train_and_evaluate.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import (BigBirdPegasusForConditionalGeneration, AutoTokenizer, \n",
    "DataCollatorForSeq2Seq, Seq2SeqTrainingArguments, Seq2SeqTrainer )\n",
    "from datasets import load_dataset, load_metric\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "data_files = 'ccdv/cnn_dailymail'\n",
    "model_name = 'google/bigbird-pegasus-large-pubmed'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Reusing dataset cnn_dailymail (/Users/johnchou/.cache/huggingface/datasets/ccdv___cnn_dailymail/3.0.0/3.0.0/0107f7388b5c6fae455a5661bcd134fc22da53ea75852027040d8d1e997f101f)\n",
      "Reusing dataset cnn_dailymail (/Users/johnchou/.cache/huggingface/datasets/ccdv___cnn_dailymail/3.0.0/3.0.0/0107f7388b5c6fae455a5661bcd134fc22da53ea75852027040d8d1e997f101f)\n"
     ]
    }
   ],
   "source": [
    "train_dataset = load_dataset(data_files, '3.0.0', split=\"train\")\n",
    "val_dataset = load_dataset(data_files, '3.0.0', split=\"validation\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "For the sake of this trial, I only select a few samples in the database,\n",
    "please comment the cell below to train the model on full dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_dataset = train_dataset.select(range(100))\n",
    "val_dataset = val_dataset.select(range(20))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "def preprossess_text(example):\n",
    "    return {'article':example['article'].replace('\\'', ''), \n",
    "            'highlights':example['highlights'].replace('\\'', ''), \n",
    "            'length': len(example['article'].split())}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "def lowercase(example):\n",
    "    return{'article':example['article'].lower(),\n",
    "          'highlights':example['highlights'].lower()}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Loading cached processed dataset at /Users/johnchou/.cache/huggingface/datasets/ccdv___cnn_dailymail/3.0.0/3.0.0/0107f7388b5c6fae455a5661bcd134fc22da53ea75852027040d8d1e997f101f/cache-736a077279fb0b7c.arrow\n",
      "Loading cached processed dataset at /Users/johnchou/.cache/huggingface/datasets/ccdv___cnn_dailymail/3.0.0/3.0.0/0107f7388b5c6fae455a5661bcd134fc22da53ea75852027040d8d1e997f101f/cache-ef00eadf1766c0ac.arrow\n",
      "Loading cached processed dataset at /Users/johnchou/.cache/huggingface/datasets/ccdv___cnn_dailymail/3.0.0/3.0.0/0107f7388b5c6fae455a5661bcd134fc22da53ea75852027040d8d1e997f101f/cache-17eca76de43ae8c7.arrow\n"
     ]
    }
   ],
   "source": [
    "train_dataset = train_dataset.map(preprossess_text)\n",
    "train_dataset = train_dataset.filter(lambda x: x['length'] > 1000)\n",
    "train_dataset = train_dataset.map(lowercase)\n",
    "train_dataset = train_dataset.remove_columns(\"id\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "23"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(train_dataset)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Loading cached processed dataset at /Users/johnchou/.cache/huggingface/datasets/ccdv___cnn_dailymail/3.0.0/3.0.0/0107f7388b5c6fae455a5661bcd134fc22da53ea75852027040d8d1e997f101f/cache-7e644a1c6ca1171e.arrow\n",
      "Loading cached processed dataset at /Users/johnchou/.cache/huggingface/datasets/ccdv___cnn_dailymail/3.0.0/3.0.0/0107f7388b5c6fae455a5661bcd134fc22da53ea75852027040d8d1e997f101f/cache-110cc2bebd3a0506.arrow\n",
      "Loading cached processed dataset at /Users/johnchou/.cache/huggingface/datasets/ccdv___cnn_dailymail/3.0.0/3.0.0/0107f7388b5c6fae455a5661bcd134fc22da53ea75852027040d8d1e997f101f/cache-d73c9e11c6ad04de.arrow\n"
     ]
    }
   ],
   "source": [
    "val_dataset = val_dataset.map(preprossess_text)\n",
    "val_dataset = val_dataset.filter(lambda x: x['length'] > 1000)\n",
    "val_dataset = val_dataset.map(lowercase)\n",
    "val_dataset = val_dataset.remove_columns(\"id\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "2"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(val_dataset)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "tokenizer = AutoTokenizer.from_pretrained(model_name)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'input_ids': [1593, 123, 116, 1102, 224, 136, 929, 112, 630, 180, 123, 116, 3114, 107, 1], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]}"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tokenizer('Let’s walk through this code to understand what’s happening.')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "max_input_length = 4096\n",
    "max_target_length = 512\n",
    "batch_size = 3"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "def tokenize(examples):\n",
    "    model_inputs = tokenizer(\n",
    "        examples[\"article\"], max_length=max_input_length, truncation=True, return_tensors='pt'\n",
    "    )\n",
    "    with tokenizer.as_target_tokenizer():\n",
    "        labels = tokenizer(\n",
    "            examples[\"highlights\"], max_length=max_target_length, truncation=True, return_tensors='pt'\n",
    "        )\n",
