What Is Together Compute?
Together Compute is an advanced AI tool that offers a high-performance cluster designed specifically for training models, akin to the renowned RedPajama model. This tool aims to provide users with the same cutting-edge capabilities through their state-of-the-art GPU clusters and proprietary training stack, while also delivering the best price-performance ratio.
Together Compute boasts premium hardware, utilizing private, state-of-the-art clusters equipped with H100 and A100 GPUs. These clusters are connected via fast 200 Gbps non-blocking Ethernet or up to 3.2 Tbps InfiniBand networks, ensuring efficient data transfer. Moreover, users have the option to choose environmentally friendly, carbon-reducing data centers for green compute alternatives.
The Together Compute’s Training Stack is a key component of the tool, offering users a suite of fast, reliable, and powerful training modules. Proprietary network compression, the Sophia optimizer, FlashAttention, and other cutting-edge modules are integrated to enhance training speed and reduce costs. Additionally, users can conveniently utilize popular frameworks like Slurm, DeepSpeed, Megatron, and others for training out of the box.
One of the standout features of Together Compute is its top-tier support. Users can collaborate with Together AI researchers and engineers, who act as an extension of their team. A playbook is provided to help users achieve the best training recipe and address issues such as sub-optimal utilization and convergence problems. Custom professional services, tailored training, and a dedicated communication channel on Discord or Slack are also available to enhance the overall experience.
Together Compute offers a powerful AI tool with high-performance clusters, a sophisticated training stack, and outstanding support, enabling users to optimize their AI model training process effectively.
