Akash Playground

The rapid development of AI models has led to a surge in innovative applications, but the lack of discoverability and ease of use hinders the growth of users in the AI ecosystem. Hugging Face Spaces has democratized the sharing and demonstration of AI models, but it doesn't provide a direct way for model builders to benefit from their work. Our platform addresses this gap by creating a marketplace that rewards model builders for their contributions.

Key Features

1. Model Upload and Deployment: Developers can upload their AI models to the platform, which will be deployed on Akash Cloud's infrastructure. This is done through sharing a Github or HuggingFace repository that runs a Gradio application.
2. Ease of Use: Users who want access to the models can just launch a playground, in a single click, from all the public models that are indexed by the platform.
3. Infrastructure Abstraction: The platform will abstract away the complexities of infrastructure management, allowing users to focus on model development and deployment.
4. Monetization Mechanism: Model builders will receive a share of the infrastructure costs when their models are used by others, providing a direct incentive for contributing to the ecosystem.
5. Compatibility with Hugging Face Spaces: The platform will be compatible with Hugging Face Spaces repositories, enabling seamless integration and collaboration.
6. Community Engagement: The platform will feature a community forum for discussion, feedback, and collaboration among users.

Benefits

1. Democratized Access: The platform will make AI models more accessible to a broader audience, including those without extensive infrastructure expertise.
2. Incentivizing Innovation: The monetization mechanism will encourage model builders to contribute to the ecosystem, driving innovation and growth.
3. Cost-Effective: The platform will provide a cost-effective solution for users, as they only pay for the infrastructure resources they use.

Technical Implementation

1. User Interface: Akash Playground is built using SvelteKit and the focus was to make it dead simple to launch a pre-made model on Akash.
2. Deployment Pipeline: The deployment pipeline features a custom docker image that prepares an environment that can run a Huggingface space. It uses a pre-processing script that makes sure that the application runs smoothly on Akash Cloud.
3. Backend: To keep things simple, we chose to use Supabase as both the OAuth solution and the database that stores metadata regarding users and playgrounds. This can be replaced with any other solution easily.
4. Reverse Proxy: After running the application, instead of directly using the endpoint generated by the provider. The application connects to a reverse proxy that generates a link. (e.g. http://123.plygrnd.live) On top of making it easy to share, it also helps to hide the provider's identity. This is the link that will be loaded in the iframe when a playground is launched.
5. Stripe integration: Users pay directly using Stripe instead of having to own a wallet. The platform will deduct from the credits in the account when there is an active playground running. Payments will be settled when the deployment is closed or when the user runs out of credits. All deployments are paid in AKT. We will use the current price of AKT when the deployment is created. In the background, what we can do to hedge against the price fluctuation is to purchase AKT twice daily to replenish what is required.

Future Work

1. Integrate with a decentralized file storage for developers to upload datasets
2. Support other UI frameworks like streamlit
3. Make playground deployments faster and more reliable
4. Add more community features
5. Streaming payments to model developers