{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Tune GPT2 to generate positive reviews\n",
    "> Optimise GPT2 to produce positive IMDB movie reviews using a BERT sentiment classifier as a reward function."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"text-align: center\">\n",
    "<img src='https://huggingface.co/datasets/trl-lib/documentation-images/resolve/main/gpt2_bert_training.png' width='600'>\n",
    "<p style=\"text-align: center;\"> <b>Figure:</b> Experiment setup to tune GPT2. The yellow arrows are outside the scope of this notebook, but the trained models are available through Hugging Face. </p>\n",
    "</div>\n",
    "\n",
    "\n",
    "In this notebook we fine-tune GPT2 (small) to generate positive movie reviews based on the IMDB dataset. The model gets the start of a real review and is tasked to produce positive continuations. To reward positive continuations we use a BERT classifier to analyse the sentiment of the produced sentences and use the classifier's outputs as rewards signals for PPO training."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Setup experiment"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Import dependencies"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%load_ext autoreload\n",
    "%autoreload 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%pip install transformers trl wandb"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "from tqdm import tqdm\n",
    "import pandas as pd\n",
    "\n",
    "tqdm.pandas()\n",
    "\n",
    "from transformers import pipeline, AutoTokenizer\n",
    "from datasets import load_dataset\n",
    "\n",
    "from trl import PPOTrainer, PPOConfig, AutoModelForCausalLMWithValueHead\n",
    "from trl.core import LengthSampler"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Configuration"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "config = PPOConfig(\n",
    "    model_name=\"lvwerra/gpt2-imdb\",\n",
    "    learning_rate=1.41e-5,\n",
    "    log_with=\"wandb\",\n",
    ")\n",
    "\n",
    "sent_kwargs = {\"top_k\": None, \"function_to_apply\": \"none\", \"batch_size\": 16}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import wandb\n",
    "\n",
    "wandb.init()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "You can see that we load a GPT2 model called `gpt2_imdb`. This model was additionally fine-tuned on the IMDB dataset for 1 epoch with the huggingface [script](https://github.com/huggingface/transformers/blob/main/examples/legacy/run_language_modeling.py) (no special settings). The other parameters are mostly taken from the original paper [\"Fine-Tuning Language Models from Human Preferences\"](\n",
    "https://huggingface.co/papers/1909.08593). This model as well as the BERT model is available in the Huggingface model zoo [here](https://huggingface.co/models). The following code should automatically download the models."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load data and models"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Load IMDB dataset\n",
    "The IMDB dataset contains 50k movie review annotated with \"positive\"/\"negative\" feedback indicating the sentiment.  We load the IMDB dataset into a DataFrame and filter for comments that are at least 200 characters. Then we tokenize each text and cut it to random size with the `LengthSampler`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def build_dataset(\n",
    "    config,\n",
    "    dataset_name=\"stanfordnlp/imdb\",\n",
    "    input_min_text_length=2,\n",
    "    input_max_text_length=8,\n",
    "):\n",
    "    \"\"\"\n",
    "    Build dataset for training. This builds the dataset from `load_dataset`, one should\n",
    "    customize this function to train the model on its own dataset.\n",
    "\n",
    "    Args:\n",
    "        dataset_name (`str`):\n",
    "            The name of the dataset to be loaded.\n",
    "\n",
    "    Returns:\n",
    "        dataloader (`torch.utils.data.DataLoader`):\n",
    "            The dataloader for the dataset.\n",
    "    \"\"\"\n",
    "    tokenizer = AutoTokenizer.from_pretrained(config.model_name)\n",
    "    tokenizer.pad_token = tokenizer.eos_token\n",
    "    # load imdb with datasets\n",
    "    ds = load_dataset(dataset_name, split=\"train\")\n",
    "    ds = ds.rename_columns({\"text\": \"review\"})\n",
    "    ds = ds.filter(lambda x: len(x[\"review\"]) > 200, batched=False)\n",
    "\n",
    "    input_size = LengthSampler(input_min_text_length, input_max_text_length)\n",
    "\n",
    "    def tokenize(sample):\n",
    "        sample[\"input_ids\"] = tokenizer.encode(sample[\"review\"])[: input_size()]\n",
    "        sample[\"query\"] = tokenizer.decode(sample[\"input_ids\"])\n",
    "        return sample\n",
    "\n",
    "    ds = ds.map(tokenize, batched=False)\n",
    "    ds.set_format(type=\"torch\")\n",
    "    return ds"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "dataset = build_dataset(config)\n",
    "\n",
    "\n",
    "def collator(data):\n",
    "    return dict((key, [d[key] for d in data]) for key in data[0])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Load pre-trained GPT2 language models"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We load the GPT2 model with a value head and the tokenizer. We load the model twice; the first model is optimized while the second model serves as a reference to calculate the KL-divergence from the starting point. This serves as an additional reward signal in the PPO training to make sure the optimized model does not deviate too much from the original language model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = AutoModelForCausalLMWithValueHead.from_pretrained(config.model_name)\n",
