HuggingFace Transformers

`pipeline()`

pipeline() function is the most high level API of 🤗 Transformers library. It connects a model with its necessary preprocessing and post-processing steps. There is three main steps involved when you pass some text to pipeline(): 1. The text is processed into a format the model can understand, i.e. numbers. 2. The processed inputs are passed to the model. 3. The predictions of the model are post-processed to get back get intelligible (i.e. human readable) answers.

flowchart LR
    A["Text"]
    A --> B(["Pre-processing"])
    B --> C(["Model"])
    C --> D(["Post-processing"])
    D --> E["Text"]

Tasks

Tasks, or pipeline() types, describe the “shape” of each model’s API, (i.e. inputs and outputs). pipeline() function support different modalities, such as text, audio, video, and even multi modal models.

Text pipelines

Sentiment Analysis

sentiment-analysis task returns one of the two labels, POSITIVE or NEGATIVE with a 'score representing the confident percentage of the returned label.

from transformers import pipeline

classifier = pipeline("sentiment-analysis")
classifier("I've been waiting for a HuggingFace course my whole life.")

>>> [{'label': 'POSITIVE', 'score': 0.9598047137260437}]

We can even pass several sentences inside a python’s list:

from transformers import pipeline

classifier = pipeline("sentiment-analysis")
classifier([
    "I've been waiting for a HuggingFace course my whole life.",
    "I hate this so much"
])

>>> [{'label': 'POSITIVE', 'score': 0.9598047137260437},
     {'label': 'NEGATIVE', 'score': 0.9994558095932007}]

Zero-shot Classifier

zero-shot-classifier is a more elaborated sentiment-analysis classifier where it can takes several labels and return the percentage of each levels, so we don’t have to rely on the labels of the pre-trained model. This pipeline is called zero-shot because you don’t need to fine-tune the model on our data. It can directly return probability scores of any list of labels we want.

from transformers import pipeline

classifier = pipeline("zero-shot-classification")
classifier(
    "This is a course about Transformers library",
    candidate_labels = ["education", "politics", "business"]
)

>>> {'sequence': 'This is a course about the Transformers library',
     'labels': ['education', 'business', 'politics'],
     'scores': [0.8445963859558105, 0.111976258456707, 0.043427448719739914]}

Text Generation

Generate text from a prompt

Arguments - num_return_sequences: control how many sequences are returned. - max_length: total length of the output text.

from transformers import pipeline

generator = pipeline("text-generation")
generator(
    "In this course, we will teach you how to",
    max_length = 30,
    num_return_sequences = 2)

>>> [{'generated_text': 'In this course, we will teach you how to manipulate the world and '
                        'move your mental and physical capabilities to your advantage.'},
     {'generated_text': 'In this course, we will teach you how to become an expert and '
                        'practice realtime, and with a hands on experience on both real '
                        'time and real'}]

Mask Filling

fill-mask helps to fill in the blanks in a given text, where mask token are positioned. By default, the mask token is <mask>. Other models might have different mask token.

arguments - top_k: controls how many possibilities we want to display.

from transformers import pipeline

unmasker = pipeline("fill-mask")
unmasker("This course will teach you all about <mask> models.", top_k = 2)

>>> [{'sequence': 'This course will teach you all about mathematical models.',
      'score': 0.19619831442832947,
      'token': 30412,
      'token_str': ' mathematical'},
     {'sequence': 'This course will teach you all about computational models.',
      'score': 0.04052725434303284,
      'token': 38163,
      'token_str': ' computational'}]

Pipeline tutorial

It exist text-generation: Generate text from a prompt text-classification: Classify text into predefined categories summarization: Create a shorter version of a text while preserving key information translation: Translate text from one language to another zero-shot-classification: Classify text without prior training on specific labels feature-extraction: Extract vector representations of text

Image pipelines

image-to-text: Generate text descriptions of images image-classification: Identify objects in an image object-detection: Locate and identify objects in images

Audio pipelines

automatic-speech-recognition: Convert speech to text audio-classification: Classify audio into categories text-to-speech: Convert text to spoken audio

Multimodal pipelines

image-text-to-text: Respond to an image based on a text prompt

Models

By default, pipeline() function select a particular pretrained model that has been fine-tuned for a specific task. The model is downloaded and cached when you call the pipeline() function. If you rerun the command, the cached model will be used instead and there is no need to download the model again.

However, we can use models avalaible on HuggingFace Models Hub that are designed for specific task. For example, if we would like to use SmolLM v2 from HuggingFace:

from transformers import pipeline

generator = pipeline("text-generation", model = "HuggingFaceTB/SmolLM2-360M")
generator(
    "In this course, we will teach you how to",
    max_length = 30,
    num_return_sequences = 2)

>>> [{'generated_text': 'In this course, we will teach you how to manipulate the world and '
                        'move your mental and physical capabilities to your advantage.'},
     {'generated_text': 'In this course, we will teach you how to become an expert and '
                        'practice realtime, and with a hands on experience on both real '
                        'time and real'}]