in model.named_parameters() if not any(nd in n for nd in no_decay)],
"weight_decay": args.weight_decay,
},
{"params": [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)], "weight_decay": 0.0},
]
optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
scheduler = get_linear_schedule_with_warmup(
optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total,
)
# Train!
logger.info("***** Running training *****")
logger.info(" Num examples = %d", len(dataloader) * args.train_batch_size)
logger.info(" Num Epochs = %d", args.num_train_epochs)
logger.info(" Instantaneous batch size per TPU core = %d", args.train_batch_size)
logger.info(
" Total train batch size (w. parallel, distributed & accumulation) = %d",
(args.train_batch_size * args.gradient_accumulation_steps * xm.xrt_world_size()),
)
logger.info(" Gradient Accumulation steps = %d", args.gradient_accumulation_steps)
logger.info(" Total optimization steps = %d", t_total)
global_step = 0
loss = None
model.zero_grad()
train_iterator = trange(int(args.num_train_epochs), desc="Epoch", disable=disable_logging)
set_seed(args.seed) # Added here for reproductibility (even between python 2 and 3)
for epoch in train_iterator:
# tpu-comment: Get TPU parallel loader which sends data to TPU in background.
train_dataloader = pl.ParallelLoader(dataloader, [args.device]).per_device_loader(args.device)
epoch_iterator = tqdm(train_dataloader, desc="Iteration", total=len(dataloader), disable=disable_logging)
for step, batch in enumerate(epoch_iterator):
# Save model checkpoint.
if args.save_steps > 0 and global_step % args.save_steps == 0:
output_dir = os.path.join(args.output_dir, "checkpoint-{}".format(global_step))
logger.info("Saving model checkpoint to %s", output_dir)
if xm.is_master_ordinal():
if not os.path.exists(output_dir):
os.makedirs(output_dir)
torch.save(args, os.path.join(output_dir, "training_args.bin"))
# Barrier to wait for saving checkpoint.
xm.rendezvous("mid_training_checkpoint")
# model.save_pretrained needs to be called by all ordinals
model.save_pretrained(output_dir)
model.train()
inputs = {"input_ids": batch[0], "attention_mask": batch[1], "labels": batch[3]}
if args.model_type != "distilbert":
# XLM, DistilBERT and RoBERTa don't use segment_ids
inputs["token_type_ids"] = batch[2] if args.model_type in ["bert", "xlnet"] else None
outputs = model(**inputs)
loss = outputs[0] # model outputs are always tuple in transformers (see doc)
if args.gradient_accumulation_steps > 1:
loss = loss / args.gradient_accumulation_steps
loss.backward()
if (step + 1) % args.gradient_accumulation_steps == 0:
torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_grad_norm)
xm.optimizer_step(optimizer)
scheduler.step() # Update learning rate schedule
model.zero_grad()
global_step += 1
if args.logging_steps > 0 and global_step % args.logging_steps == 0:
# Log metrics.
results = {}
if args.evaluate_during_training:
results = evaluate(args, model, tokenizer, disable_logging=disable_logging)
loss_scalar = loss.item()
logger.info(
"global_step: {global_step}, lr: {lr:.6f}, loss: {loss:.3f}".format(
global_step=global_step, lr=scheduler.get_lr()[0], loss=loss_scalar
)
)
if xm.is_master_ordinal():
# tpu-comment: All values must be in CPU and not on TPU device
for key, value in results.items():
tb_writer.add_scalar("eval_{}".format(key), value, global_step)
tb_writer.add_scalar("lr", scheduler.get_lr()[0], global_step)
tb_writer.add_scalar("loss", loss_scalar, global_step)
if args.max_steps > 0 and global_step > args.max_steps:
epoch_iterator.close()
break
if args.metrics_debug:
# tpu-comment: Logging debug metrics for PyTorch/XLA (compile, execute times, ops, etc.)
xm.master_print(met.metrics_report())
if args.max_steps > 0 and global_step > args.max_steps:
train_iterator.close()
break
if xm.is_master_ordinal():
tb_writer.close()
return global_step, loss.item()
def evaluate(args, model, tokenizer, prefix="", disable_logging=False):
"""Evaluate the model"""
if xm.is_master_ordinal():
# Only master writes to Tensorboard
tb_writer = SummaryWriter(args.tensorboard_logdir)
# Loop to handle MNLI double evaluation (matched, mis-matched)
eval_task_names = ("mnli", "mnli-mm") if args.task_name == "mnli" else (args.task_name,)
eval_outputs_dirs = (args.output_dir, args.output_dir + "-MM") if args.task_name == "mnli" else (args.output_dir,)
results = {}
for eval_task, eval_output_dir in zip(eval_task_names, eval_outputs_dirs):
eval_dataset = load_and_cache_examples(args, eval_task, tokenizer, evaluate=True)
eval_sampler = get_sampler(eval_dataset)
if not os.path.exists(eval_output_dir):
os.makedirs(eval_output_dir)
dataloader = DataLoader(eval_dataset, sampler=eval_sampler, batch_size=args.eval_batch_size, shuffle=False)
eval_dataloader = pl.ParallelLoader(dataloader, [args.device]).per_device_loader(args.device)
# Eval!
logger.info("***** Running evaluation {} *****".format(prefix))
logger.info(" Num examples = %d", len(dataloader) * args.eval_batch_size)
logger.info(" Batch size = %d", args.eval_batch_size)
eval_loss = 0.0
nb_eval_steps = 0
preds = None
out_label_ids = None
for batch in tqdm(eval_dataloader, desc="Evaluating", disable=disable_logging):
model.eval()
with torch.no_grad():
inputs = {"input_ids": batch[0], "attention_mask": batch[1], "labels": batch[3]}
if args.model_type != "distilbert":
# XLM, DistilBERT and RoBERTa don't use segment_ids
inputs["token_type_ids"] = batch[2] if args.model_type in ["bert", "xlnet"] else None
outputs = model(**inputs)
batch_eval_loss, logits = outputs[:2]
eval_loss += batch_eval_loss
nb_eval_steps += 1
if preds is None:
preds = logits.detach().cpu().numpy()
out_label_ids = inputs["labels"].detach().cpu().numpy()
else:
preds = np.append(preds, logits.detach().cpu().numpy(), axis=0)
out_label_ids = np.append(out_label_ids, inputs["labels"].detach().cpu().numpy(), axis=0)
# tpu-comment: Get all predictions and labels from all worker shards of eval dataset
preds = xm.mesh_reduce("eval_preds", preds, np.concatenate)