    "\n",
    "    model_inputs[\"labels\"] = labels[\"input_ids\"]\n",
    "    return model_inputs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "498d202d1d5a40c9b9fe5a1c2aca2302",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/1 [00:00<?, ?ba/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "train_dataset_tokenized = train_dataset.map(tokenize, batched=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "4f1b800f1d804252b9857f73c82e3bf6",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/1 [00:00<?, ?ba/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "val_dataset_tokenized = val_dataset.map(tokenize, batched=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Dataset({\n",
       "    features: ['article', 'highlights', 'length', 'input_ids', 'attention_mask', 'labels'],\n",
       "    num_rows: 2\n",
       "})"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "val_dataset_tokenized"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Dataset({\n",
       "    features: ['input_ids', 'attention_mask', 'labels'],\n",
       "    num_rows: 2\n",
       "})"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_dataset_tokenized.remove_columns(train_dataset.column_names)\n",
    "val_dataset_tokenized.remove_columns(val_dataset.column_names)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "data_collator = DataCollatorForSeq2Seq(tokenizer, model=model_name)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = BigBirdPegasusForConditionalGeneration.from_pretrained(model_name)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "f34a4ee1d6af48768b4748cfdae55c27",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading builder script:   0%|          | 0.00/2.16k [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "rouge = load_metric(\"rouge\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "def compute_metrics(pred):\n",
    "    labels_ids = pred.label_ids\n",
    "    pred_ids = pred.predictions\n",
    "\n",
    "    pred_str = tokenizer.batch_decode(pred_ids, skip_special_tokens=True)\n",
    "    labels_ids[labels_ids == -100] = tokenizer.pad_token_id\n",
    "    label_str = tokenizer.batch_decode(labels_ids, skip_special_tokens=True)\n",
    "    \n",
    "    rouge_output = rouge.compute(\n",
    "        predictions=pred_str, references=label_str, rouge_types=[\"rouge2\"]\n",
    "    )[\"rouge2\"].mid\n",
    "\n",
    "    return {\n",
    "        \"rouge2_precision\": round(rouge_output.precision, 4),\n",
    "        \"rouge2_recall\": round(rouge_output.recall, 4),\n",
    "        \"rouge2_fmeasure\": round(rouge_output.fmeasure, 4),\n",
    "    }"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "num_train_epochs = 8\n",
    "# Show the training loss with every epoch\n",
    "logging_steps = len(train_dataset_tokenized) // batch_size"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "args = Seq2SeqTrainingArguments(\n",
    "    output_dir=f\"./model/{model_name}-finetuned-on-cnn_news\",\n",
    "    evaluation_strategy=\"epoch\",\n",
    "    learning_rate=5.6e-5,\n",
    "    per_device_train_batch_size=batch_size,\n",
    "    per_device_eval_batch_size=batch_size,\n",
    "    weight_decay=0.01,\n",
    "    save_total_limit=3,\n",
    "    num_train_epochs=num_train_epochs,\n",
    "    predict_with_generate=True,\n",
    "    logging_steps=logging_steps,\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "trainer = Seq2SeqTrainer(\n",
    "    model,\n",
    "    args,\n",
    "    train_dataset=train_dataset_tokenized,\n",
    "    eval_dataset=val_dataset_tokenized,\n",
    "    compute_metrics=compute_metrics,\n",
    "    data_collator=data_collator,\n",
    "    tokenizer=tokenizer,\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "The following columns in the training set  don't have a corresponding argument in `BigBirdPegasusForConditionalGeneration.forward` and have been ignored: length, highlights, article. If length, highlights, article are not expected by `BigBirdPegasusForConditionalGeneration.forward`,  you can safely ignore this message.\n",
      "/Users/johnchou/opt/anaconda3/lib/python3.8/site-packages/transformers/optimization.py:306: FutureWarning: This implementation of AdamW is deprecated and will be removed in a future version. Use the PyTorch implementation torch.optim.AdamW instead, or set `no_deprecation_warning=True` to disable this warning\n",
      "  warnings.warn(\n",
      "***** Running training *****\n",
      "  Num examples = 23\n",
      "  Num Epochs = 8\n",
      "  Instantaneous batch size per device = 3\n",
      "  Total train batch size (w. parallel, distributed & accumulation) = 3\n",
      "  Gradient Accumulation steps = 1\n",
      "  Total optimization steps = 64\n",
      "Input ids are automatically padded from 2044 to 2048 to be a multiple of `config.block_size`: 64\n"
     ]
    }
   ],
   "source": [
    "trainer.train()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "trainer.evaluate()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}