Gliclass Edge V3.0

🚀 GLiClass: Generalist and Lightweight Model for Sequence Classification

GLiClass is an efficient zero-shot classifier inspired by the GLiNER work. It achieves performance comparable to cross-encoders while being more compute-efficient, as classification is completed in a single forward pass. This model can be applied to topic classification, sentiment analysis, and serve as a reranker in RAG pipelines.

The model was trained on logical tasks to induce reasoning capabilities. LoRa adapters were employed to fine-tune the model without erasing prior knowledge.

✨ Features

• Efficient Zero-Shot Classification: Performs classification in a single forward pass, offering comparable performance to cross-encoders with lower computational costs.
• Versatile Applications: Can be used for topic classification, sentiment analysis, and as a reranker in RAG pipelines.
• Reasoning Induction: Trained on logical tasks to induce reasoning capabilities.
• Knowledge Preservation: Fine-tuned using LoRa adapters without destroying previous knowledge.

📦 Installation

First, you need to install the GLiClass library:

pip install gliclass
pip install -U transformers>=4.48.0

💻 Usage Examples

Basic Usage

Initialize a model and a pipeline:

from gliclass import GLiClassModel, ZeroShotClassificationPipeline
from transformers import AutoTokenizer

model = GLiClassModel.from_pretrained("knowledgator/gliclass-edge-v3.0")
tokenizer = AutoTokenizer.from_pretrained("knowledgator/gliclass-edge-v3.0", add_prefix_space=True)
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"])

Advanced Usage

If you want to use it for NLI type of tasks, represent your premise as a text and hypothesis as a label. You can put several hypotheses, but the model works best with a single input hypothesis.

# Initialize model and multi-label pipeline
text = "The cat slept on the windowsill all afternoon"
labels = ["The cat was awake and playing outside."]
results = pipeline(text, labels, threshold=0.0)[0]
print(results)

📚 Documentation

LoRA Parameters

GLiClass-V3 Models

Benchmarks

Below are the F1 scores on several text classification datasets. All tested models were not fine-tuned on those datasets and were tested in a zero-shot setting.

GLiClass-V3

Previous GLiClass models

Cross-Encoders

Inference Speed

Each model was tested on examples with 64, 256, and 512 tokens in text and 1, 2, 4, 8, 16, 32, 64, and 128 labels on an a6000 GPU. Then, scores were averaged across text lengths.

📄 License

This project is licensed under the Apache-2.0 license.