out_label_ids = xm.mesh_reduce("eval_out_label_ids", out_label_ids, np.concatenate)
eval_loss = eval_loss / nb_eval_steps
if args.output_mode == "classification":
preds = np.argmax(preds, axis=1)
elif args.output_mode == "regression":
preds = np.squeeze(preds)
result = compute_metrics(eval_task, preds, out_label_ids)
results.update(result)
results["eval_loss"] = eval_loss.item()
output_eval_file = os.path.join(eval_output_dir, prefix, "eval_results.txt")
if xm.is_master_ordinal():
with open(output_eval_file, "w") as writer:
logger.info("***** Eval results {} *****".format(prefix))
for key in sorted(results.keys()):
logger.info(" %s = %s", key, str(results[key]))
writer.write("%s = %s\n" % (key, str(results[key])))
tb_writer.add_scalar(f"{eval_task}/{key}", results[key])
if args.metrics_debug:
# tpu-comment: Logging debug metrics for PyTorch/XLA (compile, execute times, ops, etc.)
xm.master_print(met.metrics_report())
if xm.is_master_ordinal():
tb_writer.close()
return results
def load_and_cache_examples(args, task, tokenizer, evaluate=False):
if not xm.is_master_ordinal():
xm.rendezvous("load_and_cache_examples")
processor = processors[task]()
output_mode = output_modes[task]
cached_features_file = os.path.join(
args.cache_dir,
"cached_{}_{}_{}_{}".format(
"dev" if evaluate else "train",
list(filter(None, args.model_name_or_path.split("/"))).pop(),
str(args.max_seq_length),
str(task),
),
)
# Load data features from cache or dataset file
if os.path.exists(cached_features_file) and not args.overwrite_cache:
logger.info("Loading features from cached file %s", cached_features_file)
features = torch.load(cached_features_file)
else:
logger.info("Creating features from dataset file at %s", args.data_dir)
label_list = processor.get_labels()
if task in ["mnli", "mnli-mm"] and args.model_type in ["roberta"]:
# HACK(label indices are swapped in RoBERTa pretrained model)
label_list[1], label_list[2] = label_list[2], label_list[1]
examples = (
processor.get_dev_examples(args.data_dir) if evaluate else processor.get_train_examples(args.data_dir)
)
features = convert_examples_to_features(
examples, tokenizer, max_length=args.max_seq_length, label_list=label_list, output_mode=output_mode,
)
logger.info("Saving features into cached file %s", cached_features_file)
torch.save(features, cached_features_file)
if xm.is_master_ordinal():
xm.rendezvous("load_and_cache_examples")
# Convert to Tensors and build dataset
all_input_ids = torch.tensor([f.input_ids for f in features], dtype=torch.long)
all_attention_mask = torch.tensor([f.attention_mask for f in features], dtype=torch.long)
all_token_type_ids = torch.tensor([f.token_type_ids for f in features], dtype=torch.long)
if output_mode == "classification":
all_labels = torch.tensor([f.label for f in features], dtype=torch.long)
elif output_mode == "regression":
all_labels = torch.tensor([f.label for f in features], dtype=torch.float)
dataset = TensorDataset(all_input_ids, all_attention_mask, all_token_type_ids, all_labels)
return dataset
def main(args):
if (
os.path.exists(args.output_dir)
and os.listdir(args.output_dir)
and args.do_train
and not args.overwrite_output_dir
):
raise ValueError(
(
"Output directory ({}) already exists and is not empty." " Use --overwrite_output_dir to overcome."
).format(args.output_dir)
)
# tpu-comment: Get TPU/XLA Device
args.device = xm.xla_device()
# Setup logging
logging.basicConfig(
format="[xla:{}] %(asctime)s - %(levelname)s - %(name)s - %(message)s".format(xm.get_ordinal()),
datefmt="%m/%d/%Y %H:%M:%S",
level=logging.INFO,
)
disable_logging = False
if not xm.is_master_ordinal() and args.only_log_master:
# Disable all non-master loggers below CRITICAL.
logging.disable(logging.CRITICAL)
disable_logging = True
logger.warning("Process rank: %s, device: %s, num_cores: %s", xm.get_ordinal(), args.device, args.num_cores)
# Set seed to have same initialization
set_seed(args.seed)
# Prepare GLUE task
args.task_name = args.task_name.lower()
if args.task_name not in processors:
raise ValueError("Task not found: %s" % (args.task_name))
processor = processors[args.task_name]()
args.output_mode = output_modes[args.task_name]
label_list = processor.get_labels()
num_labels = len(label_list)
if not xm.is_master_ordinal():
xm.rendezvous(
"download_only_once"
) # Make sure only the first process in distributed training will download model & vocab
# Load pretrained model and tokenizer
args.model_type = args.model_type.lower()
config_class, model_class, tokenizer_class = MODEL_CLASSES[args.model_type]
config = config_class.from_pretrained(
args.config_name if args.config_name else args.model_name_or_path,
num_labels=num_labels,
finetuning_task=args.task_name,
cache_dir=args.cache_dir if args.cache_dir else None,
xla_device=True,
)
tokenizer = tokenizer_class.from_pretrained(
args.tokenizer_name if args.tokenizer_name else args.model_name_or_path,
do_lower_case=args.do_lower_case,
cache_dir=args.cache_dir if args.cache_dir else None,
)
model = model_class.from_pretrained(
args.model_name_or_path,
from_tf=bool(".ckpt" in args.model_name_or_path),
config=config,
cache_dir=args.cache_dir if args.cache_dir else None,
)
if xm.is_master_ordinal():
xm.rendezvous("download_only_once")
# Send model to TPU/XLA device.
model.to(args.device)
logger.info("Training/evaluation parameters %s", args)
if args.do_train:
# Train the model.
train_dataset = load_and_cache_examples(args, args.task_name, tokenizer, evaluate=False)
global_step, tr_loss = train(args, train_dataset, model, tokenizer, disable_logging=disable_logging)
logger.info(" global_step = %s, average loss = %s", global_step, tr_loss)
if xm.is_master_ordinal():
# Save trained model.
# Saving best-practices: if you use defaults names for the model, you can reload it using from_pretrained()
# Create output directory if needed
if not os.path.exists(args.output_dir):
os.makedirs(args.output_dir)
logger.info("Saving model checkpoint to %s", args.output_dir)
