File size: 23,500 Bytes
7934b29 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 |
{
"nbformat": 4,
"nbformat_minor": 0,
"metadata": {
"colab": {
"name": "GLUE_Benchmark.ipynb",
"provenance": [],
"private_outputs": true,
"collapsed_sections": []
},
"kernelspec": {
"name": "python3",
"display_name": "Python 3"
},
"accelerator": "GPU",
"pycharm": {
"stem_cell": {
"cell_type": "raw",
"source": [],
"metadata": {
"collapsed": false
}
}
}
},
"cells": [
{
"cell_type": "code",
"metadata": {
"id": "o_0K1lsW1dj9",
"colab_type": "code",
"colab": {}
},
"source": [
"\"\"\"\n",
"You can run either this notebook locally (if you have all the dependencies and a GPU) or on Google Colab.\n",
"\n",
"Instructions for setting up Colab are as follows:\n",
"1. Open a new Python 3 notebook.\n",
"2. Import this notebook from GitHub (File -> Upload Notebook -> \"GITHUB\" tab -> copy/paste GitHub URL)\n",
"3. Connect to an instance with a GPU (Runtime -> Change runtime type -> select \"GPU\" for hardware accelerator)\n",
"4. Run this cell to set up dependencies.\n",
"\"\"\"\n",
"# If you're using Google Colab and not running locally, run this cell\n",
"\n",
"# install NeMo\n",
"BRANCH = 'r1.17.0'\n!python -m pip install git+https://github.com/NVIDIA/NeMo.git@$BRANCH#egg=nemo_toolkit[nlp]\n"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"pycharm": {
"name": "#%%\n"
},
"id": "JFWG-jYCfvD7",
"colab_type": "code",
"colab": {}
},
"source": [
"# If you're not using Colab, you might need to upgrade jupyter notebook to avoid the following error:\n",
"# 'ImportError: IProgress not found. Please update jupyter and ipywidgets.'\n",
"\n",
"! pip install ipywidgets\n",
"! jupyter nbextension enable --py widgetsnbextension\n",
"\n",
"# Please restart the kernel after running this cell"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "dzqD2WDFOIN-",
"colab_type": "code",
"colab": {}
},
"source": [
"from nemo.collections import nlp as nemo_nlp\n",
"from nemo.utils.exp_manager import exp_manager\n",
"\n",
"import os\n",
"import wget \n",
"import torch\n",
"import pytorch_lightning as pl\n",
"from omegaconf import OmegaConf"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"metadata": {
"id": "daYw_Xll2ZR9",
"colab_type": "text"
},
"source": [
"In this tutorial, we are going to describe how to finetune a BERT-like model based on [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805) on [GLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understanding](https://openreview.net/pdf?id=rJ4km2R5t7). \n",
"\n",
"# GLUE tasks\n",
"GLUE Benchmark includes 9 natural language understanding tasks:\n",
"\n",
"## Single-Sentence Tasks\n",
"\n",
"* CoLA - [The Corpus of Linguistic Acceptability](https://arxiv.org/abs/1805.12471) is a set of English sentences from published linguistics literature. The task is to predict whether a given sentence is grammatically correct or not.\n",
"* SST-2 - [The Stanford Sentiment Treebank](https://nlp.stanford.edu/~socherr/EMNLP2013_RNTN.pdf) consists of sentences from movie reviews and human annotations of their sentiment. The task is to predict the sentiment of a given sentence: positive or negative.\n",
"\n",
"## Similarity and Paraphrase tasks\n",
"\n",
"* MRPC - [The Microsoft Research Paraphrase Corpus](https://www.aclweb.org/anthology/I05-5002.pdf) is a corpus of sentence pairs automatically extracted from online news sources, with human annotations for whether the sentences in the pair are semantically equivalent.\n",
"* QQP - [The Quora Question Pairs](https://www.quora.com/q/quoradata/First-Quora-Dataset-Release-Question-Pairs) dataset is a collection of question pairs from the community question-answering website Quora. The task is to determine whether a pair of questions are semantically equivalent.\n",
