fastText — Fast Text Classification & Word Vectors

fastText

fastText (by Meta AI) is a lightweight, extremely fast library for text classification and word embeddings. It uses subword (character n-gram) representations to handle morphology and OOV words naturally, and can train on millions of examples in seconds on a single CPU.

Key Features

• Blazing speed: trains 10M examples/second on a single CPU
• Subword models: handles OOV and morphologically rich languages
• Multi-label classification with __label__ syntax
• Pre-trained word vectors for 157 languages (CC + Wikipedia)
• Language identification: classify text to language in microseconds
• Python (pip install fasttext) and C++ CLI
• Quantized models for memory-constrained deployment

Quick Start

import fasttext

# Train a supervised classifier
model = fasttext.train_supervised(
    input="train.txt",  # format: "__label__positive text here"
    epoch=25, lr=1.0, wordNgrams=2
)

model.test("test.txt")  # (N, precision, recall)
print(model.predict("This product is amazing!"))  # → (("__label__positive",), array([0.999]))

# Load pre-trained language ID model
lid_model = fasttext.load_model("lid.176.bin")
print(lid_model.predict("Bonjour le monde"))  # → (('__label__fr',), array([0.999]))

Install via ai-supply

npx ai-supply add fasttext-text-classification-embeddings

Curated mirror of the open-source fastText (MIT). Get it from the source.