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"
      ]
    }
  ]
}