transformers_doc/quicktour.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Transformers installation\n",
    "! pip install transformers datasets\n",
    "# To install from source instead of the last release, comment the command above and uncomment the following one.\n",
    "# ! pip install git+https://github.com/huggingface/transformers.git"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Quick tour"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Get up and running with 🤗 Transformers! Start using the [pipeline()](https://huggingface.co/docs/transformers/master/en/main_classes/pipelines#transformers.pipeline) for rapid inference, and quickly load a pretrained model and tokenizer with an [AutoClass](https://huggingface.co/docs/transformers/master/en/./model_doc/auto) to solve your text, vision or audio task.\n",
    "\n",
    "<Tip>\n",
    "\n",
    "All code examples presented in the documentation have a toggle on the top left for PyTorch and TensorFlow. If\n",
    "not, the code is expected to work for both backends without any change.\n",
    "\n",
    "</Tip>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Pipeline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "[pipeline()](https://huggingface.co/docs/transformers/master/en/main_classes/pipelines#transformers.pipeline) is the easiest way to use a pretrained model for a given task."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "cellView": "form",
    "hide_input": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<iframe width=\"560\" height=\"315\" src=\"https://www.youtube.com/embed/tiZFewofSLM?rel=0&amp;controls=0&amp;showinfo=0\" frameborder=\"0\" allowfullscreen></iframe>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#@title\n",
    "from IPython.display import HTML\n",
    "\n",
    "HTML('<iframe width=\"560\" height=\"315\" src=\"https://www.youtube.com/embed/tiZFewofSLM?rel=0&amp;controls=0&amp;showinfo=0\" frameborder=\"0\" allowfullscreen></iframe>')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The [pipeline()](https://huggingface.co/docs/transformers/master/en/main_classes/pipelines#transformers.pipeline) supports many common tasks out-of-the-box:\n",
    "\n",
    "**Text**:\n",
    "* Sentiment analysis: classify the polarity of a given text.\n",
    "* Text generation (in English): generate text from a given input.\n",
    "* Name entity recognition (NER): label each word with the entity it represents (person, date, location, etc.).\n",
    "* Question answering: extract the answer from the context, given some context and a question.\n",
    "* Fill-mask: fill in the blank given a text with masked words.\n",
    "* Summarization: generate a summary of a long sequence of text or document.\n",
    "* Translation: translate text into another language.\n",
    "* Feature extraction: create a tensor representation of the text.\n",
    "\n",
    "**Image**:\n",
    "* Image classification: classify an image.\n",
    "* Image segmentation: classify every pixel in an image.\n",
    "* Object detection: detect objects within an image.\n",
    "\n",
    "**Audio**:\n",
    "* Audio classification: assign a label to a given segment of audio.\n",
    "* Automatic speech recognition (ASR): transcribe audio data into text.\n",
    "\n",
    "<Tip>\n",
    "\n",
    "For more details about the [pipeline()](https://huggingface.co/docs/transformers/master/en/main_classes/pipelines#transformers.pipeline) and associated tasks, refer to the documentation [here](https://huggingface.co/docs/transformers/master/en/./main_classes/pipelines).\n",
    "\n",
    "</Tip>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Pipeline usage"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In the following example, you will use the [pipeline()](https://huggingface.co/docs/transformers/master/en/main_classes/pipelines#transformers.pipeline) for sentiment analysis.\n",
    "\n",
    "Install the following dependencies if you haven't already:\n",
    "\n",
    "```bash\n",
    "pip install torch\n",
    "```\n",
    "\n",
    "```bash\n",
    "pip install tensorflow\n",
    "```\n",
    "\n",
    "Import [pipeline()](https://huggingface.co/docs/transformers/master/en/main_classes/pipelines#transformers.pipeline) and specify the task you want to complete:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import pipeline\n",
    "\n",
    "classifier = pipeline(\"sentiment-analysis\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The pipeline downloads and caches a default [pretrained model](https://huggingface.co/distilbert-base-uncased-finetuned-sst-2-english) and tokenizer for sentiment analysis. Now you can use the `classifier` on your target text:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[{\"label\": \"POSITIVE\", \"score\": 0.9998}]"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "classifier(\"We are very happy to show you the 🤗 Transformers library.\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "For more than one sentence, pass a list of sentences to the [pipeline()](https://huggingface.co/docs/transformers/master/en/main_classes/pipelines#transformers.pipeline) which returns a list of dictionaries:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "label: POSITIVE, with score: 0.9998\n",
       "label: NEGATIVE, with score: 0.5309"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "results = classifier([\"We are very happy to show you the 🤗 Transformers library.\", \"We hope you don't hate it.\"])\n",
    "for result in results:\n",
