How to use Akash Network for AI model training

Overclock Labs, the power behind Akash Network, partnered with ThumperAI to launch a baseline model training effort using Akash Network GPUs.

In the year Underway by the end of 2023, this collaboration aims to push the boundaries of decentralized cloud computing by training a foundational model that represents a significant leap forward in the integration of AI and cloud computing.

Using the Akash network to train AI

The initiative sought to address the challenges associated with generative AI foundation model training. These include high costs, strict hardware requirements, and complex software requirements. Thus, decentralized computing platforms have set the benchmark for meeting and exceeding the needs of AI startups.

ThumperAI aims to strengthen the loyalty of Lora's trainer service. Facilitates AI developers in fine-tuning base models using LoRA techniques. Akash Network, on the other hand, wanted to demonstrate the feasibility and efficiency of AI model training on a decentralized cloud platform.

This initiative was not only a proof of concept, but also aimed at attracting a wider developer community, generating interest from GPU vendors, and strengthening the Aash Network's position in open source contributions and AI use case expansion.

Read More: How Artificial Intelligence (AI) Will Change Crypto?

The journey involved critical decisions and trade-offs, particularly in selecting the prototype category and base model for the training. Again, the chosen approach used a Creative Commons-licensed dataset for training. As a result of the initial training tests, Pixart-Alpha inspired architecture with an emphasis on copyright compliance and data diversity.

Despite problems with the quality and diversity of the training database, the results of this project were promising. The collaboration has successfully demonstrated the feasibility of training base models on a decentralized network such as Akash. So it has paid a chapter on the intersection of AI and blockchain technology.

“The Thumper team is looking at migrating Laura Coach to run on the Akash network to take advantage of the lower costs and different GPUs available on the network when they want to expand the service. One of the features they need before doing this is the ability to request the maximum amount of SHM via the Akash SDL. That feature will be available on the network in an update.” It is,” writes Anil Murthy, VP of Product and Engineering at Oversea Labs.

This initiative sets the stage for future developments, as both groups want to explore more options on the Akash network.