深度共创 绝影与英伟达推进AGI开发

Core Insights - The collaboration between Jueying and NVIDIA aims to advance smart automotive technology towards the AGI era by leveraging the Drive AGX high-performance platform for core technology development and model quantization [1][2] - NVIDIA's TensorRT Edge-LLM framework is designed to meet the increasing demand for high-performance edge inference, facilitating the commercial deployment of automotive AI technology [1] Group 1: Autonomous Driving - Jueying has integrated visual language model (VLM) support with TensorRT Edge-LLM to enhance the system's cognitive and decision-making capabilities in complex traffic scenarios [1] - The simplified toolchain allows for rapid adaptation of models to mainstream automotive computing platforms like NVIDIA DRIVE AGX Orin and Thor [1] Group 2: Intelligent Cockpit - Jueying has developed a high-performance multimodal interactive system for intelligent cockpits by combining advanced KV-cache management, dynamic sequence scheduling, and lightweight deployment capabilities [2] - The optimized custom attention operators and multi-precision quantization techniques enable low-latency responses to multimodal inputs on resource-constrained automotive platforms [2] - The system supports natural language dialogue, intelligent scene recommendations, and seamless cross-screen command orchestration, enhancing user interaction and creating an immersive smart travel experience [2] Group 3: Innovation and Development - Long context support allows the system to accurately handle unstructured roads and rare traffic events, addressing challenging edge scenarios [2] - The cross-compilation and benchmarking capabilities of the toolkit enable rapid iteration and optimization of autonomous driving algorithms while ensuring safety [2] - The open-source release of TensorRT Edge LLM will lower the development threshold for automotive large models, and Jueying plans to deepen collaboration with NVIDIA to explore innovative AI applications in smart mobility [2]