"* STS-B - [The Semantic Textual Similarity Benchmark](https://arxiv.org/abs/1708.00055) is a collection of sentence pairs drawn from news headlines, video, and image captions, and natural language inference data. The task is to determine how similar two sentences are.\n",
"\n",
"## Inference Tasks\n",
"\n",
"* MNLI - [The Multi-Genre Natural Language Inference Corpus](https://cims.nyu.edu/~sbowman/multinli/multinli_0.9.pdf) is a crowdsourced collection of sentence pairs with textual entailment annotations. Given a premise sentence and a hypothesis sentence, the task is to predict whether the premise entails the hypothesis (entailment), contradicts the hypothesis (contradiction), or neither (neutral). The task has the matched (in-domain) and mismatched (cross-domain) sections.\n",
"* QNLI - [The Stanford Question Answering Dataset](https://nlp.stanford.edu/pubs/rajpurkar2016squad.pdf) is a question-answering dataset consisting of question-paragraph pairs, where one of the sentences in the paragraph (drawn from Wikipedia) contains the answer to the corresponding question. The task is to determine whether the context sentence contains the answer to the question.\n",
"* RTE The Recognizing Textual Entailment (RTE) datasets come from a series of annual [textual entailment challenges](https://aclweb.org/aclwiki/Recognizing_Textual_Entailment). The task is to determine whether the second sentence is the entailment of the first one or not.\n",
"* WNLI - The Winograd Schema Challenge is a reading comprehension task in which a system must read a sentence with a pronoun and select the referent of that pronoun from a list of choices (Hector Levesque, Ernest Davis, and Leora Morgenstern. The winograd schema challenge. In Thirteenth International Conference on the Principles of Knowledge Representation and Reasoning. 2012).\n",
"\n",
"All tasks are classification tasks, except for the STS-B task which is a regression task. All classification tasks are 2-class problems, except for the MNLI task which has 3-classes.\n",
"\n",
"More details about GLUE benchmark could be found [here](https://gluebenchmark.com/)."
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "ZnuziSwJ1yEB",
"colab_type": "text"
},
"source": [
"# Datasets\n",
"\n",
"**To proceed further, you need to download the GLUE data.** For example, you can download [this script](https://gist.githubusercontent.com/W4ngatang/60c2bdb54d156a41194446737ce03e2e/raw/17b8dd0d724281ed7c3b2aeeda662b92809aadd5/download_glue_data.py) using `wget` and then execute it by running:\n",
"\n",
"`python download_glue_data.py`\n",
"\n",
"use `--tasks TASK` if datasets for only selected GLUE tasks are needed\n",
"\n",
"After running the above commands, you will have a folder `glue_data` with data folders for every GLUE task. For example, data for MRPC task would be under glue_data/MRPC.\n",
"\n",
"This tutorial and [examples/nlp/glue_benchmark/glue_benchmark.py](https://github.com/NVIDIA/NeMo/blob/stable/examples/nlp/glue_benchmark/glue_benchmark.py) work with all GLUE tasks without any modifications. For this tutorial, we are going to use MRPC task.\n",
"\n",
"\n",
"\n"
]
},
{
"cell_type": "code",
"metadata": {
"id": "--wJ2891aIIE",
"colab_type": "code",
"colab": {}
},
"source": [
"# supported task names: [\"cola\", \"sst-2\", \"mrpc\", \"sts-b\", \"qqp\", \"mnli\", \"qnli\", \"rte\", \"wnli\"]\n",
"TASK = 'mrpc'\n",
"DATA_DIR = 'glue_data/MRPC'\n",
"WORK_DIR = \"WORK_DIR\"\n",
"MODEL_CONFIG = 'glue_benchmark_config.yaml'"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "qB0oLE4R9EhJ",
"colab_type": "code",
"colab": {}
},
"source": [
"! ls -l $DATA_DIR"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"metadata": {
"id": "gMWuU69pbUDe",
"colab_type": "text"
},
"source": [
"For each task, there are 3 files: `train.tsv, dev.tsv, and test.tsv`. Note, MNLI has 2 dev sets: matched and mismatched, evaluation on both dev sets will be done automatically."