# Save a trained model, configuration and tokenizer using `save_pretrained()`.
# They can then be reloaded using `from_pretrained()`
tokenizer.save_pretrained(args.output_dir)
# Good practice: save your training arguments together with the trained.
torch.save(args, os.path.join(args.output_dir, "training_args.bin"))
xm.rendezvous("post_training_checkpoint")
# model.save_pretrained needs to be called by all ordinals
model.save_pretrained(args.output_dir)
# Load a trained model and vocabulary that you have fine-tuned
model = model_class.from_pretrained(args.output_dir)
tokenizer = tokenizer_class.from_pretrained(args.output_dir)
model.to(args.device)
# Evaluation
results = {}
if args.do_eval:
tokenizer = tokenizer_class.from_pretrained(args.output_dir, do_lower_case=args.do_lower_case)
checkpoints = [args.output_dir]
if args.eval_all_checkpoints:
checkpoints = list(
os.path.dirname(c) for c in sorted(glob.glob(args.output_dir + "/**/" + WEIGHTS_NAME, recursive=True))
)
logging.getLogger("transformers.modeling_utils").setLevel(logging.WARN) # Reduce logging
logger.info("Evaluate the following checkpoints: %s", checkpoints)
for checkpoint in checkpoints:
global_step = checkpoint.split("-")[-1] if len(checkpoints) > 1 else ""
prefix = checkpoint.split("/")[-1] if checkpoint.find("checkpoint") != -1 else ""
model = model_class.from_pretrained(checkpoint)
model.to(args.device)
result = evaluate(args, model, tokenizer, prefix=prefix, disable_logging=disable_logging)
result = dict((k + "_{}".format(global_step), v) for k, v in result.items())
results.update(result)
return results
def get_args():
parser = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
"--data_dir",
default=None,
type=str,
required=True,
help="The input data dir. Should contain the .tsv files (or other data files) for the task.",
)
parser.add_argument(
"--model_type",
default=None,
type=str,
required=True,
help="Model type selected in the list: " + ", ".join(MODEL_CLASSES.keys()),
)
parser.add_argument(
"--model_name_or_path",
default=None,
type=str,
required=True,
help="Path to pre-trained model or shortcut name selected in the list: " + ", ".join(ALL_MODELS),
)
parser.add_argument(
"--task_name",
default=None,
type=str,
required=True,
help="The name of the task to train selected in the list: " + ", ".join(processors.keys()),
)
parser.add_argument(
"--output_dir",
default=None,
type=str,
required=True,
help="The output directory where the model predictions and checkpoints will be written.",
)
# TPU Parameters
parser.add_argument("--num_cores", default=8, type=int, help="Number of TPU cores to use (1 or 8).")
parser.add_argument("--metrics_debug", action="store_true", help="Whether to print debug metrics.")
# Other parameters
parser.add_argument(
"--config_name", default="", type=str, help="Pretrained config name or path if not the same as model_name"
)
parser.add_argument(
"--tokenizer_name",
default="",
type=str,
help="Pretrained tokenizer name or path if not the same as model_name",
)
parser.add_argument(
"--cache_dir",
default="",
type=str,
help="Where do you want to store the pre-trained models downloaded and features file generated",
)
parser.add_argument(
"--max_seq_length",
default=128,
type=int,
help="The maximum total input sequence length after tokenization. Sequences longer "
"than this will be truncated, sequences shorter will be padded.",
)
parser.add_argument("--do_train", action="store_true", help="Whether to run training.")
parser.add_argument("--do_eval", action="store_true", help="Whether to run eval on the dev set.")
parser.add_argument(
"--evaluate_during_training", action="store_true", help="Rul evaluation during training at each logging step."
)
parser.add_argument(
"--do_lower_case", action="store_true", help="Set this flag if you are using an uncased model."
)
parser.add_argument("--train_batch_size", default=8, type=int, help="Per core batch size for training.")
parser.add_argument("--eval_batch_size", default=8, type=int, help="Per core batch size for evaluation.")
parser.add_argument(
"--gradient_accumulation_steps",
type=int,
default=1,
help="Number of updates steps to accumulate before performing a backward/update pass.",
)
parser.add_argument("--learning_rate", default=5e-5, type=float, help="The initial learning rate for Adam.")
parser.add_argument("--weight_decay", default=0.0, type=float, help="Weight deay if we apply some.")
parser.add_argument("--adam_epsilon", default=1e-8, type=float, help="Epsilon for Adam optimizer.")
parser.add_argument("--max_grad_norm", default=1.0, type=float, help="Max gradient norm.")
parser.add_argument(
"--num_train_epochs", default=3.0, type=float, help="Total number of training epochs to perform."
)
parser.add_argument(
"--max_steps",
default=-1,
type=int,
help="If > 0: set total number of training steps to perform. Override num_train_epochs.",
)
parser.add_argument("--warmup_steps", default=0, type=int, help="Linear warmup over warmup_steps.")
parser.add_argument("--tensorboard_logdir", default="./runs", type=str, help="Where to write tensorboard metrics.")
parser.add_argument("--logging_steps", type=int, default=50, help="Log every X update steps.")
parser.add_argument("--only_log_master", action="store_true", help="Whether to log only from each hosts master.")
parser.add_argument("--save_steps", type=int, default=50, help="Save checkpoint every X update steps.")
parser.add_argument(
"--eval_all_checkpoints",
action="store_true",
help="Evaluate all checkpoints starting with the same prefix as model_name ending and ending with step number",
)
parser.add_argument(
"--overwrite_output_dir", action="store_true", help="Overwrite the content of the output directory"
)
parser.add_argument(
"--overwrite_cache", action="store_true", help="Overwrite the cached training and evaluation sets"
)
parser.add_argument("--seed", type=int, default=42, help="random seed for initialization")
return parser.parse_args()
def _mp_fn(rank, args):
main(args)
def main_cli():
args = get_args()
xmp.spawn(_mp_fn, args=(args,), nprocs=args.num_cores)
if __name__ == "__main__":
main_cli()

