AI4Bharat, the AI research lab associated with IIT Madras, has recently launched Airavata, an instruction-tuned model tailored for the Hindi language. This model, derived from fine-tuning Sarvam AI’s OpenHathi, aims to enhance performance in assistive tasks through the incorporation of diverse, instruction-tuning Hindi datasets.
Airavata’s Development Approach
AI4Bharat emphasizes a sustainable approach to developing Airavata. The model’s development involves human-curated, license-friendly instruction-tuned datasets, steering clear of data generated from commercial models like GPT-4. This approach ensures cost-effectiveness and facilitates unrestricted usage in downstream applications due to the absence of licensing restrictions.
Also Read: India’s AI Leap 🇮🇳 : 6 LLMs that are Built in India
Addressing the Hindi Language Challenge
Leveraging IndicTrans2, an advanced open-source machine translation model for Indian languages, the team translates well-constructed English-supervised instruction-tuning datasets into Hindi. This method tackles the challenge of data scarcity for Hindi, aligning with AI4Bharat’s commitment to fostering advancements in Indic language models.
Comprehensive Release of Airavata
AI4Bharat not only introduced Airavata but also shared the instruction tuning datasets for the model. This step encourages innovation in the Indic language model domain, enabling researchers and developers to contribute to the evolution of Hindi language models.
The Larger Context
This release by AI4Bharat comes at a time when there is a growing interest in large language models worldwide. The recent focus has been on English-centric models, leaving a gap in support for Indian languages. The collaboration with Sarvam AI to launch OpenHathi laid the foundation, and now, with Airavata, AI4Bharat is taking a significant step forward in addressing the language model needs of Hindi.
Looking Ahead
As AI4Bharat continues to push boundaries in AI research, Airavata stands as a testament to the lab’s commitment to innovation and sustainability. The model’s performance on natural language understanding (NLU) tasks is noteworthy, indicating the potential for broader applications in various domains.
Also Read: Stability AI’s Small but Mighty Leap with Stable LM 2 1.6B Language Model
Our Say
The launch of Airavata is a milestone for AI4Bharat, paving the way for advancements in Indic language models. It aligns with the global shift towards more inclusive language models, emphasizing comprehensive solutions beyond English-centric approaches. Airavata’s impact on Hindi language processing could herald further advancements in the broader landscape of AI language models.
Follow us on Google News to stay updated with the latest innovations in the world of AI, Data Science, & GenAI.
