Meta and Microsoft present the next generation of Llamas

Recent breakthroughs in artificial intelligence, and generative AI in particular, have captured the public’s imagination and shown what those developing these technologies have long known – they have the potential to help people do incredible things, creating a new era of economic and social opportunity, and giving. Individuals, creators, and businesses find new ways to express themselves and connect with people.

We believe that an open approach is the right approach for developing existing AI models, especially those in the generative space where technology is advancing rapidly. By making AI models available openly, it can benefit everyone. Giving corporations, startups, entrepreneurs, and researchers access to tools developed at a scale that would challenge them to build themselves, powered by computing power they may not have access to otherwise, will open up a world of opportunities for them to experiment and innovate in exciting ways, and to benefit economically and socially in their future. end of the day.

And we think it’s safer. Opening up access to existing AI models means that a generation of developers and researchers can rigorously test them, quickly identify and fix problems, as a community. By knowing how these tools are used by others, our teams can learn from them, improve those tools, and fix security vulnerabilities.

Meta has put exploratory research, open source, and collaboration with academic and industry partners at the heart of our AI efforts for more than a decade. We’ve seen firsthand how innovation in the open spaces can lead to technologies that benefit more people. Dozens of large language paradigms have already been released and are driving progress by developers and researchers. They are used by companies as key components of new AI-powered generation experiences. We’ve been blown away by the overwhelming demand for Llama 1 from researchers – with over 100,000 requests for access to the large language model – and the amazing things they’ve achieved with it. build on top of it.

We are now ready to open source the next version of Llama 2 and build it available Free shipping For research and commercial use. We include form weights, pre-tested form starting code, and chat optimized versions as well. As Satya Nadella announced on stage at Microsoft Inspire, We’re taking our partnership to the next level with Microsoft as our partner of choice for Llama 2 and expanding our generative AI efforts. Starting today, Llama 2 will be available in the Azure AI Model Catalog, enabling developers using Microsoft Azure to build with it and take advantage of cloud-native tools for content filtering and security features. It’s also optimized to run natively on Windows, giving developers a seamless workflow as they deliver generative AI experiences to customers across different platforms. Llama 2 will be available through Amazon Web Services (AWS), Hugging Face, and other service providers as well.

Individuals and businesses have benefited from the long-standing partnership between Microsoft and Meta. We filed a file An open ecosystem for interchangeable AI frameworks, and we have co-authored papers to advance the latest in AI technology. We have cooperated to expand the scope of adoption PyTorch – Today’s leading AI framework built by Meta and the AI ​​community – on Azure, and we’re among the founding members of the PyTorch Foundation. Microsoft and Meta recently joined a group of supporters who endorse Partnership in the framework of artificial intelligence For teamwork in creating and sharing synthetic media. Our partnership extends beyond artificial intelligence and extends to the other side as well Deliver immersive experiences For the future of work and play.

Now, with this expanded partnership, Microsoft and Meta are supporting an open approach to providing increased access to core AI technologies to benefit businesses globally. Meta and Microsoft just don’t believe in democratizing access to existing AI models. we’ve got Wide range of diverse supporters Around the world believe in this approach too – including companies that gave us early feedback and are excited to build new products with Llama 2, cloud providers who will include Llama 2 in their offerings to customers, research institutions that collaborate with us in the secure and responsible deployment of large generative models, and people across Technology, academia and politics who see the benefits as we do.

Focus on responsibility

Our open source approach promotes transparency and access. We know that although artificial intelligence has made huge progress in society, it also has risks. We are committed to building responsibly and offer a number of resources to help those using Llama 2 do so as well.

• Red band exercises: Our accurate models are standardized – tested for safety – through internal and external efforts. The team worked to generate hostile claims to make it easier to fine-tune the model. In addition, we outsourced external adversarial testing across our micromodels to similarly identify gaps in performance. These safety fine-tunings are frequent; We will continue to invest in safety through fine-tuning and benchmarking and plan to release thoroughly updated models based on these efforts.
• Transparency chart: We explain our modification and evaluation methods for the model and identify its shortcomings. Our transparency scheme, which is located within Research paperreveals known challenges and issues we encountered and provides insight into the mitigations that have been taken and future ones we intend to explore.
• Responsible use guide: Six past five This guide As a resource to support developers with best practices for responsible development and safety assessments. It identifies best practices that reflect the latest, current research on responsible generative AI that is being discussed across the industry and the AI ​​research community.
• Acceptable Use Policy: We put a Policy Certain use cases are prohibited to help ensure that these forms are used fairly and responsibly.

Meta has also created new initiatives to harness the insight and creativity of individuals, researchers, and developers around the world to get feedback on how models are performing and how they can be improved.

• Open Innovation AI Research Community: Today, we also launched a new partnership program for academic researchers It aims to deepen our understanding of the responsible development and sharing of large language paradigms. Researchers may apply to join a community of practitioners to share what they have learned about this important topic, and the community will shape a research agenda to pursue moving forward.
• Llama effect challenge: We want to revitalize the community of innovators who aspire to use llamas to solve difficult problems. We are launching a challenge to encourage a variety of public, nonprofit, and for-profit entities to use Llama 2 to address important environmental, educational, and other challenges. the Challenge rules It will be available before it starts.

Conclusion

Throughout our company’s history, we’ve experienced the benefits of an open source approach when innovating in other areas of the business. Our engineers have developed and shared frameworks that are now industry standards – like React, which is the leading framework for building web and mobile apps, and PyTorch, which is now the leading framework for AI. This infrastructure has become commonly used by the entire technology industry. We believe that sharing today’s large language models publicly will support the development of useful and safer AI as well.

We look forward to seeing what the world builds with the Llama 2.