2
examples/summarization/bart/finetune.py
View File

@ -7,7 +7,7 @@ import time
import torch import torch
from torch.utils.data import DataLoader from torch.utils.data import DataLoader
from transformer_base import BaseTransformer, add_generic_args, generic_train, get_linear_schedule_with_warmup from lightning_base import BaseTransformer, add_generic_args, generic_train, get_linear_schedule_with_warmup
try: try:

2
examples/summarization/bart/run_train.sh
View File

@ -5,7 +5,7 @@ export OUTPUT_DIR=${CURRENT_DIR}/${OUTPUT_DIR_NAME}
# Make output directory if it doesn't exist # Make output directory if it doesn't exist
mkdir -p $OUTPUT_DIR mkdir -p $OUTPUT_DIR
# Add parent directory to python path to access transformer_base.py # Add parent directory to python path to access lightning_base.py
export PYTHONPATH="../../":"${PYTHONPATH}" export PYTHONPATH="../../":"${PYTHONPATH}"
python finetune.py \ python finetune.py \

2
examples/summarization/bart/run_train_tiny.sh
View File

@ -12,7 +12,7 @@ export OUTPUT_DIR=${CURRENT_DIR}/${OUTPUT_DIR_NAME}
# Make output directory if it doesn't exist # Make output directory if it doesn't exist
mkdir -p $OUTPUT_DIR mkdir -p $OUTPUT_DIR
# Add parent directory to python path to access transformer_base.py and utils.py # Add parent directory to python path to access lightning_base.py and utils.py
export PYTHONPATH="../../":"${PYTHONPATH}" export PYTHONPATH="../../":"${PYTHONPATH}"
python finetune.py \ python finetune.py \
--data_dir=cnn_tiny/ \ --data_dir=cnn_tiny/ \

85
examples/test_examples.py
View File

@ -16,10 +16,20 @@
import argparse import argparse
import logging import logging
import os
import sys import sys
import unittest import unittest
from unittest.mock import patch from unittest.mock import patch
SRC_DIRS = [
os.path.join(os.path.dirname(__file__), dirname)
for dirname in ["text-generation", "text-classification", "language-modeling", "question-answering"]
]
sys.path.extend(SRC_DIRS)
if SRC_DIRS is not None:
import run_generation import run_generation
import run_glue import run_glue
import run_language_modeling import run_language_modeling
@ -43,24 +53,24 @@ class ExamplesTests(unittest.TestCase):
stream_handler = logging.StreamHandler(sys.stdout) stream_handler = logging.StreamHandler(sys.stdout)
logger.addHandler(stream_handler) logger.addHandler(stream_handler)
testargs = [ testargs = """
"run_glue.py", run_glue.py
"--data_dir=./examples/tests_samples/MRPC/", --model_name_or_path bert-base-uncased
"--task_name=mrpc", --data_dir ./tests/fixtures/tests_samples/MRPC/
"--do_train", --task_name mrpc
"--do_eval", --do_train
"--output_dir=./examples/tests_samples/temp_dir", --do_eval
"--per_gpu_train_batch_size=2", --output_dir ./tests/fixtures/tests_samples/temp_dir
"--per_gpu_eval_batch_size=1", --per_gpu_train_batch_size=2
"--learning_rate=1e-4", --per_gpu_eval_batch_size=1
"--max_steps=10", --learning_rate=1e-4
"--warmup_steps=2", --max_steps=10
"--overwrite_output_dir", --warmup_steps=2
"--seed=42", --overwrite_output_dir
"--max_seq_length=128", --seed=42
] --max_seq_length=128
model_name = "--model_name_or_path=bert-base-uncased" """.split()
with patch.object(sys, "argv", testargs + [model_name]): with patch.object(sys, "argv", testargs):
result = run_glue.main() result = run_glue.main()
del result["loss"] del result["loss"]
for value in result.values(): for value in result.values():
@ -78,7 +88,7 @@ class ExamplesTests(unittest.TestCase):
--line_by_line --line_by_line
--train_data_file ./tests/fixtures/sample_text.txt --train_data_file ./tests/fixtures/sample_text.txt
--eval_data_file ./tests/fixtures/sample_text.txt --eval_data_file ./tests/fixtures/sample_text.txt
--output_dir ./tests/fixtures --output_dir ./tests/fixtures/tests_samples/temp_dir
--overwrite_output_dir --overwrite_output_dir
--do_train --do_train
--do_eval --do_eval
@ -93,24 +103,25 @@ class ExamplesTests(unittest.TestCase):
stream_handler = logging.StreamHandler(sys.stdout) stream_handler = logging.StreamHandler(sys.stdout)
logger.addHandler(stream_handler) logger.addHandler(stream_handler)
testargs = [ testargs = """
"run_squad.py", run_squad.py
"--data_dir=./examples/tests_samples/SQUAD", --model_type=bert
"--model_name=bert-base-uncased", --model_name_or_path=bert-base-uncased
"--output_dir=./examples/tests_samples/temp_dir", --data_dir=./tests/fixtures/tests_samples/SQUAD
"--max_steps=10", --model_name=bert-base-uncased
"--warmup_steps=2", --output_dir=./tests/fixtures/tests_samples/temp_dir
"--do_train", --max_steps=10
"--do_eval", --warmup_steps=2
"--version_2_with_negative", --do_train
"--learning_rate=2e-4", --do_eval
"--per_gpu_train_batch_size=2", --version_2_with_negative
"--per_gpu_eval_batch_size=1", --learning_rate=2e-4
"--overwrite_output_dir", --per_gpu_train_batch_size=2
"--seed=42", --per_gpu_eval_batch_size=1
] --overwrite_output_dir
model_type, model_name = ("--model_type=bert", "--model_name_or_path=bert-base-uncased") --seed=42
with patch.object(sys, "argv", testargs + [model_type, model_name]): """.split()
with patch.object(sys, "argv", testargs):
result = run_squad.main() result = run_squad.main()
self.assertGreaterEqual(result["f1"], 30) self.assertGreaterEqual(result["f1"], 30)
self.assertGreaterEqual(result["exact"], 30) self.assertGreaterEqual(result["exact"], 30)

