Mistral AI and NVIDIA today released Mistral NeMo 12B, a new state-of-the-art language model that developers can easily customize and deploy in enterprise applications supporting chatbots, multilingual tasks, coding and summarization.
Combining Mistral AI's training data expertise with NVIDIA's optimized hardware and software ecosystem, Mistral NeMo models deliver high performance across a range of applications.
“We are fortunate to collaborate with the NVIDIA team and leverage their best-in-class hardware and software,” said Guillaume Lample, co-founder and chief scientist at Mistral AI. “With the introduction of NVIDIA AI Enterprise, we have worked together to develop models with unprecedented accuracy, flexibility, high efficiency and enterprise-grade support and security.”
Mistral NeMo was trained on the NVIDIA DGX Cloud AI platform, which provides dedicated, scalable access to the latest NVIDIA architecture.
NVIDIA TensorRT-LLM, which accelerates inference performance on large-scale language models, and the NVIDIA NeMo development platform for building custom generative AI models were also used to advance and optimize the process.
This collaboration highlights NVIDIA's commitment to supporting the Model Builder ecosystem.
Unprecedented precision, flexibility and efficiency
Excelling in multi-turn conversation, mathematics, common sense reasoning, world knowledge and coding, this enterprise-grade AI model delivers accurate and reliable performance across a range of tasks.
With a context length of 128K, Mistral NeMo processes a wide range of complex information more consistently and accurately, ensuring context-appropriate output.
Released under the Apache 2.0 license to foster innovation and support the broader AI community, Mistral NeMo is a 12 billion parameter model. Additionally, the model uses the FP8 data format for model inference, reducing memory size and speeding up deployment without compromising accuracy.
This means that the models learn tasks better and handle a variety of scenarios more effectively, making them ideal for enterprise use cases.
Mistral NeMo is packaged as an NVIDIA NIM inference microservice and provides performance-optimized inference powered by the NVIDIA TensorRT-LLM engine.
This containerized format allows for easy deployment anywhere and increased flexibility for a wide range of applications.
As a result, models can be deployed anywhere in minutes instead of days.
NIM features enterprise-grade software that is part of NVIDIA AI Enterprise, with dedicated feature branches, rigorous validation processes and enterprise-grade security and support.
Includes comprehensive support, direct access to NVIDIA AI experts, and defined service level agreements to ensure reliable, consistent performance.
The open model license allows companies to seamlessly integrate Mistral NeMo into their commercial applications.
The Mistral NeMo NIM is designed to fit into the memory of a single NVIDIA L40S, NVIDIA GeForce RTX 4090, or NVIDIA RTX 4500 GPU, delivering high efficiency, low computing costs, and enhanced security and privacy.
Advanced Model Development and Customization
The combined expertise of Mistral AI and NVIDIA engineers has optimized training and inference for Mistral NeMo.
The model is trained using Mistral AI's expertise, particularly in multi-language, code and multi-turn content, and benefits from fast training with NVIDIA's full stack.
It is designed to deliver optimal performance by leveraging efficient model parallelism techniques, scalability, and mixed precision with Megatron-LM.
The model was trained using Megatron-LM, part of NVIDIA NeMo. The 3,072 H100 80GB Tensor Core GPUs on DGX Cloud are comprised of NVIDIA AI architecture, including accelerated computing, network fabric and software for training efficiency.
Availability and Deployment
With the flexibility to run anywhere – in the cloud, in the data center, or on an RTX workstation – Mistral NeMo is poised to revolutionize AI applications across platforms.
Experience Mistral NeMo as an NVIDIA NIM today at ai.nvidia.com, with a downloadable NIM coming soon.
look News About software product information.
