Researchers baked 3x inference speedups directly into LLM weights — without speculative decoding

Researchers from the University of Maryland, Lawrence Livermore, Columbia and TogetherAI have developed a training technique that triples LLM inference speed without auxiliary models or infrastructure ...
MediaNama

Here’s What Companies Unveiled at India AI Impact Summit 2026

AI companies like Sarvam AI, BharatGen, Gnani.ai etc launched a range of domain-specific AI systems at the India AI Impact ...