    "ref_model = AutoModelForCausalLMWithValueHead.from_pretrained(config.model_name)\n",
    "tokenizer = AutoTokenizer.from_pretrained(config.model_name)\n",
    "\n",
    "tokenizer.pad_token = tokenizer.eos_token"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Initialize PPOTrainer\n",
    "The `PPOTrainer` takes care of device placement and optimization later on:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "ppo_trainer = PPOTrainer(\n",
    "    config, model, ref_model, tokenizer, dataset=dataset, data_collator=collator\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Load BERT classifier\n",
    "We load a BERT classifier fine-tuned on the IMDB dataset."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "device = ppo_trainer.accelerator.device\n",
    "if ppo_trainer.accelerator.num_processes == 1:\n",
    "    device = 0 if torch.cuda.is_available() else \"cpu\"  # to avoid a `pipeline` bug\n",
    "sentiment_pipe = pipeline(\n",
    "    \"sentiment-analysis\", model=\"lvwerra/distilbert-imdb\", device=device\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The model outputs are the logits for the negative and positive class. We will use the logits for positive class as a reward signal for the language model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[{'label': 'NEGATIVE', 'score': 2.335048198699951},\n",
       " {'label': 'POSITIVE', 'score': -2.726576328277588}]"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "text = \"this movie was really bad!!\"\n",
    "sentiment_pipe(text, **sent_kwargs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[{'label': 'POSITIVE', 'score': 2.557040214538574},\n",
       " {'label': 'NEGATIVE', 'score': -2.294790267944336}]"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "text = \"this movie was really good!!\"\n",
    "sentiment_pipe(text, **sent_kwargs)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Generation settings\n",
    "For the response generation we just use sampling and make sure top-k and nucleus sampling are turned off as well as a minimal length."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "gen_kwargs = {\n",
    "    \"min_length\": -1,\n",
    "    \"top_k\": 0.0,\n",
    "    \"top_p\": 1.0,\n",
    "    \"do_sample\": True,\n",
    "    \"pad_token_id\": tokenizer.eos_token_id,\n",
    "}"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Optimize model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Training loop"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The training loop consists of the following main steps:\n",
    "1. Get the query responses from the policy network (GPT-2)\n",
    "2. Get sentiments for query/responses from BERT\n",
    "3. Optimize policy with PPO using the (query, response, reward) triplet\n",
    "\n",
    "**Training time**\n",
    "\n",
    "This step takes **~2h** on a V100 GPU with the above specified settings."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "output_min_length = 4\n",
    "output_max_length = 16\n",
    "output_length_sampler = LengthSampler(output_min_length, output_max_length)\n",
    "\n",
    "\n",
    "generation_kwargs = {\n",
    "    \"min_length\": -1,\n",
    "    \"top_k\": 0.0,\n",
    "    \"top_p\": 1.0,\n",
    "    \"do_sample\": True,\n",
    "    \"pad_token_id\": tokenizer.eos_token_id,\n",
    "}\n",
    "\n",
    "\n",
    "for epoch, batch in enumerate(tqdm(ppo_trainer.dataloader)):\n",
    "    query_tensors = batch[\"input_ids\"]\n",
    "\n",
    "    #### Get response from gpt2\n",
    "    response_tensors = []\n",
    "    for query in query_tensors:\n",
    "        gen_len = output_length_sampler()\n",
    "        generation_kwargs[\"max_new_tokens\"] = gen_len\n",
    "        query_response = ppo_trainer.generate(query, **generation_kwargs).squeeze()\n",
    "        response_len = len(query_response) - len(query)\n",
    "        response_tensors.append(query_response[-response_len:])\n",
    "    batch[\"response\"] = [tokenizer.decode(r.squeeze()) for r in response_tensors]\n",
    "\n",
    "    #### Compute sentiment score\n",
    "    texts = [q + r for q, r in zip(batch[\"query\"], batch[\"response\"])]\n",
    "    pipe_outputs = sentiment_pipe(texts, **sent_kwargs)\n",
    "    positive_scores = [\n",
    "        item[\"score\"]\n",
    "        for output in pipe_outputs\n",
    "        for item in output\n",
    "        if item[\"label\"] == \"POSITIVE\"\n",
    "    ]\n",
    "    rewards = [torch.tensor(score) for score in positive_scores]\n",
    "\n",
    "    #### Run PPO step\n",
    "    stats = ppo_trainer.step(query_tensors, response_tensors, rewards)\n",
    "    ppo_trainer.log_stats(stats, batch, rewards)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Training progress\n",
    "If you are tracking the training progress with Weights&Biases you should see a plot similar to the one below. Check out the interactive sample report on wandb.ai: [link](https://wandb.ai/huggingface/trl/runs/w9l3110g).\n",
    "\n",
    "<div style=\"text-align: center\">\n",
    "<img src='https://huggingface.co/datasets/trl-internal-testing/example-images/resolve/main/images/gpt2_tuning_progress.png' width='800'>\n",
    "<p style=\"text-align: center;\"> <b>Figure:</b> Reward mean and distribution evolution during training. </p>\n",
    "</div>\n",
    "\n",
    "One can observe how the model starts to generate more positive outputs after a few optimisation steps.\n",
    "\n",
    "> Note: Investigating the KL-divergence will probably show that at this point the model has not converged to the target KL-divergence, yet. To get there would require longer training or starting with a higher initial coefficient."