299
examples/text-classification/README.md Normal file
View File

@ -0,0 +1,299 @@
## GLUE Benchmark
# Run TensorFlow 2.0 version
Based on the script [`run_tf_glue.py`](https://github.com/huggingface/transformers/blob/master/examples/run_tf_glue.py).
Fine-tuning the library TensorFlow 2.0 Bert model for sequence classification on the MRPC task of the GLUE benchmark: [General Language Understanding Evaluation](https://gluebenchmark.com/).
This script has an option for mixed precision (Automatic Mixed Precision / AMP) to run models on Tensor Cores (NVIDIA Volta/Turing GPUs) and future hardware and an option for XLA, which uses the XLA compiler to reduce model runtime.
Options are toggled using `USE_XLA` or `USE_AMP` variables in the script.
These options and the below benchmark are provided by @tlkh.
Quick benchmarks from the script (no other modifications):
| GPU | Mode | Time (2nd epoch) | Val Acc (3 runs) |
| --------- | -------- | ----------------------- | ----------------------|
| Titan V | FP32 | 41s | 0.8438/0.8281/0.8333 |
| Titan V | AMP | 26s | 0.8281/0.8568/0.8411 |
| V100 | FP32 | 35s | 0.8646/0.8359/0.8464 |
| V100 | AMP | 22s | 0.8646/0.8385/0.8411 |
| 1080 Ti | FP32 | 55s | - |
Mixed precision (AMP) reduces the training time considerably for the same hardware and hyper-parameters (same batch size was used).
# Run PyTorch version
Based on the script [`run_glue.py`](https://github.com/huggingface/transformers/blob/master/examples/text-classification/run_glue.py).
Fine-tuning the library models for sequence classification on the GLUE benchmark: [General Language Understanding
Evaluation](https://gluebenchmark.com/). This script can fine-tune the following models: BERT, XLM, XLNet and RoBERTa.
GLUE is made up of a total of 9 different tasks. We get the following results on the dev set of the benchmark with an
uncased BERT base model (the checkpoint `bert-base-uncased`). All experiments ran single V100 GPUs with a total train
batch sizes between 16 and 64. Some of these tasks have a small dataset and training can lead to high variance in the results
between different runs. We report the median on 5 runs (with different seeds) for each of the metrics.
| Task | Metric | Result |
|-------|------------------------------|-------------|
| CoLA | Matthew's corr | 49.23 |
| SST-2 | Accuracy | 91.97 |
| MRPC | F1/Accuracy | 89.47/85.29 |
| STS-B | Person/Spearman corr. | 83.95/83.70 |
| QQP | Accuracy/F1 | 88.40/84.31 |
| MNLI | Matched acc./Mismatched acc. | 80.61/81.08 |
| QNLI | Accuracy | 87.46 |
| RTE | Accuracy | 61.73 |
| WNLI | Accuracy | 45.07 |
Some of these results are significantly different from the ones reported on the test set
of GLUE benchmark on the website. For QQP and WNLI, please refer to [FAQ #12](https://gluebenchmark.com/faq) on the webite.
Before running any one 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`.
```bash
export GLUE_DIR=/path/to/glue
export TASK_NAME=MRPC
python run_glue.py \
--model_type bert \
--model_name_or_path bert-base-cased \
--task_name $TASK_NAME \
--do_train \
--do_eval \
--data_dir $GLUE_DIR/$TASK_NAME \
--max_seq_length 128 \
--per_gpu_train_batch_size 32 \
--learning_rate 2e-5 \
--num_train_epochs 3.0 \
--output_dir /tmp/$TASK_NAME/
```
where task name can be one of CoLA, SST-2, MRPC, STS-B, QQP, MNLI, QNLI, RTE, WNLI.
The dev set results will be present within the text file `eval_results.txt` in the specified output_dir.
In case of MNLI, since there are two separate dev sets (matched and mismatched), there will be a separate
output folder called `/tmp/MNLI-MM/` in addition to `/tmp/MNLI/`.
The code has not been tested with half-precision training with apex on any GLUE task apart from MRPC, MNLI,
CoLA, SST-2. The following section provides details on how to run half-precision training with MRPC. With that being
said, there shouldn’t be any issues in running half-precision training with the remaining GLUE tasks as well,
since the data processor for each task inherits from the base class DataProcessor.
## Running on TPUs
You can accelerate your workloads on Google's TPUs. For information on how to setup your TPU environment refer to this
[README](https://github.com/pytorch/xla/blob/master/README.md).
The following are some examples of running the `*_tpu.py` finetuning scripts on TPUs. All steps for data preparation are
identical to your normal GPU + Huggingface setup.
For running your GLUE task on MNLI dataset you can run something like the following:
```
export XRT_TPU_CONFIG="tpu_worker;0;$TPU_IP_ADDRESS:8470"
export GLUE_DIR=/path/to/glue
export TASK_NAME=MNLI
python run_glue_tpu.py \
--model_type bert \
--model_name_or_path bert-base-cased \
--task_name $TASK_NAME \
--do_train \
--do_eval \
--data_dir $GLUE_DIR/$TASK_NAME \
--max_seq_length 128 \
--train_batch_size 32 \
--learning_rate 3e-5 \
--num_train_epochs 3.0 \
--output_dir /tmp/$TASK_NAME \
--overwrite_output_dir \
--logging_steps 50 \
--save_steps 200 \
--num_cores=8 \
--only_log_master
```
