fastc

Preview：

Introduce：

fastc is a simple and lightweight text classification tool based on embedding of large language models. It focuses on CPU execution, using efficient models such as deepset/tinyroberta-6l-768d to generate embed. Text classification is realized by cosine similarity classification instead of fine tuning. It can also run multiple classifiers using the same model without additional overhead.

Stakeholders:
The target audience is developers and data scientists who need to classify text, especially if you have limited computing resources or want to quickly deploy a text classification model.
Usage Scenario Examples:

• Social media sentiment analysis to quickly determine the emotional tendencies of user comments.
• Product review categorization automatically categorizes user feedback as positive or negative.
• Subject classification of news articles, automatically distribute news to the corresponding subject column.

The features of the tool:

• Focus on CPU execution with efficient model generation embedding.
• Use cosine similarity for text classification without fine tuning.
• Supports multiple classifier execution, sharing the same model embeddings.
• Support model training and export for future use.
• Models can be published to the HuggingFace model library.
• Supports loading pre-trained models from directories or HuggingFace.
• Provides class prediction capabilities, including single item and batch prediction.

Steps for Use:

• Install the fastc library: Install fastc through liili, Python’s package management tool.
• Prepare the data set: Collect and organize the text data used to train the classifier.
• Training model: Use the SentenceClassifier class provided by fastc to train the text classifier.
• Save the model: After the training is complete, use the save_model method to save the model for later use.
• Load model: The SentenceClassifier class loads the pre-trained model on the local or HuggingFace.
• Make predictions: Use liredict_one or liredict methods to make sentiment classification predictions for new text.

Tool’s Tabs: Text classification, cosine similarity