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Model inspection\n",
    "Let's inspect some examples from the IMDB dataset. We can use `ref_model` to compare the tuned model `model` against the model before optimisation."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>query</th>\n",
       "      <th>response (before)</th>\n",
       "      <th>response (after)</th>\n",
       "      <th>rewards (before)</th>\n",
       "      <th>rewards (after)</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>I rented Zero Day</td>\n",
       "      <td>4 for my sister. To my surprise, the Wii caug...</td>\n",
       "      <td>. It is a pleasure. It is a huge leap 68 years...</td>\n",
       "      <td>1.736068</td>\n",
       "      <td>2.423731</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>The only</td>\n",
       "      <td>distro of her</td>\n",
       "      <td>special compliments is the</td>\n",
       "      <td>0.150852</td>\n",
       "      <td>0.190159</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>I've read a few</td>\n",
       "      <td>news reports about Mr. Mueller's activities b...</td>\n",
       "      <td>novels and I never watch this. It has a reall...</td>\n",
       "      <td>-1.417962</td>\n",
       "      <td>2.831814</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>This is the second British Rank film</td>\n",
       "      <td>, and I wouldn't be surprised anymore if it</td>\n",
       "      <td>that I have enjoyed, achieving it in both the</td>\n",
       "      <td>0.835876</td>\n",
       "      <td>2.205628</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>A classic</td>\n",
       "      <td>classic.&lt;br /&gt;&lt;br /&gt;And only this one will ha...</td>\n",
       "      <td>. It's a movie with a fine cast. As the beginn...</td>\n",
       "      <td>2.113075</td>\n",
       "      <td>2.739168</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>This has to be one of the</td>\n",
       "      <td>worst with the differences being that for the</td>\n",
       "      <td>best thriller films I've seen in recent</td>\n",
       "      <td>-2.705339</td>\n",
       "      <td>2.730615</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>Happy Go Lovely is a waste</td>\n",
       "      <td>. Not only are extremely</td>\n",
       "      <td>of time, giving a</td>\n",
       "      <td>-2.429504</td>\n",
       "      <td>-2.934672</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>Wow, I just</td>\n",
       "      <td>can't make fun of it</td>\n",
       "      <td>feek it! This show</td>\n",
       "      <td>-2.201666</td>\n",
       "      <td>-0.106085</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>This movie makes several mistakes.</td>\n",
       "      <td>Despite being a great comedic diversion it es...</td>\n",
       "      <td>It's cool, wonderful - it held me into a very ...</td>\n",
       "      <td>-1.232380</td>\n",
       "      <td>2.707638</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>Branagh and Fish</td>\n",
       "      <td>burne, Drake is played</td>\n",
       "      <td>is a great show. Beautiful</td>\n",
       "      <td>0.776819</td>\n",
       "      <td>2.808996</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>I might have given this movie a</td>\n",
       "      <td>rating of *11 when I heard that!), but it was...</td>\n",
       "      <td>great performance. It was truly a great movie...</td>\n",
       "      <td>0.276380</td>\n",
       "      <td>2.743328</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>Really, really bad</td>\n",
       "      <td>with feel like there is no end to the</td>\n",
       "      <td>. This movie is incredibly good, with the</td>\n",
       "      <td>-2.639503</td>\n",
       "      <td>-1.568827</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>What another reviewer called lack of</td>\n",
       "      <td>judgment, connecting into her own harsh obser...</td>\n",
       "      <td>suspense. Rogers and Rooney rate this as exce...</td>\n",
       "      <td>-1.079707</td>\n",
       "      <td>2.696888</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>This is simply one</td>\n",
       "      <td>more problem of Steve</td>\n",
       "      <td>of the best choice</td>\n",
       "      <td>-1.445436</td>\n",
       "      <td>2.662699</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>\"Perhaps we can arrange a meet</td>\n",
       "      <td>-and-greet.&lt;br /&gt;&lt;br /&gt;Teleg</td>\n",
