Thetune

Thetune Tuner
Select Dataset
Training on Progress
Chatbot interface to preview model
Deploy to ThetaEdgeCloud
Deployed Model, API Ready (OpenAI Specs)
Thetune's Architecture

Inspiration

In today's rapidly evolving tech landscape, large language models (LLMs) hold immense potential to revolutionize various industries. However, fine-tuning and deploying these models often require specialized technical skills, leaving many non-tech users at a disadvantage. The inspiration for Thetune emerged from a desire to democratize access to LLM technology, enabling anyone, regardless of their technical background, to harness the power of AI. By leveraging the capabilities of Theta EdgeCloud, Thetune aims to simplify the process and make LLM fine-tuning and deployment accessible to all.

What it does

Thetune is a user-friendly application designed to help users fine-tune and deploy LLMs effortlessly. With an intuitive interface, users can:

Training using various datasets or using your own dataset (Huggingface datasets public or private)
Configure training parameters through simple options
Evaluate and Predict the trained model
Get Preview by chat trained model
Deploy the fine-tuned model and API on Theta EdgeCloud with a single click

Thetune eliminates the need for complex coding or deep technical knowledge, providing a seamless experience from start to finish.

How we built it

The development of Thetune involved several key steps:

User Interface Design: Built using Gradio, focused on creating an intuitive and accessible UI, clear visual cues. Landing page using Framer.
Backend Development: FastAPI was used to create a robust and efficient backend that handles user requests, data processing, and interactions with the training pipeline.
Containerization: Docker was employed to ensure the application is easily deployable and scalable, facilitating smooth integration with Theta EdgeCloud.
Model Repository: Huggingface Hub was used to store fine-tuned model
Deployment Automation: Automated the deployment process by Theta EdgeCloud and Huggingface API, enabling users to deploy their models on Theta EdgeCloud with minimal effort.

Here is the Thetune's Architecture:

Challenges we ran into

Limited time, and resources to deploy. API integration documentation for Theta EdgeCloud does not exist yet, so to make it work need an alternative way by using x-auth-id and x-auth-token from the request header in Theta EdgeCloud's Dashboard.