GLiClass Modern Base v2.0
| Property | Value |
|---|---|
| Parameter Count | 151M |
| Model Type | Zero-shot Classifier |
| Architecture | ModernBERT-base with layer-wise feature selection |
| Model URL | knowledgator/gliclass-modern-base-v2.0-init |
What is gliclass-modern-base-v2.0-init?
GLiClass Modern Base v2.0 is an efficient zero-shot classifier designed for sequence classification tasks. Built on the ModernBERT-base architecture, it implements layer-wise feature selection to better understand different levels of language. The model achieves comparable performance to cross-encoders while being more computationally efficient through single-pass classification.
Implementation Details
The model utilizes a sophisticated architecture that processes long sequences effectively through the ModernBERT-base backbone. It demonstrates strong performance across various benchmarks, achieving an F1 score of 0.8264 on IMDB and 0.6637 on AG_NEWS in zero-shot settings.
- Trained on synthetic and commercially-licensed data
- Supports multi-label classification
- Optimized for single forward path computation
- Implements layer-wise feature selection
Core Capabilities
- Topic Classification
- Sentiment Analysis
- RAG Pipeline Reranking
- Natural Language Inference Tasks
- Zero-shot Classification
Frequently Asked Questions
Q: What makes this model unique?
The model combines efficient computation with strong performance through its layer-wise feature selection mechanism and ModernBERT backbone, making it particularly suitable for production environments where computational efficiency is crucial.
Q: What are the recommended use cases?
The model excels in text classification tasks, sentiment analysis, and as a reranker in RAG pipelines. It's particularly effective when used for zero-shot classification scenarios where training data isn't available. The model can also handle NLI-type tasks when configured appropriately.