    "    print(f\"label: {result['label']}, with score: {round(result['score'], 4)}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The [pipeline()](https://huggingface.co/docs/transformers/master/en/main_classes/pipelines#transformers.pipeline) can also iterate over an entire dataset. Start by installing the [🤗 Datasets](https://huggingface.co/docs/datasets/) library:\n",
    "\n",
    "```bash\n",
    "pip install datasets \n",
    "```\n",
    "\n",
    "Create a [pipeline()](https://huggingface.co/docs/transformers/master/en/main_classes/pipelines#transformers.pipeline) with the task you want to solve for and the model you want to use. Set the `device` parameter to `0` to place the tensors on a CUDA device:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import pipeline\n",
    "\n",
    "speech_recognizer = pipeline(\"automatic-speech-recognition\", model=\"facebook/wav2vec2-base-960h\", device=0)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Next, load a dataset (see the 🤗 Datasets [Quick Start](https://huggingface.co/docs/datasets/quickstart.html) for more details) you'd like to iterate over. For example, let's load the [SUPERB](https://huggingface.co/datasets/superb) dataset:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import datasets\n",
    "\n",
    "dataset = datasets.load_dataset(\"superb\", name=\"asr\", split=\"test\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now you can iterate over the dataset with the pipeline. `KeyDataset` retrieves the item in the dictionary returned by the dataset:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{\"text\": \"HE HOPED THERE WOULD BE STEW FOR DINNER TURNIPS AND CARROTS AND BRUISED POTATOES AND FAT MUTTON PIECES TO BE LADLED OUT IN THICK PEPPERED FLOWER FAT AND SAUCE\"}"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from transformers.pipelines.pt_utils import KeyDataset\n",
    "from tqdm.auto import tqdm\n",
    "\n",
    "for out in tqdm(speech_recognizer(KeyDataset(dataset, \"file\"))):\n",
    "    print(out)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Use another model and tokenizer in the pipeline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The [pipeline()](https://huggingface.co/docs/transformers/master/en/main_classes/pipelines#transformers.pipeline) can accommodate any model from the [Model Hub](https://huggingface.co/models), making it easy to adapt the [pipeline()](https://huggingface.co/docs/transformers/master/en/main_classes/pipelines#transformers.pipeline) for other use-cases. For example, if you'd like a model capable of handling French text, use the tags on the Model Hub to filter for an appropriate model. The top filtered result returns a multilingual [BERT model](https://huggingface.co/nlptown/bert-base-multilingual-uncased-sentiment) fine-tuned for sentiment analysis. Great, let's use this model!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_name = \"nlptown/bert-base-multilingual-uncased-sentiment\""
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Use the [AutoModelForSequenceClassification](https://huggingface.co/docs/transformers/master/en/model_doc/auto#transformers.AutoModelForSequenceClassification) and ['AutoTokenizer'] to load the pretrained model and it's associated tokenizer (more on an `AutoClass` below):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import AutoTokenizer, AutoModelForSequenceClassification\n",
    "\n",
    "model = AutoModelForSequenceClassification.from_pretrained(model_name)\n",
    "tokenizer = AutoTokenizer.from_pretrained(model_name)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import AutoTokenizer, TFAutoModelForSequenceClassification\n",
    "\n",
    "model = TFAutoModelForSequenceClassification.from_pretrained(model_name)\n",
    "tokenizer = AutoTokenizer.from_pretrained(model_name)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Then you can specify the model and tokenizer in the [pipeline()](https://huggingface.co/docs/transformers/master/en/main_classes/pipelines#transformers.pipeline), and apply the `classifier` on your target text:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[{\"label\": \"5 stars\", \"score\": 0.7272651791572571}]"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "classifier = pipeline(\"sentiment-analysis\", model=model, tokenizer=tokenizer)\n",
    "classifier(\"Nous sommes très heureux de vous présenter la bibliothèque 🤗 Transformers.\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "If you can't find a model for your use-case, you will need to fine-tune a pretrained model on your data. Take a look at our [fine-tuning tutorial](https://huggingface.co/docs/transformers/master/en/./training) to learn how. Finally, after you've fine-tuned your pretrained model, please consider sharing it (see tutorial [here](https://huggingface.co/docs/transformers/master/en/./model_sharing)) with the community on the Model Hub to democratize NLP for everyone! 🤗"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## AutoClass"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "cellView": "form",
    "hide_input": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<iframe width=\"560\" height=\"315\" src=\"https://www.youtube.com/embed/AhChOFRegn4?rel=0&amp;controls=0&amp;showinfo=0\" frameborder=\"0\" allowfullscreen></iframe>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#@title\n",
    "from IPython.display import HTML\n",
    "\n",