]
},
{
"cell_type": "code",
"metadata": {
"id": "6UDPgadLN6SG",
"colab_type": "code",
"colab": {}
},
"source": [
"# let's take a look at the training data \n",
"! head -n 5 {DATA_DIR}/train.tsv"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"metadata": {
"id": "_whKCxfTMo6Y",
"colab_type": "text"
},
"source": [
"# Model configuration\n",
"\n",
"Now, let's take a closer look at the model's configuration and learn to train the model.\n",
"\n",
"GLUE model is comprised of the pretrained [BERT](https://arxiv.org/pdf/1810.04805.pdf) model followed by a Sequence Regression module (for STS-B task) or Sequence classifier module (for the rest of the tasks).\n",
"\n",
"The model is defined in a config file which declares multiple important sections. They are:\n",
"- **model**: All arguments that are related to the Model - language model, a classifier, optimizer and schedulers, datasets and any other related information\n",
"\n",
"- **trainer**: Any argument to be passed to PyTorch Lightning"
]
},
{
"cell_type": "code",
"metadata": {
"id": "T1gA8PsJ13MJ",
"colab_type": "code",
"colab": {}
},
"source": [
"# download the model's configuration file \n",
"config_dir = WORK_DIR + '/configs/'\n",
"os.makedirs(config_dir, exist_ok=True)\n",
"if not os.path.exists(config_dir + MODEL_CONFIG):\n",
" print('Downloading config file...')\n",
" wget.download(f'https://raw.githubusercontent.com/NVIDIA/NeMo/{BRANCH}/examples/nlp/glue_benchmark/' + MODEL_CONFIG, config_dir)\n",
"else:\n",
" print ('config file is already exists')"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "mX3KmWMvSUQw",
"colab_type": "code",
"colab": {}
},
"source": [
"# this line will print the entire config of the model\n",
"config_path = f'{WORK_DIR}/configs/{MODEL_CONFIG}'\n",
"print(config_path)\n",
"config = OmegaConf.load(config_path)\n",
"print(OmegaConf.to_yaml(config))"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"metadata": {
"id": "ZCgWzNBkaQLZ",
"colab_type": "text"
},
"source": [
"# Model Training\n",
"## Setting up Data within the config\n",
"\n",
"Among other things, the config file contains dictionaries called **dataset**, **train_ds** and **validation_ds**. These are configurations used to setup the Dataset and DataLoaders of the corresponding config.\n",
"\n",
"We assume that both training and evaluation files are located in the same directory, and use the default names mentioned during the data download step. \n",
"So, to start model training, we simply need to specify `model.dataset.data_dir`, like we are going to do below.\n",
"\n",
"Also notice that some config lines, including `model.dataset.data_dir`, have `???` in place of paths, this means that values for these fields are required to be specified by the user.\n",
"\n",
"Let's now add the data directory path, task name and output directory for saving predictions to the config."
]
},
{
"cell_type": "code",
"metadata": {
"id": "LQHCJN-ZaoLp",
"colab_type": "code",
"colab": {}
},
"source": [
"config.model.task_name = TASK\n",
"config.model.output_dir = WORK_DIR\n",
"config.model.dataset.data_dir = DATA_DIR"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"metadata": {
"id": "nB96-3sTc3yk",
"colab_type": "text"
},
"source": [
"## Building the PyTorch Lightning Trainer\n",
"\n",
"NeMo models are primarily PyTorch Lightning modules - and therefore are entirely compatible with the PyTorch Lightning ecosystem.\n",
"\n",
"Let's first instantiate a Trainer object"
]
},
{
"cell_type": "code",
"metadata": {
"id": "1tG4FzZ4Ui60",
"colab_type": "code",
"colab": {}
},
"source": [
"print(\"Trainer config - \\n\")\n",
"print(OmegaConf.to_yaml(config.trainer))"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "knF6QeQQdMrH",
"colab_type": "code",
"colab": {}
},
"source": [
"# lets modify some trainer configs\n",
"# checks if we have GPU available and uses it\n",
"accelerator = 'gpu' if torch.cuda.is_available() else 'cpu'\n",
"config.trainer.devices = 1\n",
