Gliclass Large V1.0 Init

⭐ GLiClass: Generalist and Lightweight Model for Sequence Classification

GLiClass is an efficient zero - shot classifier inspired by the GLiNER work. It achieves the same performance as a cross - encoder but is more compute - efficient since classification is done in a single forward pass. This model can be applied to topic classification, sentiment analysis, and serve as a reranker in RAG pipelines. Trained on synthetic data, it is suitable for commercial applications. Note that this model has not been additionally fine - tuned on any dataset other than the initial one (MoritzLaurer/synthetic_zeroshot_mixtral_v0.1).

🚀 Quick Start

Installation

First, install the GLiClass library:

pip install gliclass

Usage

Then, initialize the model and the pipeline:

from gliclass import GLiClassModel, ZeroShotClassificationPipeline
from transformers import AutoTokenizer

model = GLiClassModel.from_pretrained("knowledgator/gliclass-large-v1.0-init")
tokenizer = AutoTokenizer.from_pretrained("knowledgator/gliclass-large-v1.0-init")

pipeline = ZeroShotClassificationPipeline(model, tokenizer, classification_type='multi-label', device='cuda:0')

text = "One day I will see the world!"
labels = ["travel", "dreams", "sport", "science", "politics"]
results = pipeline(text, labels, threshold=0.5)[0] #because we have one text

for result in results:
 print(result["label"], "=>", result["score"])

✨ Features

• Efficient Zero - Shot Classification: Performs classification in a single forward pass, offering comparable performance to cross - encoders with better compute efficiency.
• Versatile Applications: Can be used for topic classification, sentiment analysis, and as a reranker in RAG pipelines.
• Commercial - Ready: Trained on synthetic data, it can be used in commercial applications.

📚 Documentation

Benchmarks

The table below shows the F1 scores of several text classification models on different datasets. All models were tested in a zero - shot setting without fine - tuning on these datasets.

📄 License

This project is licensed under the Apache - 2.0 license.

📦 Model Information