Foundation Model Architectures · Evaluation, Reliability & Trustworthy AI

Project Unity

Indic language models and safety systems. Ganga is the open Hindi model family. UnityAI-Guard is the multilingual toxicity-detection system.

Themes
Foundation Model Architectures · Evaluation, Reliability & Trustworthy AI
Models
Ganga
Safety system
UnityAI-Guard

Ganga

Open Hindi language models on Hugging Face.

Ganga-1B

Base Hindi language model pretrained from scratch on curated public-domain and native-speaker-filtered text.

Hugging Face

Ganga-2-1B

Instruct-tuned follow-on of Ganga-1B.

Hugging Face

Ganga-en-hi-1B

Ganga-1B fine-tuned for English–Hindi translation.

Hugging Face

UnityAI-Guard

Multilingual toxicity detection and guardrails for Indian-language generation. EMNLP 2025 System Demonstrations.

UNITYAI-GUARD: Pioneering Toxicity Detection Across Low-Resource Indian Languages

Beniwal et al. EMNLP 2025 System Demonstrations.

PaperarXiv

All projects →All publications →