Say hello to Pareto, the world’s first GenAI log-math number system to make GenAI inference compute orders of magnitude faster and cheaper.
🔴 AI calculations need a lot of power
Modern AI transformer models like Llama, Falcon, or Mistral involve trillions of multiplications every second - and multiplications are exponentially growing as AI becomes more capable. But adding numbers uses a lot fewer silicon transistors and consumes far less power than multiplying them.
🔵 Pareto makes AI calculations more efficient
By turning multiplications and divisions into additions and subtractions, Pareto makes GenAI inference compute smaller, faster, and more energy-efficient, all while keeping accuracy extremely high.
Extensive testing on various AI models, including Mixtral-8x22B, Llama3-70B, Falcon-180B, Stable Diffusion XL, and Llama3.1-405B shows that Pareto achieves a relative accuracy of over 99.9% compared to the trained high-precision baseline model and consumes significantly less power.
🟢 Pareto makes GenAI inference more profitable
>> Pareto helps data centers and enterprises to maximize utilization of compute, space, and energy - drastically lowering total cost of ownership.
>> Pareto helps developers to bring new models to production in less time and with higher accuracy - generating revenue more quickly and with better profit margins.
In February of this year, we raised $103M Series C funding to double down on our strategic focus and our mission to accelerate the world’s AI ambitions. The round was co-led by Celesta Capital and GreatPoint Ventures. With special thanks to Ashok Krishnamurthi, Sriram Viswanathan, and Kevin Johnson.
If you want to know more about Pareto’s nitty gritty details, join us at NeurIPS 2024.