NVIDIA and Mistral AI Collaborate to Launch a New Open Model Family

Recently, NVIDIA announced a collaboration with Mistral AI, a large language model developer, aimed at accelerating the development and deployment of a series of new open-source models. According to the partnership agreement, NVIDIA will use its powerful platform to support Mistral's latest model family - Mistral 3.

The Mistral 3 series is described as open-source, multilingual, and multimodal, and is optimized for NVIDIA's supercomputing and edge platforms. The model uses a mixture of experts (MoE) architecture, which activates only the parts relevant to specific tasks, thus achieving more efficient and accurate deployment. NVIDIA stated that combining its GB200 NVL72 system with Mistral's MoE architecture will enable enterprises to deploy and scale large models more efficiently, fully utilizing "advanced parallelism" and hardware optimization.

The Mistral 3 model has 41 billion active parameters, 675 billion total parameters, and a large context window of 256K, providing good scalability, efficiency, and adaptability for enterprise AI workloads. Both parties also revealed that the new model will be fully launched on December 2nd and can be used in the cloud, data centers, and edge devices.

This collaboration continues the previous foundation between the two companies, including the Mistral NeMo 12B language model developed for chatbots and coding tasks. In addition, Mistral also released nine small language models, designed to help developers run AI on various devices. These models are designed to run on NVIDIA's hardware, including Spark, RTX PCs and laptops, as well as Jetson devices. Developers can access these models through AI frameworks such as Llama.cpp and Ollama.

The Mistral 3 family is open to researchers and developers, an initiative considered part of efforts to "democratize" cutting-edge AI. Notably, NVIDIA also announced an investment of $2 billion in chipmaker Synopsys within the same week, further highlighting the company's strength in AI and computing.

Key Points:

🌟 NVIDIA has partnered with Mistral AI to launch the open-source Mistral 3 model family, aiming to improve the efficiency of AI model development and deployment.

🔍 The Mistral 3 series uses a mixture of experts architecture, featuring 41 billion active parameters and 675 billion total parameters, suitable for enterprise AI applications.

💻 Nine small language models were also released, supporting various hardware and promoting the popularization and application of AI technology.