### MRPC
#### Fine-tuning example
The following examples fine-tune BERT on the Microsoft Research Paraphrase Corpus (MRPC) corpus and runs in less
than 10 minutes on a single K-80 and in 27 seconds (!) on single tesla V100 16GB with apex installed.
Before running any one of these GLUE tasks you should download the
[GLUE data](https://gluebenchmark.com/tasks) by running
[this script](https://gist.github.com/W4ngatang/60c2bdb54d156a41194446737ce03e2e)
and unpack it to some directory `$GLUE_DIR`.
```bash
export GLUE_DIR=/path/to/glue
python run_glue.py \
--model_name_or_path bert-base-cased \
--task_name MRPC \
--do_train \
--do_eval \
--data_dir $GLUE_DIR/MRPC/ \
--max_seq_length 128 \
--per_gpu_train_batch_size 32 \
--learning_rate 2e-5 \
--num_train_epochs 3.0 \
--output_dir /tmp/mrpc_output/
```
Our test ran on a few seeds with [the original implementation hyper-
parameters](https://github.com/google-research/bert#sentence-and-sentence-pair-classification-tasks) gave evaluation
results between 84% and 88%.
#### Using Apex and mixed-precision
Using Apex and 16 bit precision, the fine-tuning on MRPC only takes 27 seconds. First install
[apex](https://github.com/NVIDIA/apex), then run the following example:
```bash
export GLUE_DIR=/path/to/glue
python run_glue.py \
--model_name_or_path bert-base-cased \
--task_name MRPC \
--do_train \
--do_eval \
--data_dir $GLUE_DIR/MRPC/ \
--max_seq_length 128 \
--per_gpu_train_batch_size 32 \
--learning_rate 2e-5 \
--num_train_epochs 3.0 \
--output_dir /tmp/mrpc_output/ \
--fp16
```
#### Distributed training
Here is an example using distributed training on 8 V100 GPUs. The model used is the BERT whole-word-masking and it
reaches F1 > 92 on MRPC.
```bash
export GLUE_DIR=/path/to/glue
python -m torch.distributed.launch \
--nproc_per_node 8 run_glue.py \
--model_name_or_path bert-base-cased \
--task_name MRPC \
--do_train \
--do_eval \
--data_dir $GLUE_DIR/MRPC/ \
--max_seq_length 128 \
--per_gpu_train_batch_size 8 \
--learning_rate 2e-5 \
--num_train_epochs 3.0 \
--output_dir /tmp/mrpc_output/
```
Training with these hyper-parameters gave us the following results:
```bash
acc = 0.8823529411764706
acc_and_f1 = 0.901702786377709
eval_loss = 0.3418912578906332
f1 = 0.9210526315789473
global_step = 174
loss = 0.07231863956341798
```
### MNLI
The following example uses the BERT-large, uncased, whole-word-masking model and fine-tunes it on the MNLI task.
```bash
export GLUE_DIR=/path/to/glue
python -m torch.distributed.launch \
--nproc_per_node 8 run_glue.py \
--model_name_or_path bert-base-cased \
--task_name mnli \
--do_train \
--do_eval \
--data_dir $GLUE_DIR/MNLI/ \
--max_seq_length 128 \
--per_gpu_train_batch_size 8 \
--learning_rate 2e-5 \
--num_train_epochs 3.0 \
--output_dir output_dir \
```
The results are the following:
```bash
***** Eval results *****
acc = 0.8679706601466992
eval_loss = 0.4911287787382479
global_step = 18408
loss = 0.04755385363816904
***** Eval results *****
acc = 0.8747965825874695
eval_loss = 0.45516540421714036
global_step = 18408
loss = 0.04755385363816904
```
# Run PyTorch version using PyTorch-Lightning
Run `bash run_pl.sh` from the `glue` directory. This will also install `pytorch-lightning` and the requirements in `examples/requirements.txt`. It is a shell pipeline that will automatically download, pre-process the data and run the specified models. Logs are saved in `lightning_logs` directory.
Pass `--n_gpu` flag to change the number of GPUs. Default uses 1. At the end, the expected results are:
```
TEST RESULTS {'val_loss': tensor(0.0707), 'precision': 0.852427800698191, 'recall': 0.869537067011978, 'f1': 0.8608974358974358}
```
# XNLI
Based on the script [`run_xnli.py`](https://github.com/huggingface/transformers/blob/master/examples/run_xnli.py).
[XNLI](https://www.nyu.edu/projects/bowman/xnli/) is crowd-sourced dataset based on [MultiNLI](http://www.nyu.edu/projects/bowman/multinli/). It is an evaluation benchmark for cross-lingual text representations. Pairs of text are labeled with textual entailment annotations for 15 different languages (including both high-resource language such as English and low-resource languages such as Swahili).
#### Fine-tuning on XNLI
This example code fine-tunes mBERT (multi-lingual BERT) on the XNLI dataset. It runs in 106 mins
on a single tesla V100 16GB. The data for XNLI can be downloaded with the following links and should be both saved (and un-zipped) in a
`$XNLI_DIR` directory.
* [XNLI 1.0](https://www.nyu.edu/projects/bowman/xnli/XNLI-1.0.zip)
* [XNLI-MT 1.0](https://www.nyu.edu/projects/bowman/xnli/XNLI-MT-1.0.zip)
```bash
export XNLI_DIR=/path/to/XNLI
python run_xnli.py \
--model_type bert \
--model_name_or_path bert-base-multilingual-cased \
--language de \
--train_language en \
--do_train \
--do_eval \
--data_dir $XNLI_DIR \
--per_gpu_train_batch_size 32 \
--learning_rate 5e-5 \
--num_train_epochs 2.0 \
--max_seq_length 128 \
--output_dir /tmp/debug_xnli/ \
--save_steps -1
```
Training with the previously defined hyper-parameters yields the following results on the **test** set:
```bash
acc = 0.7093812375249501
```

