reproduceability

gpt2ia_v1.ipynb

@@ -0,0 +1,260 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import pandas as pd\n",
+    "from transformers import GPT2LMHeadModel, GPT2Tokenizer\n",
+    "import numpy as np\n",
+    "import random\n",
+    "import torch\n",
+    "from torch.utils.data import Dataset, DataLoader\n",
+    "from transformers import GPT2Tokenizer, GPT2LMHeadModel, AdamW, get_linear_schedule_with_warmup\n",
+    "from tqdm import tqdm, trange\n",
+    "import torch.nn.functional as F\n",
+    "import csv\n",
+    "from transformers import TextDataset, DataCollatorForLanguageModeling\n",
+    "from transformers import GPT2Tokenizer, GPT2LMHeadModel\n",
+    "from transformers import Trainer, TrainingArguments"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "100000\n"
+     ]
+    }
+   ],
+   "source": [
+    "with open(\"./titulos.txt\") as file:\n",
+    "    manchetes = [line.rstrip() for line in file]\n",
+    "print(len(manchetes))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def load_dataset(file_path, tokenizer, block_size = 128):\n",
+    "    dataset = TextDataset(\n",
+    "        tokenizer = tokenizer,\n",
+    "        file_path = file_path,\n",
+    "        block_size = block_size,\n",
+    "    )\n",
+    "    return dataset\n",
+    "\n",
+    "\n",
+    "def load_data_collator(tokenizer, mlm = False):\n",
+    "    data_collator = DataCollatorForLanguageModeling(\n",
+    "        tokenizer=tokenizer,\n",
+    "        mlm=mlm,\n",
+    "    )\n",
+    "    return data_collator\n",
+    "\n",
+    "\n",
+    "def train(train_file_path,model_name,\n",
+    "          output_dir,\n",
+    "          overwrite_output_dir,\n",
+    "          per_device_train_batch_size,\n",
+    "          num_train_epochs,\n",
+    "          save_steps):\n",
+    "  tokenizer = GPT2Tokenizer.from_pretrained(model_name)\n",
+    "  train_dataset = load_dataset(train_file_path, tokenizer)\n",
+    "  data_collator = load_data_collator(tokenizer)\n",
+    "\n",
+    "  tokenizer.save_pretrained(output_dir)\n",
+    "\n",
+    "  model = GPT2LMHeadModel.from_pretrained(model_name)\n",
+    "\n",
+    "  model.save_pretrained(output_dir)\n",
+    "\n",
+    "  training_args = TrainingArguments(\n",
+    "          output_dir=output_dir,\n",
+    "          overwrite_output_dir=overwrite_output_dir,\n",
+    "          per_device_train_batch_size=per_device_train_batch_size,\n",
+    "          num_train_epochs=num_train_epochs,\n",
+    "      )\n",
+    "\n",
+    "  trainer = Trainer(\n",
+    "          model=model,\n",
+    "          args=training_args,\n",
+    "          data_collator=data_collator,\n",
+    "          train_dataset=train_dataset,\n",
+    "  )\n",
+    "\n",
+    "  trainer.train()\n",
+    "  trainer.save_model()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "train_file_path = \"./titulos.txt\"\n",
+    "model_name = 'gpt2'\n",
+    "output_dir = './result'\n",
+    "overwrite_output_dir = False\n",
+    "per_device_train_batch_size = 8\n",
+    "num_train_epochs = 5.0\n",
+    "save_steps = 500"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "c:\\Users\\yamgc\\miniconda3\\envs\\choqueianotebook1\\lib\\site-packages\\transformers\\data\\datasets\\language_modeling.py:53: FutureWarning: This dataset will be removed from the library soon, preprocessing should be handled with the 🤗 Datasets library. You can have a look at this example script for pointers: https://github.com/huggingface/transformers/blob/main/examples/pytorch/language-modeling/run_mlm.py\n",
+      "  warnings.warn(\n"
+     ]
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "f791b828d56f4844a369b75f66ec071a",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "  0%|          | 0/11455 [00:00<?, ?it/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "{'loss': 4.0109, 'learning_rate': 4.7817546922741165e-05, 'epoch': 0.22}\n",
