In-Short
- Mistral AI partners with NVIDIA to launch a 12B model named NeMo with a 128,000 token context window.
- NeMo is open-source under the Apache 2.0 license, promoting widespread adoption and research.
- The model features quantisation awareness for efficient FP8 inference and introduces a new tokeniser, Tekken.
- Mistral NeMo is designed for global use, excelling in multiple languages and available as an NVIDIA NIM inference microservice.
Summary of Mistral AI and NVIDIA’s NeMo Model
Mistral AI, in collaboration with NVIDIA, has unveiled NeMo, a powerful 12B model that sets a new standard in language model performance. NeMo’s impressive context window can handle up to 128,000 tokens, positioning it as a leader in reasoning, world knowledge, and coding accuracy within its size category. The partnership aims to make NeMo a straightforward replacement for systems using the Mistral 7B model, leveraging a standard architecture for ease of transition.
The model’s open-source availability under the Apache 2.0 license is a strategic move to foster research and practical application in the AI community. NeMo’s quantisation awareness during training is a standout feature, allowing for FP8 inference that maintains high performance while optimizing efficiency—a boon for organizations deploying large language models.
Mistral NeMo’s multilingual capabilities are robust, with training in function calling and strong performance in numerous languages, including English, French, German, Spanish, Italian, Portuguese, Chinese, Japanese, Korean, Arabic, and Hindi. The introduction of Tekken, a new tokeniser trained on over 100 languages, enhances compression efficiency for natural language text and source code, outperforming previous tokenisers and offering significant benefits for languages like Korean and Arabic.
Integrated as an NVIDIA NIM inference microservice, Mistral NeMo is readily accessible for those within NVIDIA’s AI ecosystem, simplifying deployment processes. This release marks a significant advancement in making cutting-edge AI models more accessible and versatile for a variety of applications across industries and research domains.
Further Information and Credits
For a deeper dive into the capabilities and applications of the Mistral NeMo model, readers are encouraged to visit the original source. Click here for more information.
Image credit: Photo by David Clode on Unsplash