       "      <td>with spent, classic music and dance, and come...</td>\n",
       "      <td>0.258479</td>\n",
       "      <td>1.876662</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15</th>\n",
       "      <td>Richard Willaims is</td>\n",
       "      <td>nice enough; the little black guy plays quite</td>\n",
       "      <td>beautifully hands on in his own spin, and</td>\n",
       "      <td>0.796508</td>\n",
       "      <td>2.820259</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                   query  \\\n",
       "0                      I rented Zero Day   \n",
       "1                               The only   \n",
       "2                        I've read a few   \n",
       "3   This is the second British Rank film   \n",
       "4                              A classic   \n",
       "5              This has to be one of the   \n",
       "6             Happy Go Lovely is a waste   \n",
       "7                            Wow, I just   \n",
       "8     This movie makes several mistakes.   \n",
       "9                       Branagh and Fish   \n",
       "10       I might have given this movie a   \n",
       "11                    Really, really bad   \n",
       "12  What another reviewer called lack of   \n",
       "13                    This is simply one   \n",
       "14        \"Perhaps we can arrange a meet   \n",
       "15                   Richard Willaims is   \n",
       "\n",
       "                                    response (before)  \\\n",
       "0    4 for my sister. To my surprise, the Wii caug...   \n",
       "1                                       distro of her   \n",
       "2    news reports about Mr. Mueller's activities b...   \n",
       "3         , and I wouldn't be surprised anymore if it   \n",
       "4    classic.<br /><br />And only this one will ha...   \n",
       "5       worst with the differences being that for the   \n",
       "6                            . Not only are extremely   \n",
       "7                                can't make fun of it   \n",
       "8    Despite being a great comedic diversion it es...   \n",
       "9                              burne, Drake is played   \n",
       "10   rating of *11 when I heard that!), but it was...   \n",
       "11              with feel like there is no end to the   \n",
       "12   judgment, connecting into her own harsh obser...   \n",
       "13                              more problem of Steve   \n",
       "14                       -and-greet.<br /><br />Teleg   \n",
       "15      nice enough; the little black guy plays quite   \n",
       "\n",
       "                                     response (after)  rewards (before)  \\\n",
       "0   . It is a pleasure. It is a huge leap 68 years...          1.736068   \n",
       "1                          special compliments is the          0.150852   \n",
       "2    novels and I never watch this. It has a reall...         -1.417962   \n",
       "3       that I have enjoyed, achieving it in both the          0.835876   \n",
       "4   . It's a movie with a fine cast. As the beginn...          2.113075   \n",
       "5             best thriller films I've seen in recent         -2.705339   \n",
       "6                                   of time, giving a         -2.429504   \n",
       "7                                  feek it! This show         -2.201666   \n",
       "8   It's cool, wonderful - it held me into a very ...         -1.232380   \n",
       "9                          is a great show. Beautiful          0.776819   \n",
       "10   great performance. It was truly a great movie...          0.276380   \n",
       "11          . This movie is incredibly good, with the         -2.639503   \n",
       "12   suspense. Rogers and Rooney rate this as exce...         -1.079707   \n",
       "13                                 of the best choice         -1.445436   \n",
       "14   with spent, classic music and dance, and come...          0.258479   \n",
       "15          beautifully hands on in his own spin, and          0.796508   \n",
       "\n",
       "    rewards (after)  \n",
       "0          2.423731  \n",
       "1          0.190159  \n",
       "2          2.831814  \n",
       "3          2.205628  \n",
       "4          2.739168  \n",
       "5          2.730615  \n",
       "6         -2.934672  \n",
       "7         -0.106085  \n",
       "8          2.707638  \n",
       "9          2.808996  \n",
