Chinese startup 01.ai has stunned the AI community by training a state-of-the-art LLM with just 2,000 GPUs and $3 million.
This stands in contrast to OpenAIâs estimated $80-100 million spent on GPT-4, showcasing the power of innovation over brute force.
Kai-Fu Lee, the founder of 01.ai, stated, âThe thing that shocks my friends in Silicon Valley is not just our performance, but that we trained the model with only $3 million.â
By turning computational problems into memory-focused tasks, they have built a multi-layer caching system and designed a specialized inference engine. This approach allowed them to optimize GPU usage and drastically lower inference costs.
Their model Yi-Lightning ranks 6th in performance in LMSYS 'Overall' benchmark.
It's clear that the future of AI lies in smart engineering rather than just scaling & spending power.
Checkout their public models at https://lnkd.in/g6NPS4FN
Chatbot Arena: https://lmarena.ai
â
Are you a developer? Check out https://AlphaSignal.ai for a daily summary of breakthrough models, repos, and papers in AI. Read by over 200,000 developers.