Vikasit-AI-Writer-08B

About Model

The Sovereign MissionVikasit AI Writer 0.8B is a cornerstone of the Vikasit AI initiative by Chandorkar Technologies Pvt Ltd. Our mission is to provide India with a "Sovereign AI System"—one that is trained, optimized, and deployed locally to ensure data privacy, cultural alignment, and technological independence from global tech giants.Advanced Hybrid ArchitectureThis model is built upon the cutting-edge Qwen 3.5 foundation, featuring a sophisticated Hybrid Gated DeltaNet design.The Power of DeltaNet: By utilizing Gated Delta Networks alongside traditional Gated Attention, the model achieves $O(1)$ memory complexity for its linear attention layers.Efficiency at Scale: This hybrid approach allows the 0.8B model to punch significantly above its weight class, delivering logic and coherence usually reserved for models 2-3x its size.Key Technical PillarsMassive 256K Context: Unlike standard lightweight models, Vikasit AI 0.8B handles up to 256,000 tokens natively. This is ideal for analyzing long Indian legal statutes, multi-page business reports, or large code repositories.Ultra-Lightweight Footprint: Optimized via custom iMatrix IQ4_XS quantization, the model resides comfortably under 510MB. It is designed to run at high speeds on consumer CPUs and edge devices.Native Indian Identity: The model is fine-tuned to understand the nuances of Indian English and professional standards. It identifies as "Vikasit AI," a helpful assistant dedicated to the Indian ecosystem.Deployment & SovereigntyWe believe AI should be accessible to every Indian developer. By releasing this model under the Apache 2.0 license on AI Kosh and Hugging Face, we are ensuring that the building blocks of India's digital future remain open and sovereign.