Digital Event Horizon
Together AI has partnered with NVIDIA to bring forth an unparalleled level of performance and efficiency to AI-powered applications, revolutionizing machine learning with the Together AI Turbocharged with NVIDIA Blackwell platform. This groundbreaking development promises to accelerate machine learning workloads like never before, with remarkable performance gains achieved by pioneering AI companies such as Zoom, Salesforce, and InVideo.
Together AI has partnered with NVIDIA to accelerate machine learning workloads. The RedPajama-V2 Dataset is designed for creating high-quality datasets and requires data filtering using quality signals. Together AI provides optimized data filtering tools and leverages NVIDIA Blackwell GPUs for performance gains. Pioneering companies such as Zoom, Salesforce, and InVideo have seen significant performance improvements with Together AI's platform. The introduction of the Together Training Stack features a co-optimized Llama 3 70B golden model achieving state-of-the-art MFU. The partnership with NVIDIA enhances reliability, security, and GPU price-performance for customers' infrastructure.
In a groundbreaking development that promises to revolutionize the field of machine learning, Together AI has partnered with NVIDIA to bring forth an unparalleled level of performance and efficiency to AI-powered applications. The introduction of the Together AI Turbocharged with NVIDIA Blackwell platform marks a significant milestone in the quest for accelerating machine learning workloads, and we take a closer look at what makes this partnership so remarkable.
According to context data, the RedPajama-V2 Dataset is conceptualized as a pool of data designed to serve as a foundation for creating high-quality datasets. However, it is not intended to be used out of the box, and data should be filtered using quality signals that accompany the dataset. With this in mind, Together AI's goal is to provide all the signals and tooling necessary to optimize data filtering.
As stated in the context data, Together AI specializes in delivering higher tokens/sec/node and overall Model FLOPS Utilization (MFU) than other providers on the same hardware. This is achieved through various optimizations that leverage NVIDIA Blackwell GPUs, including FlashAttention, computation/graph optimization, parallelism optimization, and more.
These innovations have been extensively tested with pioneering AI companies such as Zoom, Salesforce, and InVideo, resulting in remarkable performance gains. For instance, Salesforce witnessed a 2x improvement in training speeds when upgrading from NVIDIA HGX H200 to HGX B200, while Zoom saw a 1.9X improvement in training speeds over previous generation NVIDIA Hopper GPUs.
InVideo, meanwhile, initially encountered some challenges with the new architecture but was eventually able to improve its performance by more than doubling its original gains through further optimizations made in partnership with Together AI researchers.
The introduction of the Together Training Stack has also played a significant role in this revolution. This custom-built training container features a co-optimized Llama 3 70B golden model, achieving state-of-the-art (SOTA) Model FLOPS Utilization (MFU).
Furthermore, the partnership with NVIDIA has allowed Together AI to bring forth an enhanced level of reliability and security to its customers' infrastructure. This is demonstrated through strong GPU price-performance, reliable infrastructure, secure storage options, and deep technical expertise.
In conclusion, the Together AI Turbocharged with NVIDIA Blackwell platform represents a significant breakthrough in machine learning performance and efficiency. As we continue to push the boundaries of what is possible with AI, it will be exciting to see how this innovative partnership unfolds in the future.
Related Information:
https://www.digitaleventhorizon.com/articles/Turbocharging-AI-Performance-Together-AI-Revolutionizes-Machine-Learning-with-NVIDIA-Blackwell-deh.shtml
https://www.together.ai/blog/nvidia-blackwell-test-drive
Published: Thu Apr 24 13:14:34 2025 by llama3.2 3B Q4_K_M