0
examples/run_glue.py → examples/text-classification/run_glue.py
View File

2
examples/glue/run_pl.sh → examples/text-classification/run_pl.sh
View File

@ -20,7 +20,7 @@ export OUTPUT_DIR=${CURRENT_DIR}/${OUTPUT_DIR_NAME}
# Make output directory if it doesn't exist # Make output directory if it doesn't exist
mkdir -p $OUTPUT_DIR mkdir -p $OUTPUT_DIR
# Add parent directory to python path to access transformer_base.py # Add parent directory to python path to access lightning_base.py
export PYTHONPATH="../":"${PYTHONPATH}" export PYTHONPATH="../":"${PYTHONPATH}"
python3 run_pl_glue.py --data_dir $DATA_DIR \ python3 run_pl_glue.py --data_dir $DATA_DIR \

2
examples/glue/run_pl_glue.py → examples/text-classification/run_pl_glue.py
View File

@ -8,7 +8,7 @@ import numpy as np
import torch import torch
from torch.utils.data import DataLoader, TensorDataset from torch.utils.data import DataLoader, TensorDataset
from transformer_base import BaseTransformer, add_generic_args, generic_train from lightning_base import BaseTransformer, add_generic_args, generic_train
from transformers import glue_compute_metrics as compute_metrics from transformers import glue_compute_metrics as compute_metrics
from transformers import glue_convert_examples_to_features as convert_examples_to_features from transformers import glue_convert_examples_to_features as convert_examples_to_features
from transformers import glue_output_modes from transformers import glue_output_modes

0
examples/run_tf_glue.py → examples/text-classification/run_tf_glue.py
View File

2
examples/run_xnli.py → examples/text-classification/run_xnli.py
View File

@ -14,7 +14,7 @@
# See the License for the specific language governing permissions and # See the License for the specific language governing permissions and
# limitations under the License. # limitations under the License.
""" Finetuning multi-lingual models on XNLI (Bert, DistilBERT, XLM). """ Finetuning multi-lingual models on XNLI (Bert, DistilBERT, XLM).
Adapted from `examples/run_glue.py`""" Adapted from `examples/text-classification/run_glue.py`"""
import argparse import argparse

15
examples/text-generation/README.md Normal file
View File

@ -0,0 +1,15 @@
## Language generation
Based on the script [`run_generation.py`](https://github.com/huggingface/transformers/blob/master/examples/run_generation.py).
Conditional text generation using the auto-regressive models of the library: GPT, GPT-2, Transformer-XL, XLNet, CTRL.
A similar script is used for our official demo [Write With Transfomer](https://transformer.huggingface.co), where you
can try out the different models available in the library.
Example usage:
```bash
python run_generation.py \
--model_type=gpt2 \
--model_name_or_path=gpt2
```

0
examples/pplm/README.md → examples/text-generation/pplm/README.md
View File

0
examples/pplm/imgs/headfigure.png → examples/text-generation/pplm/imgs/headfigure.png
View File

Side by Side Swipe Overlay

Before

Width:  |  Height:  |  Size: 653 KiB

After

Width:  |  Height:  |  Size: 653 KiB

0
examples/pplm/imgs/wooly.png → examples/text-generation/pplm/imgs/wooly.png
View File

Side by Side Swipe Overlay

Before

Width:  |  Height:  |  Size: 664 KiB

After

Width:  |  Height:  |  Size: 664 KiB

0
examples/pplm/pplm_classification_head.py → examples/text-generation/pplm/pplm_classification_head.py
View File

0
examples/pplm/run_pplm.py → examples/text-generation/pplm/run_pplm.py
View File

0
examples/pplm/run_pplm_discrim_train.py → examples/text-generation/pplm/run_pplm_discrim_train.py
View File

0
examples/run_generation.py → examples/text-generation/run_generation.py
View File

0
examples/ner/README.md → examples/token-classification/README.md
View File

0
examples/ner/run.sh → examples/token-classification/run.sh
View File

0
examples/ner/run_ner.py → examples/token-classification/run_ner.py
View File

2
examples/ner/run_pl.sh → examples/token-classification/run_pl.sh
View File

@ -27,7 +27,7 @@ export CURRENT_DIR=${PWD}
export OUTPUT_DIR=${CURRENT_DIR}/${OUTPUT_DIR_NAME} export OUTPUT_DIR=${CURRENT_DIR}/${OUTPUT_DIR_NAME}
mkdir -p $OUTPUT_DIR mkdir -p $OUTPUT_DIR
# Add parent directory to python path to access transformer_base.py # Add parent directory to python path to access lightning_base.py
export PYTHONPATH="../":"${PYTHONPATH}" export PYTHONPATH="../":"${PYTHONPATH}"
python3 run_pl_ner.py --data_dir ./ \ python3 run_pl_ner.py --data_dir ./ \

2
examples/ner/run_pl_ner.py → examples/token-classification/run_pl_ner.py
View File

@ -9,7 +9,7 @@ from seqeval.metrics import f1_score, precision_score, recall_score
from torch.nn import CrossEntropyLoss from torch.nn import CrossEntropyLoss
from torch.utils.data import DataLoader, TensorDataset from torch.utils.data import DataLoader, TensorDataset
from transformer_base import BaseTransformer, add_generic_args, generic_train from lightning_base import BaseTransformer, add_generic_args, generic_train
from utils_ner import convert_examples_to_features, get_labels, read_examples_from_file from utils_ner import convert_examples_to_features, get_labels, read_examples_from_file

0
examples/ner/run_tf_ner.py → examples/token-classification/run_tf_ner.py
View File

6
examples/ner/test_ner_examples.py → examples/token-classification/test_ner_examples.py
View File

@ -18,10 +18,10 @@ class ExamplesTests(unittest.TestCase):
testargs = """ testargs = """
--model_name distilbert-base-german-cased --model_name distilbert-base-german-cased
--output_dir ./examples/tests_samples/temp_dir --output_dir ./tests/fixtures/tests_samples/temp_dir
--overwrite_output_dir --overwrite_output_dir
--data_dir ./examples/tests_samples/GermEval --data_dir ./tests/fixtures/tests_samples/GermEval
--labels ./examples/tests_samples/GermEval/labels.txt --labels ./tests/fixtures/tests_samples/GermEval/labels.txt
--max_seq_length 128 --max_seq_length 128
--num_train_epochs 6 --num_train_epochs 6
--logging_steps 1 --logging_steps 1

