DEKUBE

DEKUBE offers a cost-effective, scalable and efficient way to get the computing resources needed to train and fine-tune LLM. The company's goal is to democratize AI innovation by creating the world's largest distributed AI computing network.

DEKUBE lets you run jobs in parallel on multiple nodes, speeding up the time it takes to get results. You can scale up or down depending on your training needs, with unlimited capacity, high availability and lower costs. The service uses asymmetric encryption to protect data, has a security monitoring system and certification for computational nodes.

Among the features:

• Infinite Scalability: Can train super-large parameter models with the computing power of millions of GPUs.
• Cost Reduction: Can cut the cost of a full-parameter training job by up to 40% compared with a centralized system.
• Speed Improvement: Can speed up training jobs by up to 30% with the Load Balance System.
• Elastic Supply: Can dynamically allocate resources and supply them on a pay-as-you-go basis.
• GPU Access: Can handle more than 10,000 GPUs and 1,000 nodes for distributed computing.
• Fair POW: Lets GPU device owners get rewards based on how much they contribute.

The distributed network uses blockchain technology and Kubernetes to manage resources and jobs efficiently. DEKUBE also offers a client for GPU providers to easily set up and manage their own computing resources.

Pricing isn't clear, but DEKUBE offers a pay-as-you-go pricing model so you only pay for what you use. GPU providers can get rewards based on their contributions, with different rewards for different levels of participation.

DEKUBE is useful for anyone who needs a cost-effective and scalable way to train and fine-tune LLM, especially those who need a lot of computing horsepower. By using a distributed network, you can get results faster without worrying about data privacy and security.

Published on July 11, 2024