Quick tour

Get up and running with 🤗 Transformers! Whether you’re a developer or an everyday user, this quick tour will help you get started and show you how to use the pipeline() for inference, load a pretrained model and preprocessor with an AutoClass, and quickly train a model with PyTorch or TensorFlow. If you’re a beginner, we recommend checking out our tutorials or course next for more in-depth explanations of the concepts introduced here.

Before you begin, make sure you have all the necessary libraries installed:

!pip install transformers datasets

You’ll also need to install your preferred machine learning framework:

pip install torch

pip install tensorflow

Pipeline

The pipeline() is the easiest way to use a pretrained model for inference. You can use the pipeline() out-of-the-box for many tasks across different modalities. Take a look at the table below for some supported tasks:

Start by creating an instance of pipeline() and specifying a task you want to use it for. You can use the pipeline() for any of the previously mentioned tasks, and for a complete list of supported tasks, take a look at the pipeline API reference. In this guide though, you’ll use the pipeline() for sentiment analysis as an example:

>>> from transformers import pipeline

>>> classifier = pipeline("sentiment-analysis")

The pipeline() downloads and caches a default pretrained model and tokenizer for sentiment analysis. Now you can use the classifier on your target text:

>>> classifier("We are very happy to show you the 🤗 Transformers library.")
[{'label': 'POSITIVE', 'score': 0.9998}]

If you have more than one input, pass your inputs as a list to the pipeline() to return a list of dictionaries:

>>> results = classifier(["We are very happy to show you the 🤗 Transformers library.", "We hope you don't hate it."])
>>> for result in results:
...     print(f"label: {result['label']}, with score: {round(result['score'], 4)}")
label: POSITIVE, with score: 0.9998
label: NEGATIVE, with score: 0.5309

The pipeline() can also iterate over an entire dataset for any task you like. For this example, let’s choose automatic speech recognition as our task:

>>> import torch
>>> from transformers import pipeline

>>> speech_recognizer = pipeline("automatic-speech-recognition", model="facebook/wav2vec2-base-960h")

Load an audio dataset (see the 🤗 Datasets Quick Start for more details) you’d like to iterate over. For example, load the MInDS-14 dataset:

>>> from datasets import load_dataset, Audio

>>> dataset = load_dataset("PolyAI/minds14", name="en-US", split="train")

You need to make sure the sampling rate of the dataset matches the sampling rate facebook/wav2vec2-base-960h was trained on:

>>> dataset = dataset.cast_column("audio", Audio(sampling_rate=speech_recognizer.feature_extractor.sampling_rate))

The audio files are automatically loaded and resampled when calling the "audio" column. Extract the raw waveform arrays from the first 4 samples and pass it as a list to the pipeline:

>>> result = speech_recognizer(dataset[:4]["audio"])
>>> print([d["text"] for d in result])
['I WOULD LIKE TO SET UP A JOINT ACCOUNT WITH MY PARTNER HOW DO I PROCEED WITH DOING THAT', "FODING HOW I'D SET UP A JOIN TO HET WITH MY WIFE AND WHERE THE AP MIGHT BE", "I I'D LIKE TOY SET UP A JOINT ACCOUNT WITH MY PARTNER I'M NOT SEEING THE OPTION TO DO IT ON THE AP SO I CALLED IN TO GET SOME HELP CAN I JUST DO IT OVER THE PHONE WITH YOU AND GIVE YOU THE INFORMATION OR SHOULD I DO IT IN THE AP AND I'M MISSING SOMETHING UQUETTE HAD PREFERRED TO JUST DO IT OVER THE PHONE OF POSSIBLE THINGS", 'HOW DO I THURN A JOIN A COUNT']

For larger datasets where the inputs are big (like in speech or vision), you’ll want to pass a generator instead of a list to load all the inputs in memory. Take a look at the pipeline API reference for more information.

Use another model and tokenizer in the pipeline

The pipeline() can accommodate any model from the Hub, making it easy to adapt the pipeline() for other use-cases. For example, if you’d like a model capable of handling French text, use the tags on the Hub to filter for an appropriate model. The top filtered result returns a multilingual BERT model finetuned for sentiment analysis you can use for French text:

>>> model_name = "nlptown/bert-base-multilingual-uncased-sentiment"

>>> from transformers import AutoTokenizer, AutoModelForSequenceClassification

>>> model = AutoModelForSequenceClassification.from_pretrained(model_name)
>>> tokenizer = AutoTokenizer.from_pretrained(model_name)

>>> from transformers import AutoTokenizer, TFAutoModelForSequenceClassification

>>> model = TFAutoModelForSequenceClassification.from_pretrained(model_name)
>>> tokenizer = AutoTokenizer.from_pretrained(model_name)

Specify the model and tokenizer in the pipeline(), and now you can apply the classifier on French text:

>>> classifier = pipeline("sentiment-analysis", model=model, tokenizer=tokenizer)
>>> classifier("Nous sommes très heureux de vous présenter la bibliothèque 🤗 Transformers.")
[{'label': '5 stars', 'score': 0.7273}]

If you can’t find a model for your use-case, you’ll need to finetune a pretrained model on your data. Take a look at our finetuning tutorial to learn how. Finally, after you’ve finetuned your pretrained model, please consider sharing the model with the community on the Hub to democratize machine learning for everyone! 🤗

AutoClass

Under the hood, the AutoModelForSequenceClassification and AutoTokenizer classes work together to power the pipeline() you used above. An AutoClass 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 preprocessing class.