    "HTML('<iframe width=\"560\" height=\"315\" src=\"https://www.youtube.com/embed/AhChOFRegn4?rel=0&amp;controls=0&amp;showinfo=0\" frameborder=\"0\" allowfullscreen></iframe>')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Under the hood, the [AutoModelForSequenceClassification](https://huggingface.co/docs/transformers/master/en/model_doc/auto#transformers.AutoModelForSequenceClassification) and [AutoTokenizer](https://huggingface.co/docs/transformers/master/en/model_doc/auto#transformers.AutoTokenizer) classes work together to power the [pipeline()](https://huggingface.co/docs/transformers/master/en/main_classes/pipelines#transformers.pipeline). An [AutoClass](https://huggingface.co/docs/transformers/master/en/./model_doc/auto) is a shortcut that automatically retrieves the architecture of a pretrained model from it's name or path. You only need to select the appropriate `AutoClass` for your task and it's associated tokenizer with [AutoTokenizer](https://huggingface.co/docs/transformers/master/en/model_doc/auto#transformers.AutoTokenizer). \n",
    "\n",
    "Let's return to our example and see how you can use the `AutoClass` to replicate the results of the [pipeline()](https://huggingface.co/docs/transformers/master/en/main_classes/pipelines#transformers.pipeline)."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### AutoTokenizer"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "A tokenizer is responsible for preprocessing text into a format that is understandable to the model. First, the tokenizer will split the text into words called *tokens*. There are multiple rules that govern the tokenization process, including how to split a word and at what level (learn more about tokenization [here](https://huggingface.co/docs/transformers/master/en/./tokenizer_summary)). The most important thing to remember though is you need to instantiate the tokenizer with the same model name to ensure you're using the same tokenization rules a model was pretrained with.\n",
    "\n",
    "Load a tokenizer with [AutoTokenizer](https://huggingface.co/docs/transformers/master/en/model_doc/auto#transformers.AutoTokenizer):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import AutoTokenizer\n",
    "\n",
    "model_name = \"nlptown/bert-base-multilingual-uncased-sentiment\"\n",
    "tokenizer = AutoTokenizer.from_pretrained(model_name)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Next, the tokenizer converts the tokens into numbers in order to construct a tensor as input to the model. This is known as the model's *vocabulary*.\n",
    "\n",
    "Pass your text to the tokenizer:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{\"input_ids\": [101, 2057, 2024, 2200, 3407, 2000, 2265, 2017, 1996, 100, 19081, 3075, 1012, 102],\n",
       " \"attention_mask\": [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]}"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "encoding = tokenizer(\"We are very happy to show you the 🤗 Transformers library.\")\n",
    "print(encoding)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The tokenizer will return a dictionary containing:\n",
    "\n",
    "* [input_ids](https://huggingface.co/docs/transformers/master/en/./glossary#input-ids): numerical representions of your tokens.\n",
    "* [atttention_mask](https://huggingface.co/docs/transformers/master/en/.glossary#attention-mask): indicates which tokens should be attended to.\n",
    "\n",
    "Just like the [pipeline()](https://huggingface.co/docs/transformers/master/en/main_classes/pipelines#transformers.pipeline), the tokenizer will accept a list of inputs. In addition, the tokenizer can also pad and truncate the text to return a batch with uniform length:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "pt_batch = tokenizer(\n",
    "    [\"We are very happy to show you the 🤗 Transformers library.\", \"We hope you don't hate it.\"],\n",
    "    padding=True,\n",
    "    truncation=True,\n",
    "    max_length=512,\n",
    "    return_tensors=\"pt\",\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "tf_batch = tokenizer(\n",
    "    [\"We are very happy to show you the 🤗 Transformers library.\", \"We hope you don't hate it.\"],\n",
    "    padding=True,\n",
    "    truncation=True,\n",
    "    max_length=512,\n",
    "    return_tensors=\"tf\",\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Read the [preprocessing](https://huggingface.co/docs/transformers/master/en/./preprocessing) tutorial for more details about tokenization."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### AutoModel"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "🤗 Transformers provides a simple and unified way to load pretrained instances. This means you can load an [AutoModel](https://huggingface.co/docs/transformers/master/en/model_doc/auto#transformers.AutoModel) like you would load an [AutoTokenizer](https://huggingface.co/docs/transformers/master/en/model_doc/auto#transformers.AutoTokenizer). The only difference is selecting the correct [AutoModel](https://huggingface.co/docs/transformers/master/en/model_doc/auto#transformers.AutoModel) for the task. Since you are doing text - or sequence - classification, load [AutoModelForSequenceClassification](https://huggingface.co/docs/transformers/master/en/model_doc/auto#transformers.AutoModelForSequenceClassification). The TensorFlow equivalent is simply [TFAutoModelForSequenceClassification](https://huggingface.co/docs/transformers/master/en/model_doc/auto#transformers.TFAutoModelForSequenceClassification):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import AutoModelForSequenceClassification\n",
