IndicBART
| Property | Value |
|---|---|
| Developer | AI4Bharat |
| Training Data | 452M sentences (9B tokens) |
| Languages Supported | 12 (11 Indian + English) |
| License | MIT |
| Paper | arXiv:2109.02903 |
What is IndicBART?
IndicBART is a groundbreaking multilingual sequence-to-sequence pre-trained model specifically designed for Indic languages and English. Built on the mBART architecture, it represents a significant advancement in natural language processing for Indian languages, offering support for 11 Indian languages while being more computationally efficient than its predecessors.
Implementation Details
The model utilizes a unique approach where all Indian languages are represented in Devanagari script to facilitate transfer learning. It's implemented using the transformers library and can be easily integrated into existing NLP pipelines. The model was trained using a text-infilling objective similar to mBART, making it suitable for various generation tasks.
- Smaller model size compared to mBART and mT5-base, requiring less computational resources
- Trained on extensive Indic corpora including Indian English content
- Unified script representation for enhanced cross-lingual learning
- Compatible with transformers library version 4.3.2 and above
Core Capabilities
- Machine Translation between supported languages
- Text Summarization in Indic languages
- Question Generation tasks
- Text completion and infilling
- Cross-lingual generation tasks
Frequently Asked Questions
Q: What makes this model unique?
IndicBART's uniqueness lies in its specialized focus on Indian languages, unified script approach, and efficient architecture that makes it more accessible for deployment. It supports languages not covered by mBART50 and mT5, while maintaining lower computational requirements.
Q: What are the recommended use cases?
The model is ideal for natural language generation tasks in Indian languages, including machine translation, text summarization, and question generation. It's particularly useful for organizations working with multiple Indian languages who need efficient, accurate language processing capabilities.