Let’s return to the example from the previous section and see how you can use the AutoClass to replicate the results of the pipeline().

AutoTokenizer

A tokenizer is responsible for preprocessing text into an array of numbers as inputs to a model. There are multiple rules that govern the tokenization process, including how to split a word and at what level words should be split (learn more about tokenization in the tokenizer summary). The most important thing to remember is you need to instantiate a tokenizer with the same model name to ensure you’re using the same tokenization rules a model was pretrained with.

Load a tokenizer with AutoTokenizer:

>>> from transformers import AutoTokenizer

>>> model_name = "nlptown/bert-base-multilingual-uncased-sentiment"
>>> tokenizer = AutoTokenizer.from_pretrained(model_name)

Pass your text to the tokenizer:

>>> encoding = tokenizer("We are very happy to show you the 🤗 Transformers library.")
>>> print(encoding)
{'input_ids': [101, 11312, 10320, 12495, 19308, 10114, 11391, 10855, 10103, 100, 58263, 13299, 119, 102],
 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]}

The tokenizer returns a dictionary containing:

• input_ids: numerical representations of your tokens.
• attention_mask: indicates which tokens should be attended to.

A tokenizer can also accept a list of inputs, and pad and truncate the text to return a batch with uniform length:

>>> pt_batch = tokenizer(
...     ["We are very happy to show you the 🤗 Transformers library.", "We hope you don't hate it."],
...     padding=True,
...     truncation=True,
...     max_length=512,
...     return_tensors="pt",
... )

>>> tf_batch = tokenizer(
...     ["We are very happy to show you the 🤗 Transformers library.", "We hope you don't hate it."],
...     padding=True,
...     truncation=True,
...     max_length=512,
...     return_tensors="tf",
... )

AutoModel

>>> from transformers import AutoModelForSequenceClassification

>>> model_name = "nlptown/bert-base-multilingual-uncased-sentiment"
>>> pt_model = AutoModelForSequenceClassification.from_pretrained(model_name)

>>> pt_outputs = pt_model(**pt_batch)

>>> from torch import nn

>>> pt_predictions = nn.functional.softmax(pt_outputs.logits, dim=-1)
>>> print(pt_predictions)
tensor([[0.0021, 0.0018, 0.0115, 0.2121, 0.7725],
        [0.2084, 0.1826, 0.1969, 0.1755, 0.2365]], grad_fn=<SoftmaxBackward0>)

>>> from transformers import TFAutoModelForSequenceClassification

>>> model_name = "nlptown/bert-base-multilingual-uncased-sentiment"
>>> tf_model = TFAutoModelForSequenceClassification.from_pretrained(model_name)

>>> tf_outputs = tf_model(tf_batch)

>>> import tensorflow as tf

>>> tf_predictions = tf.nn.softmax(tf_outputs.logits, axis=-1)
>>> tf_predictions

Save a model

>>> pt_save_directory = "./pt_save_pretrained"
>>> tokenizer.save_pretrained(pt_save_directory)
>>> pt_model.save_pretrained(pt_save_directory)

>>> pt_model = AutoModelForSequenceClassification.from_pretrained("./pt_save_pretrained")

>>> tf_save_directory = "./tf_save_pretrained"
>>> tokenizer.save_pretrained(tf_save_directory)
>>> tf_model.save_pretrained(tf_save_directory)

>>> tf_model = TFAutoModelForSequenceClassification.from_pretrained("./tf_save_pretrained")

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:

>>> from transformers import AutoModel

>>> tokenizer = AutoTokenizer.from_pretrained(tf_save_directory)
>>> pt_model = AutoModelForSequenceClassification.from_pretrained(tf_save_directory, from_tf=True)

>>> from transformers import TFAutoModel

>>> tokenizer = AutoTokenizer.from_pretrained(pt_save_directory)
>>> tf_model = TFAutoModelForSequenceClassification.from_pretrained(pt_save_directory, from_pt=True)

Custom model builds

You can modify the model’s configuration class to change how a model is built. The configuration specifies a model’s attributes, such as the number of hidden layers or attention heads. You start from scratch when you initialize a model from a custom configuration class. The model attributes are randomly initialized, and you’ll need to train the model before you can use it to get meaningful results.

Start by importing AutoConfig, and then load the pretrained model you want to modify. Within AutoConfig.from_pretrained(), you can specify the attribute you want to change, such as the number of attention heads:

>>> from transformers import AutoConfig

>>> my_config = AutoConfig.from_pretrained("distilbert-base-uncased", n_heads=12)

>>> from transformers import AutoModel

>>> my_model = AutoModel.from_config(my_config)

>>> from transformers import TFAutoModel

>>> my_model = TFAutoModel.from_config(my_config)

Take a look at the Create a custom architecture guide for more information about building custom configurations.