"config.trainer.accelerator = accelerator\n",
"\n",
"config.trainer.precision = 16 if torch.cuda.is_available() else 32\n",
"\n",
"# for mixed precision training, uncomment the line below (precision should be set to 16 and amp_level to O1):\n",
"# config.trainer.amp_level = O1\n",
"\n",
"# remove distributed training flags\n",
"config.trainer.strategy = None\n",
"\n",
"# setup max number of steps to reduce training time for demonstration purposes of this tutorial\n",
"config.trainer.max_steps = 128\n",
"\n",
"trainer = pl.Trainer(**config.trainer)"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"metadata": {
"id": "8IlEMdVxdr6p",
"colab_type": "text"
},
"source": [
"## Setting up a NeMo Experiment\n",
"\n",
"NeMo has an experiment manager that handles logging and checkpointing for us, so let's use it:"
]
},
{
"cell_type": "code",
"metadata": {
"id": "8uztqGAmdrYt",
"colab_type": "code",
"colab": {}
},
"source": [
"exp_dir = exp_manager(trainer, config.get(\"exp_manager\", None))\n",
"\n",
"# the exp_dir provides a path to the current experiment for easy access\n",
"exp_dir = str(exp_dir)\n",
"exp_dir"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"metadata": {
"id": "8tjLhUvL_o7_",
"colab_type": "text"
},
"source": [
"Before initializing the model, we might want to modify some of the model configs. For example, we might want to modify the pretrained BERT model and use [Megatron-LM BERT](https://arxiv.org/abs/1909.08053) or [AlBERT model](https://arxiv.org/abs/1909.11942):"
]
},
{
"cell_type": "code",
"metadata": {
"id": "Xeuc2i7Y_nP5",
"colab_type": "code",
"colab": {}
},
"source": [
"# get the list of supported BERT-like models, for the complete list of HugginFace models, see https://huggingface.co/models\n",
"print(nemo_nlp.modules.get_pretrained_lm_models_list(include_external=True))\n",
"\n",
"# specify BERT-like model, you want to use, for example, \"megatron-bert-345m-uncased\" or 'bert-base-uncased'\n",
"PRETRAINED_BERT_MODEL = \"albert-base-v1\""
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "RK2xglXyAUOO",
"colab_type": "code",
"colab": {}
},
"source": [
"# add the specified above model parameters to the config\n",
"config.model.language_model.pretrained_model_name = PRETRAINED_BERT_MODEL"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"metadata": {
"id": "fzNZNAVRjDD-",
"colab_type": "text"
},
"source": [
"Now, we are ready to initialize our model. During the model initialization call, the dataset and data loaders we'll be prepared for training and evaluation.\n",
"Also, the pretrained BERT model will be downloaded, note it can take up to a few minutes depending on the size of the chosen BERT model."
]
},
{
"cell_type": "code",
"metadata": {
"id": "NgsGLydWo-6-",
"colab_type": "code",
"colab": {}
},
"source": [
"model = nemo_nlp.models.GLUEModel(cfg=config.model, trainer=trainer)"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"metadata": {
"id": "kQ592Tx4pzyB",
"colab_type": "text"
},
"source": [
"## Monitoring training progress\n",
"Optionally, you can create a Tensorboard visualization to monitor training progress."
]
},
{
"cell_type": "code",
"metadata": {
"id": "mTJr16_pp0aS",
"colab_type": "code",
"colab": {}
},
"source": [
"try:\n",
" from google import colab\n",
" COLAB_ENV = True\n",
"except (ImportError, ModuleNotFoundError):\n",
" COLAB_ENV = False\n",
"\n",
"# Load the TensorBoard notebook extension\n",
"if COLAB_ENV:\n",
" %load_ext tensorboard\n",
" %tensorboard --logdir {exp_dir}\n",
"else:\n",
" print(\"To use tensorboard, please use this notebook in a Google Colab environment.\")"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"metadata": {
"id": "CFgAlaIdndjW",
"colab_type": "text"
},
"source": [
"Note, it’s recommended to finetune the model on each task separately. Also, based on [GLUE Benchmark FAQ#12](https://gluebenchmark.com/faq), there are might be some differences in dev/test distributions for QQP task and in train/dev for WNLI task."