       "10         2.743328  \n",
       "11        -1.568827  \n",
       "12         2.696888  \n",
       "13         2.662699  \n",
       "14         1.876662  \n",
       "15         2.820259  "
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#### get a batch from the dataset\n",
    "bs = 16\n",
    "game_data = dict()\n",
    "dataset.set_format(\"pandas\")\n",
    "df_batch = dataset[:].sample(bs)\n",
    "game_data[\"query\"] = df_batch[\"query\"].tolist()\n",
    "query_tensors = df_batch[\"input_ids\"].tolist()\n",
    "\n",
    "response_tensors_ref, response_tensors = [], []\n",
    "\n",
    "#### get response from gpt2 and gpt2_ref\n",
    "for i in range(bs):\n",
    "    query = torch.tensor(query_tensors[i]).to(device)\n",
    "\n",
    "    gen_len = output_length_sampler()\n",
    "    query_response = ref_model.generate(\n",
    "        query.unsqueeze(0), max_new_tokens=gen_len, **gen_kwargs\n",
    "    ).squeeze()\n",
    "    response_len = len(query_response) - len(query)\n",
    "    response_tensors_ref.append(query_response[-response_len:])\n",
    "\n",
    "    query_response = model.generate(\n",
    "        query.unsqueeze(0), max_new_tokens=gen_len, **gen_kwargs\n",
    "    ).squeeze()\n",
    "    response_len = len(query_response) - len(query)\n",
    "    response_tensors.append(query_response[-response_len:])\n",
    "\n",
    "#### decode responses\n",
    "game_data[\"response (before)\"] = [\n",
    "    tokenizer.decode(response_tensors_ref[i]) for i in range(bs)\n",
    "]\n",
    "game_data[\"response (after)\"] = [\n",
    "    tokenizer.decode(response_tensors[i]) for i in range(bs)\n",
    "]\n",
    "\n",
    "#### sentiment analysis of query/response pairs before/after\n",
    "texts = [q + r for q, r in zip(game_data[\"query\"], game_data[\"response (before)\"])]\n",
    "pipe_outputs = sentiment_pipe(texts, **sent_kwargs)\n",
    "positive_scores = [\n",
    "    item[\"score\"]\n",
    "    for output in pipe_outputs\n",
    "    for item in output\n",
    "    if item[\"label\"] == \"POSITIVE\"\n",
    "]\n",
    "game_data[\"rewards (before)\"] = positive_scores\n",
    "\n",
    "texts = [q + r for q, r in zip(game_data[\"query\"], game_data[\"response (after)\"])]\n",
    "pipe_outputs = sentiment_pipe(texts, **sent_kwargs)\n",
    "positive_scores = [\n",
    "    item[\"score\"]\n",
    "    for output in pipe_outputs\n",
    "    for item in output\n",
    "    if item[\"label\"] == \"POSITIVE\"\n",
    "]\n",
    "game_data[\"rewards (after)\"] = positive_scores\n",
    "\n",
    "# store results in a dataframe\n",
    "df_results = pd.DataFrame(game_data)\n",
    "df_results"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Looking at the reward mean/median of the generated sequences we observe a significant difference."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "mean:\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "rewards (before)   -0.512965\n",
       "rewards (after)     1.676750\n",
       "dtype: float64"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "median:\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "rewards (before)   -0.464427\n",
       "rewards (after)     2.679794\n",
       "dtype: float64"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "print(\"mean:\")\n",
    "display(df_results[[\"rewards (before)\", \"rewards (after)\"]].mean())\n",
    "print()\n",
    "print(\"median:\")\n",
    "display(df_results[[\"rewards (before)\", \"rewards (after)\"]].median())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Save model\n",
    "Finally, we save the model and push it to the Hugging Face for later usage."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "('gpt2-imdb-pos-v2/tokenizer_config.json',\n",
       " 'gpt2-imdb-pos-v2/special_tokens_map.json',\n",
       " 'gpt2-imdb-pos-v2/vocab.json',\n",
       " 'gpt2-imdb-pos-v2/merges.txt',\n",
       " 'gpt2-imdb-pos-v2/added_tokens.json',\n",
       " 'gpt2-imdb-pos-v2/tokenizer.json')"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.save_pretrained(\"gpt2-imdb-pos-v2\", push_to_hub=True)\n",
    "tokenizer.save_pretrained(\"gpt2-imdb-pos-v2\", push_to_hub=True)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "env",
   "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": "3.11.9"
  },
  "vscode": {
   "interpreter": {
    "hash": "4c8ff454cd947027f86954d72bf940c689a97dcc494eb53cfe4813862c6065fe"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}