Trainer - a PyTorch optimized training loop

All models are a standard torch.nn.Module so you can use them in any typical training loop. While you can write your own training loop, 🤗 Transformers provides a Trainer class for PyTorch, which contains the basic training loop and adds additional functionality for features like distributed training, mixed precision, and more.

Depending on your task, you’ll typically pass the following parameters to Trainer:

1. A PreTrainedModel or a torch.nn.Module:

   Copied

   >>> from transformers import AutoModelForSequenceClassification
   
   >>> model = AutoModelForSequenceClassification.from_pretrained("distilbert-base-uncased")

2. TrainingArguments contains the model hyperparameters you can change like learning rate, batch size, and the number of epochs to train for. The default values are used if you don’t specify any training arguments:

   Copied

   >>> from transformers import TrainingArguments
   
   >>> training_args = TrainingArguments(
   ...     output_dir="path/to/save/folder/",
   ...     learning_rate=2e-5,
   ...     per_device_train_batch_size=8,
   ...     per_device_eval_batch_size=8,
   ...     num_train_epochs=2,
   ... )

3. A preprocessing class like a tokenizer, image processor, feature extractor, or processor:

   Copied

   >>> from transformers import AutoTokenizer
   
   >>> tokenizer = AutoTokenizer.from_pretrained("distilbert-base-uncased")

4. Load a dataset:

   Copied

   >>> from datasets import load_dataset
   
   >>> dataset = load_dataset("rotten_tomatoes")  # doctest: +IGNORE_RESULT

5. Create a function to tokenize the dataset, and apply it over the entire dataset with map:

   Copied

   >>> def tokenize_dataset(dataset):
   ...     return tokenizer(dataset["text"])

dataset = dataset.map(tokenize_dataset, batched=True)

Copied

6. A DataCollatorWithPadding to create a batch of examples from your dataset:

   Copied

   >>> from transformers import DataCollatorWithPadding
   
   >>> data_collator = DataCollatorWithPadding(tokenizer=tokenizer)

Now gather all these classes in Trainer:

>>> from transformers import Trainer

>>> trainer = Trainer(
...     model=model,
...     args=training_args,
...     train_dataset=dataset["train"],
...     eval_dataset=dataset["test"],
...     tokenizer=tokenizer,
...     data_collator=data_collator,
... )  # doctest: +SKIP

When you’re ready, call train() to start training:

>>> trainer.train()

You can customize the training loop behavior by subclassing the methods inside Trainer. This allows you to customize features such as the loss function, optimizer, and scheduler. Take a look at the Trainer reference for which methods can be subclassed.

The other way to customize the training loop is by using Callbacks. You can use callbacks to integrate with other libraries and inspect the training loop to report on progress or stop the training early. Callbacks do not modify anything in the training loop itself. To customize something like the loss function, you need to subclass the Trainer instead.

Train with TensorFlow

All models are a standard tf.keras.Model so they can be trained in TensorFlow with the Keras API. 🤗 Transformers provides the prepare_tf_dataset() method to easily load your dataset as a tf.data.Dataset so you can start training right away with Keras’ compile and fit methods.

1. You’ll start with a TFPreTrainedModel or a tf.keras.Model:

   Copied

   >>> from transformers import TFAutoModelForSequenceClassification
   
   >>> model = TFAutoModelForSequenceClassification.from_pretrained("distilbert-base-uncased")

2. A preprocessing class like a tokenizer, image processor, feature extractor, or processor:

   Copied

   >>> from transformers import AutoTokenizer
   
   >>> tokenizer = AutoTokenizer.from_pretrained("distilbert-base-uncased")

3. Create a function to tokenize the dataset:

   Copied

   >>> def tokenize_dataset(dataset):
   ...     return tokenizer(dataset["text"])  # doctest: +SKIP

4. Apply the tokenizer over the entire dataset with map and then pass the dataset and tokenizer to prepare_tf_dataset(). You can also change the batch size and shuffle the dataset here if you’d like:

   Copied

   >>> dataset = dataset.map(tokenize_dataset)  # doctest: +SKIP
   >>> tf_dataset = model.prepare_tf_dataset(
   ...     dataset, batch_size=16, shuffle=True, tokenizer=tokenizer
   ... )  # doctest: +SKIP

5. When you’re ready, you can call compile and fit to start training:

   Copied

   >>> from tensorflow.keras.optimizers import Adam
   
   >>> model.compile(optimizer=Adam(3e-5))
   >>> model.fit(dataset)  # doctest: +SKIP

What's next?

Now that you’ve completed the 🤗 Transformers quick tour, check out our guides and learn how to do more specific things like writing a custom model, fine-tuning a model for a task, and how to train a model with a script. If you’re interested in learning more about 🤗 Transformers core concepts, grab a cup of coffee and take a look at our Conceptual Guides!