]
},
{
"cell_type": "code",
"metadata": {
"id": "hUvnSpyjp0Dh",
"colab_type": "code",
"colab": {}
},
"source": [
"# start model training\n",
"trainer.fit(model)"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"metadata": {
"id": "ref1qSonGNhP",
"colab_type": "text"
},
"source": [
"## Training Script\n",
"\n",
"If you have NeMo installed locally, you can also train the model with [examples/nlp/glue_benchmark/glue_benchmark.py](https://github.com/NVIDIA/NeMo/blob/stable/examples/nlp/glue_benchmark/glue_benchmark.py).\n",
"\n",
"To run training script, use:\n",
"\n",
"`python glue_benchmark.py \\\n",
" model.dataset.data_dir=PATH_TO_DATA_DIR \\\n",
" model.task_name=TASK`\n"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "KVPFofXaoKNE",
"colab_type": "text"
},
"source": [
"Average results after 3 runs:\n",
"\n",
"| Task | Metric | ALBERT-large | ALBERT-xlarge | Megatron-345m | BERT base paper | BERT large paper |\n",
"|-------|--------------------------|--------------|---------------|---------------|-----------------|------------------|\n",
"| CoLA | Matthew's correlation | 54.94 | 61.72 | 64.56 | 52.1 | 60.5 |\n",
"| SST-2 | Accuracy | 92.74 | 91.86 | 95.87 | 93.5 | 94.9 |\n",
"| MRPC | F1/Accuracy | 92.05/88.97 | 91.87/88.61 | 92.36/89.46 | 88.9/- | 89.3/- |\n",
"| STS-B | Person/Spearman corr. | 90.41/90.21 | 90.07/90.10 | 91.51/91.61 | -/85.8 | -/86.5 |\n",
"| QQP | F1/Accuracy | 88.26/91.26 | 88.80/91.65 | 89.18/91.91 | 71.2/- | 72.1/- |\n",
"| MNLI | Matched /Mismatched acc. | 86.69/86.81 | 88.66/88.73 | 89.86/89.81 | 84.6/83.4 | 86.7/85.9 |\n",
"| QNLI | Accuracy | 92.68 | 93.66 | 94.33 | 90.5 | 92.7 |\n",
"| RTE | Accuracy | 80.87 | 82.86 | 83.39 | 66.4 | 70.1 |\n",
"\n",
"WNLI task was excluded from the experiments due to the problematic WNLI set.\n",
"The dev sets were used for evaluation for ALBERT and Megatron models, and the test sets results for [the BERT paper](https://arxiv.org/abs/1810.04805).\n",
"\n",
"Hyperparameters used to get the results from the above table, could be found in the table below. Some tasks could be further finetuned to improve performance numbers, the tables are for a baseline reference only.\n",
"Each cell in the table represents the following parameters:\n",
"Number of GPUs used/ Batch Size/ Learning Rate/ Number of Epochs. For not specified parameters, please refer to the default parameters in the training script.\n",
"\n",
"| Task | ALBERT-large | ALBERT-xlarge | Megatron-345m |\n",
"|-------|--------------|---------------|---------------|\n",
"| CoLA | 1 / 32 / 1e-5 / 3 | 1 / 32 / 1e-5 / 10 | 4 / 16 / 2e-5 / 12 |\n",
"| SST-2 | 4 / 16 / 2e-5 / 5 | 4 / 16 / 2e-5 /12 | 4 / 16 / 2e-5 / 12 |\n",
"| MRPC | 1 / 32 / 1e-5 / 5 | 1 / 16 / 2e-5 / 5 | 1 / 16 / 2e-5 / 10 |\n",
"| STS-B | 1 / 16 / 2e-5 / 5 | 1 / 16 / 4e-5 / 12 | 4 / 16 / 3e-5 / 12 |\n",
"| QQP | 1 / 16 / 2e-5 / 5 | 4 / 16 / 1e-5 / 12 | 4 / 16 / 1e-5 / 12 |\n",
"| MNLI | 4 / 64 / 1e-5 / 5 | 4 / 32 / 1e-5 / 5 | 4 / 32 / 1e-5 / 5 | \n",
"| QNLI | 4 / 16 / 1e-5 / 5 | 4 / 16 / 1e-5 / 5 | 4 / 16 / 2e-5 / 5 | \n",
"| RTE | 1 / 16 / 1e-5 / 5 | 1 / 16 / 1e-5 / 12 | 4 / 16 / 3e-5 / 12 |\n"
]
}
]
} |