Nvidia introduces Nemotron-4 4B Instruct, a small language model designed to enhance the conversational capabilities of game characters. The model runs directly on GeForce RTX graphics cards and is designed to deliver more natural interactions in games.
Nvidia has unveiled a new language model called Nemotron-4 4B Instruct, specifically designed for use in video games. The model is part of Nvidia's ACE technology and aims to improve the conversational abilities of game characters by more intuitively understanding player instructions and responding more naturally. A unique feature is that the model runs directly on GeForce RTX graphics cards.
The new language model is currently being demonstrated in "Mecha BREAK," a multiplayer mech game developed by Amazing Seasun Games. In a demo shown at Gamescom, players interact with a mechanic NPC who helps them select and customize mechs. In addition to the language model, other AI technologies are being used, including Nvidia Audio2Face-3D and OpenAI's Whisper for facial animation and speech recognition, as well as ElevenLabs for cloud-based speech output.
Small language models for games could run on over 100 million GeForce cards
Nemotron-4 4B Instruct is available as an Nvidia NIM microservice. The model is optimized for low memory usage and fast response times. Nvidia emphasizes that developers could benefit from over 100 million GeForce RTX PCs and laptops with this technology.
Other companies are also leveraging Nvidia's digital assistant technologies. Game developer Perfect World Games is expanding its Nvidia ACE demo "Legends" with AI-powered vision capabilities. In this demo, a game character named Yun Ni can recognize players and objects in the real world through the computer camera. The technology is based on GPT-4o and adds an augmented reality component to the gaming experience.
