Promptzl Documentation

Turn state-of-the-art LLMs into zero-shot PyTorch classifiers in just a few lines of code.

Promptzl offers:

• 🤖 Zero-shot classification with LLMs

• 🤗 Turning causal and masked LMs into classifiers without any training

• 📦 Batch processing on your device for efficiency

• 🚀 Speed-up over calling an online API

• 🔎 Transparency and accessibility by using the model locally

• 📈 Distribution over labels

• ✂️ No need to extract the predictions from the answer.

Links

GitHub | PyPI | GitHub Issue Tracker

Installation

pip install -U promptzl

How does it Work?

Language models predict a token given a specific context by calculating a distribution over the vocabulary. When classifying sentences, only a few tokens are relevant for the classification task. Extracting the tokens’ logits and forming a distribution over them allows turning the LLM into a classifier. This is what Promptzl does. A simple example can be found in Tutorial - Basic Usage.

Background

• Background

Documentation

• Prompt Classes
• LM Classifiers
• Utilities

Tutorials

• Tutorial - Basic Usage
• Tutorial - Causal-Language-Model-Based Classifiers
• Tutorial - Masked-Language-Model-Based Classifiers

Benchmark

• Benchmark