Behind the Compute: Using the New AI Supercomputer
Behind the Compute is a series of blog posts that chronicle elements of our business, offering insights for others to harness the power of generative AI.
In our last installment, we spoke about why it made sense for us to explore alternatives to GPUs when building a new AI Supercomputer to power our next gen generative AI models.
In this installment, we’ll talk about how we plan to utilize this state-of-the-art AI Supercomputer.
This supercomputer ranks among the top 15 globally in terms of sheer power and efficiency. It’s run with an impressive 4,000 Intel Gaudi2 processors and resides in the state of Texas, USA - and as they say, "everything is bigger in Texas!"
We aim to harness the supercomputer's capabilities in four key ways:
Efficient Model Training and Optimization: The supercomputer will enable us to refine and condense our language models, a process known as quantization. This is akin to removing the “L” from “LLM,” making our models more efficient and compact.
Advanced Models Anywhere: Imagine running a 1.6B parameter model from your laptop. This becomes feasible as the supercomputer compresses these models for edge computing, bringing the computational power accessible for heavy tasks while managing and controlling the process from the convenience of a laptop. This capability is especially beneficial for managing large datasets, which previously could not be processed on local machines.
Multimodal Capabilities: The system is equipped to handle models in various modalities - image, video, audio, language and text - with ease. Beyond just refining and reducing the size of models, we continue to enhance our datasets to ensure our modalities remain cutting-edge.
Transparent Training: The need for transparent AI in heavily regulated industries is paramount. By building models across various modalities with auditable datasets on our Gaudi2 infrastructure, we aim to foster broader enterprise adoption by ensuring understanding and trust in how our models are developed.
Stay tuned for more insights in our next installment of "Behind the Compute".
To stay updated on our progress follow us on Twitter, Instagram, LinkedIn, and join our Discord Community.
