Announcement: Checkout out our new BOLT2.5B LLM – World’s first Generative LLM trained exclusively on CPUs

Introducing the ThirdAI Tech Blog

Showcasing the science and engineering behind unlocking the power of deep learning on everyday CPU devices
Hello World! This is the first post from ThirdAI’s new science and engineering blog, where we will showcase the exciting product innovations and developments from our talented team.
Founded in 2021, ThirdAI is an early-stage startup dedicated to the mission of democratizing artificial intelligence technologies through algorithmic and software innovations that make these techniques dramatically more sustainable and scalable. Our goal, quite simply, is to radically reduce the cost of state-of-the-art artificial intelligence while eliminating barriers to accessing this technology. We are driven by the conviction that one should not need to incur a massive carbon footprint or exorbitant GPU bill to take advantage of AI and begin integrating these capabilities into practical applications.

To achieve these goals, we have pioneered new approaches to training neural networks that enable machine learning researchers and practitioners to build billion-parameter models on cost-effective CPU hardware as opposed to expensive, specialized chips such as GPUs. Furthermore, we have developed a suite of novel, automated hyper-parameter tuning capabilities to provide a truly hands-off-the-wheel library, called Universal Deep Transformers (UDT). This AutoML interface allows anyone with a standard laptop to get started with training state-of-the-art deep learning models without the need for tedious manual tuning or sophisticated machine learning expertise.

And, as the cherry on top, our novel sparse inference techniques allow users to deploy our models with sub-millisecond latencies on CPUs — without the need for expensive model compression techniques such as knowledge distillation or pruning that are now commonplace in large-scale deep learning pipelines. With ThirdAI, one can train a model and immediately deploy it in production.
ThirdAI delivers revolutionary efficiency improvements in neural network training through proprietary sparse algorithms that activate only a small subset of neurons for a given input.
These advances are the culmination of years of academic research in applying hashing, sketching, and randomized algorithmic techniques to the domain of machine learning. In fact, much of this work was driven by current employees at ThirdAI, where we have assembled a collection of experts uniquely positioned to take on this once-in-a-lifetime challenge of democratizing AI.

For more updates on our product releases, research developments, and engineering advances, look for regular updates in this space. In the meantime, we invite you to try out our software through demo notebooks hosted on Google Colab and visit our website. To use ThirdAI , please reach out to us by requesting a trial license for our software.

ThirdAI is on a mission to make sophisticated large language models (LLMs) and other cutting-edge AI technologies accessible for everyone. Our goal is to build customized, private AI that is trained on commodity hardware with ultra-low latency inference for every organization.

USEFUL LINKS

Technology

Solutions

Newsroom
Company

Contact

  • Address
    3200 Southwest Fwy, Suite 3120, Houston, Texas 77027
  • Email contact@thirdai.com

Follow US

Twitter Instagram Linkedin Youtube