+      "{'loss': 3.3335, 'learning_rate': 4.563509384548232e-05, 'epoch': 0.44}\n",
+      "{'loss': 3.0906, 'learning_rate': 4.3452640768223485e-05, 'epoch': 0.65}\n",
+      "{'loss': 2.9653, 'learning_rate': 4.127018769096465e-05, 'epoch': 0.87}\n",
+      "{'loss': 2.855, 'learning_rate': 3.9087734613705804e-05, 'epoch': 1.09}\n",
+      "{'loss': 2.764, 'learning_rate': 3.690528153644697e-05, 'epoch': 1.31}\n",
+      "{'loss': 2.7204, 'learning_rate': 3.472282845918813e-05, 'epoch': 1.53}\n",
+      "{'loss': 2.6883, 'learning_rate': 3.2540375381929286e-05, 'epoch': 1.75}\n",
+      "{'loss': 2.6503, 'learning_rate': 3.035792230467045e-05, 'epoch': 1.96}\n",
+      "{'loss': 2.5871, 'learning_rate': 2.817546922741161e-05, 'epoch': 2.18}\n",
+      "{'loss': 2.5675, 'learning_rate': 2.5993016150152772e-05, 'epoch': 2.4}\n",
+      "{'loss': 2.5442, 'learning_rate': 2.3810563072893935e-05, 'epoch': 2.62}\n",
+      "{'loss': 2.5271, 'learning_rate': 2.1628109995635095e-05, 'epoch': 2.84}\n",
+      "{'loss': 2.4952, 'learning_rate': 1.9445656918376258e-05, 'epoch': 3.06}\n",
+      "{'loss': 2.4653, 'learning_rate': 1.7263203841117417e-05, 'epoch': 3.27}\n",
+      "{'loss': 2.4566, 'learning_rate': 1.5080750763858579e-05, 'epoch': 3.49}\n",
+      "{'loss': 2.4551, 'learning_rate': 1.2898297686599738e-05, 'epoch': 3.71}\n",
+      "{'loss': 2.4445, 'learning_rate': 1.07158446093409e-05, 'epoch': 3.93}\n",
+      "{'loss': 2.4247, 'learning_rate': 8.533391532082061e-06, 'epoch': 4.15}\n",
+      "{'loss': 2.4052, 'learning_rate': 6.3509384548232216e-06, 'epoch': 4.36}\n",
+      "{'loss': 2.4033, 'learning_rate': 4.168485377564382e-06, 'epoch': 4.58}\n",
+      "{'loss': 2.3983, 'learning_rate': 1.9860323003055434e-06, 'epoch': 4.8}\n",
+      "{'train_runtime': 2081.9794, 'train_samples_per_second': 44.004, 'train_steps_per_second': 5.502, 'train_loss': 2.6816321317621945, 'epoch': 5.0}\n"
+     ]
+    }
+   ],
+   "source": [
+    "train(\n",
+    "    train_file_path=train_file_path,\n",
+    "    model_name=model_name,\n",
+    "    output_dir=output_dir,\n",
+    "    overwrite_output_dir=overwrite_output_dir,\n",
+    "    per_device_train_batch_size=per_device_train_batch_size,\n",
+    "    num_train_epochs=num_train_epochs,\n",
+    "    save_steps=save_steps\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 8,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from transformers import PreTrainedTokenizerFast, GPT2LMHeadModel, GPT2TokenizerFast, GPT2Tokenizer\n",
+    "\n",
+    "def load_model(model_path):\n",
+    "    model = GPT2LMHeadModel.from_pretrained(model_path)\n",
+    "    return model\n",
+    "\n",
+    "\n",
+    "def load_tokenizer(tokenizer_path):\n",
+    "    tokenizer = GPT2Tokenizer.from_pretrained(tokenizer_path)\n",
+    "    return tokenizer\n",
+    "\n",
+    "\n",
+    "def generate_text(sequence, max_length):\n",
+    "    model_path = \"./result\"\n",
+    "    model = load_model(model_path)\n",
+    "    tokenizer = load_tokenizer(model_path)\n",
+    "    ids = tokenizer.encode(f'{sequence}', return_tensors='pt')\n",
+    "    final_outputs = model.generate(\n",
+    "        ids,\n",
+    "        do_sample=True,\n",
+    "        max_length=max_length,\n",
+    "        pad_token_id=model.config.eos_token_id,\n",
+    "        top_k=50,\n",
+    "        top_p=0.95,\n",
+    "    )\n",
+    "    print(tokenizer.decode(final_outputs[0], skip_special_tokens=True))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Juliette rebate críticas à mostra em clique sexy em evento\n",
+      "\n"
+     ]
+    }
+   ],
+   "source": [
+    "sequence = \"Juliette\"\n",
+    "max_len = 19\n",
+    "generate_text(sequence, max_len)"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "choqueianotebook1",
+   "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": "3.9.18"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}