0
examples/ner/utils_ner.py → examples/token-classification/utils_ner.py
View File

2
model_cards/SparkBeyond/roberta-large-sts-b/README.md
View File

@ -4,7 +4,7 @@
This model is a fine tuned RoBERTA model over STS-B. This model is a fine tuned RoBERTA model over STS-B.
It was trained with these params: It was trained with these params:
!python /content/transformers/examples/run_glue.py \ !python /content/transformers/examples/text-classification/run_glue.py \
--model_type roberta \ --model_type roberta \
--model_name_or_path roberta-large \ --model_name_or_path roberta-large \
--task_name STS-B \ --task_name STS-B \

2
src/transformers/trainer.py
View File

@ -333,7 +333,7 @@ class Trainer:
total_train_batch_size = ( total_train_batch_size = (
self.args.train_batch_size self.args.train_batch_size
* self.args.gradient_accumulation_steps * self.args.gradient_accumulation_steps
* (torch.distributed.get_world_size() if self.args.local_rank != -1 else 1), * (torch.distributed.get_world_size() if self.args.local_rank != -1 else 1)
) )
logger.info("***** Running training *****") logger.info("***** Running training *****")
logger.info(" Num examples = %d", num_examples) logger.info(" Num examples = %d", num_examples)

0
examples/tests_samples/.gitignore → tests/fixtures/tests_samples/.gitignore vendored
View File

0
examples/tests_samples/GermEval/dev.txt → tests/fixtures/tests_samples/GermEval/dev.txt vendored
View File

0
examples/tests_samples/GermEval/labels.txt → tests/fixtures/tests_samples/GermEval/labels.txt vendored
View File

0
examples/tests_samples/GermEval/train.txt → tests/fixtures/tests_samples/GermEval/train.txt vendored
View File

0
examples/tests_samples/MRPC/dev.tsv → tests/fixtures/tests_samples/MRPC/dev.tsv vendored
View File

Can't render this file because it contains an unexpected character in line 3 and column 155.

0
examples/tests_samples/MRPC/train.tsv → tests/fixtures/tests_samples/MRPC/train.tsv vendored
View File

Can't render this file because it contains an unexpected character in line 3 and column 155.

0
examples/tests_samples/SQUAD/dev-v2.0.json → tests/fixtures/tests_samples/SQUAD/dev-v2.0.json vendored
View File

0
examples/tests_samples/SQUAD/train-v2.0.json → tests/fixtures/tests_samples/SQUAD/train-v2.0.json vendored
View File

0
examples/tests_samples/STS-B/dev.tsv → tests/fixtures/tests_samples/STS-B/dev.tsv vendored
View File

0
examples/tests_samples/STS-B/train.tsv → tests/fixtures/tests_samples/STS-B/train.tsv vendored
View File

6
tests/test_trainer.py
View File

@ -28,7 +28,7 @@ class DataCollatorIntegrationTest(unittest.TestCase):
MODEL_ID = "bert-base-cased-finetuned-mrpc" MODEL_ID = "bert-base-cased-finetuned-mrpc"
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID) tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
data_args = GlueDataTrainingArguments( data_args = GlueDataTrainingArguments(
task_name="mrpc", data_dir="./examples/tests_samples/MRPC", overwrite_cache=True task_name="mrpc", data_dir="./tests/fixtures/tests_samples/MRPC", overwrite_cache=True
) )
dataset = GlueDataset(data_args, tokenizer=tokenizer, evaluate=True) dataset = GlueDataset(data_args, tokenizer=tokenizer, evaluate=True)
data_collator = DefaultDataCollator() data_collator = DefaultDataCollator()
@ -39,7 +39,7 @@ class DataCollatorIntegrationTest(unittest.TestCase):
MODEL_ID = "distilroberta-base" MODEL_ID = "distilroberta-base"
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID) tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
data_args = GlueDataTrainingArguments( data_args = GlueDataTrainingArguments(
task_name="sts-b", data_dir="./examples/tests_samples/STS-B", overwrite_cache=True task_name="sts-b", data_dir="./tests/fixtures/tests_samples/STS-B", overwrite_cache=True
) )
dataset = GlueDataset(data_args, tokenizer=tokenizer, evaluate=True) dataset = GlueDataset(data_args, tokenizer=tokenizer, evaluate=True)
data_collator = DefaultDataCollator() data_collator = DefaultDataCollator()
@ -91,7 +91,7 @@ class TrainerIntegrationTest(unittest.TestCase):
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID) tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
model = AutoModelForSequenceClassification.from_pretrained(MODEL_ID) model = AutoModelForSequenceClassification.from_pretrained(MODEL_ID)
data_args = GlueDataTrainingArguments( data_args = GlueDataTrainingArguments(
task_name="mrpc", data_dir="./examples/tests_samples/MRPC", overwrite_cache=True task_name="mrpc", data_dir="./tests/fixtures/tests_samples/MRPC", overwrite_cache=True
) )
eval_dataset = GlueDataset(data_args, tokenizer=tokenizer, evaluate=True) eval_dataset = GlueDataset(data_args, tokenizer=tokenizer, evaluate=True)

4
valohai.yaml
View File

@ -1,13 +1,13 @@
--- ---
- step: - step:
name: Execute python examples/run_glue.py name: Execute python examples/text-classification/run_glue.py
image: pytorch/pytorch:nightly-devel-cuda10.0-cudnn7 image: pytorch/pytorch:nightly-devel-cuda10.0-cudnn7
command: command:
- python /valohai/repository/utils/download_glue_data.py --data_dir=/glue_data - python /valohai/repository/utils/download_glue_data.py --data_dir=/glue_data
- pip install -e . - pip install -e .
- pip install -r examples/requirements.txt - pip install -r examples/requirements.txt
- python examples/run_glue.py --do_train --data_dir=/glue_data/{parameter-value:task_name} {parameters} - python examples/text-classification/run_glue.py --do_train --data_dir=/glue_data/{parameter-value:task_name} {parameters}
parameters: parameters:
- name: model_type - name: model_type
pass-as: --model_type={v} pass-as: --model_type={v}