At Rhymes AI, we are excited to announce our collaboration with AMD as we prepare to showcase our latest AI advancements at AMD's Advancing AI 2024.

This partnership brings together our newly announced multimodal AI technology with next-generation AMD EPYC™ processors and AMD Instinct™ accelerators, creating a powerful synergy that drives AI performance to new heights. With the AMD Instinct MI300 series accelerators driving our latest release, the consumer AI search app BeaGo, we are unlocking the transformative potential of this collaboration.

Powering the Future of AI with AMD

AMD’s cutting-edge hardware serves as the foundation for Rhymes AI to optimize our models, enabling unmatched performance and scalability. With AMD's advanced accelerators, we push the boundaries of AI, particularly when it comes to handling complex, multi-modality workloads.

Ramine Roane, Corporate Vice President, Artificial Intelligence Group at AMD said:

At AMD, we're dedicated to fueling the innovation of AI leaders like Rhymes AI by providing the advanced infrastructure they need to break new ground. With our EPYC processors and Instinct GPUs, we're driving next-gen AI solutions that set new standards in speed, scalability, and multimodal capabilities.

With AMD's advanced infrastructure, we've shipped AI models that are not only faster but also more efficient, underscoring how AMD's growing AI solutions ecosystem is becoming an essential part of the AI innovation ecosystem.

Introducing BeaGo: Ultra-Fast AI Search Powered by AMD

As part of Advancing AI 2024, we are proud to debut our AI search app BeaGo, as Rhymes AI's first consumer app. BeaGo showcases the high performance of AMD MI300X, delivering lightning-fast, comprehensive AI search results for both text and images using a model specifically optimized for speed and efficiency through AMD's hardware.

This launch highlights the powerful synergy between AMD and Rhymes AI, fundamentally reshaping how we think about search. BeaGo skips traditional search results displaying links, providing answers directly from question to response. With BeaGo, at Rhymes AI, we are delivering on our mission of powerful AI, built for all, making advanced AI search accessible and highly efficient.

Through our collaboration with AMD, we achieved significant inference throughput by leveraging the optimized ROCm software stack, along with the high computational power and 192GB memory capacity with the fast bandwidth of the MI300X. Working closely with the AMD team, we applied full-stack optimization techniques, including improvements to inference engine efficiency, kernel optimizations (such as Flash Attention and fine-tuned GEMM operations), and enhanced RCCL All-Reduce functionality.

In addition, we fine-tuned inference hyperparameters to maximize the advanced features of the MI300X, enabling larger throughput. These combined efforts resulted in Rhymes AI achieving a 1.5x increase in inference throughput on the AMD Instinct MI300X compared to the Nvidia H100 in offline mode. This demonstrates state-of-the-art performance, delivering an exceptional user experience in the large language model (LLM) applications we've developed.

Looking Ahead: Joint Innovation and Developer Community Engagement

As we launch BeaGo at Advancing AI 2024, we are excited to further engage with AMD's hardware, software and developer ecosystem to showcase how our collaboration will drive AI forward.

Together, Rhymes AI and AMD are committed to enabling businesses and developers to create disruptive AI solutions that will continue to push the frontier of the future of AI.

Rhymes AI x AMD: A Shared Vision for the Future of AI

This collaboration reflects our shared vision with AMD: to lead the future of AI with cutting-edge technology, performance, and scalability. As we continue to innovate and launch AI applications like BeaGo, we are proud to work alongside AMD to create a robust and innovative developer community that can leverage the best of both platforms. In addition to BeaGo, we've launched Aria, the first open source natively multimodal Mixture-of-Experts (MoE) model.

Stay tuned for more updates from Rhymes AI and AMD as we continue to deliver breakthrough AI innovations and further our collaboration.