    "\n",
    "model_name = \"nlptown/bert-base-multilingual-uncased-sentiment\"\n",
    "pt_model = AutoModelForSequenceClassification.from_pretrained(model_name)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import TFAutoModelForSequenceClassification\n",
    "\n",
    "model_name = \"nlptown/bert-base-multilingual-uncased-sentiment\"\n",
    "tf_model = TFAutoModelForSequenceClassification.from_pretrained(model_name)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<Tip>\n",
    "\n",
    "See the [task summary](https://huggingface.co/docs/transformers/master/en/./task_summary) for which [AutoModel](https://huggingface.co/docs/transformers/master/en/model_doc/auto#transformers.AutoModel) class to use for which task.\n",
    "\n",
    "</Tip>\n",
    "\n",
    "Now you can pass your preprocessed batch of inputs directly to the model. If you are using a PyTorch model, unpack the dictionary by adding `**`. For TensorFlow models, pass the dictionary keys directly to the tensors:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "pt_outputs = pt_model(**pt_batch)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "tf_outputs = tf_model(tf_batch)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The model outputs the final activations in the `logits` attribute. Apply the softmax function to the `logits` to retrieve the probabilities:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[2.2043e-04, 9.9978e-01],\n",
       "        [5.3086e-01, 4.6914e-01]], grad_fn=<SoftmaxBackward>)"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from torch import nn\n",
    "\n",
    "pt_predictions = nn.functional.softmax(pt_outputs.logits, dim=-1)\n",
    "print(pt_predictions)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tf.Tensor(\n",
       "[[2.2043e-04 9.9978e-01]\n",
       " [5.3086e-01 4.6914e-01]], shape=(2, 2), dtype=float32)"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "\n",
    "tf_predictions = tf.nn.softmax(tf_outputs.logits, axis=-1)\n",
    "print(tf_predictions)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<Tip>\n",
    "\n",
    "All 🤗 Transformers models (PyTorch or TensorFlow) outputs the tensors *before* the final activation\n",
    "function (like softmax) because the final activation function is often fused with the loss.\n",
    "\n",
    "</Tip>\n",
    "\n",
    "Models are a standard [`torch.nn.Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) or a [`tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model) so you can use them in your usual training loop. However, to make things easier, 🤗 Transformers provides a [Trainer](https://huggingface.co/docs/transformers/master/en/main_classes/trainer#transformers.Trainer) class for PyTorch that adds functionality for distributed training, mixed precision, and more. For TensorFlow, you can use the `fit` method from [Keras](https://keras.io/). Refer to the [training tutorial](https://huggingface.co/docs/transformers/master/en/./training) for more details.\n",
    "\n",
    "<Tip>\n",
    "\n",
    "🤗 Transformers model outputs are special dataclasses so their attributes are autocompleted in an IDE.\n",
    "The model outputs also behave like a tuple or a dictionary (e.g., you can index with an integer, a slice or a string) in which case the attributes that are `None` are ignored.\n",
    "\n",
    "</Tip>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Save a model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Once your model is fine-tuned, you can save it with its tokenizer using [PreTrainedModel.save_pretrained()](https://huggingface.co/docs/transformers/master/en/main_classes/model#transformers.PreTrainedModel.save_pretrained):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "pt_save_directory = \"./pt_save_pretrained\"\n",
    "tokenizer.save_pretrained(pt_save_directory)\n",
    "pt_model.save_pretrained(pt_save_directory)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "tf_save_directory = \"./tf_save_pretrained\"\n",
    "tokenizer.save_pretrained(tf_save_directory)\n",
    "tf_model.save_pretrained(tf_save_directory)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "When you are ready to use the model again, reload it with [PreTrainedModel.from_pretrained()](https://huggingface.co/docs/transformers/master/en/main_classes/model#transformers.PreTrainedModel.from_pretrained):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "pt_model = AutoModelForSequenceClassification.from_pretrained(\"./pt_save_pretrained\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "tf_model = TFAutoModelForSequenceClassification.from_pretrained(\"./tf_save_pretrained\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "One particularly cool 🤗 Transformers feature is the ability to save a model and reload it as either a PyTorch or TensorFlow model. The `from_pt` or `from_tf` parameter can convert the model from one framework to the other:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import AutoModel\n",
    "\n",
    "tokenizer = AutoTokenizer.from_pretrained(tf_save_directory)\n",
    "pt_model = AutoModelForSequenceClassification.from_pretrained(tf_save_directory, from_tf=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import TFAutoModel\n",
    "\n",
    "tokenizer = AutoTokenizer.from_pretrained(pt_save_directory)\n",
    "tf_model = TFAutoModelForSequenceClassification.from_pretrained(pt_save_directory, from_pt=True)"
   ]
  }
 ],
 "metadata": {},
 "nbformat": 4,
 "nbformat_minor": 4
}