You are browsing as a guest. Sign up (or log in) to start making projects!

@batow133 on trainbox

16m 37s logged

Devlog #1: Trainbox

Hey! I am Nikolas, and I am really obsessed with local AI and LLM training. But I discovered a problem… If someone wants to fine-tune a model, they will have to buy pricey hardware or a very expensive cloud subscription. So, that leads to LLM fine-tuning (and training) being very hard for beginners to start.

Solution: My project (it’s called Trainbox) is a local large language model fine-tuning center that helps people train LLMs with local hardware easily. This project will consist of different parts, like a fine-tuning center, a dataset creator (for context, datasets are the organized data LLMs are trained on), a fully local LLM inference engine (so you can chat with your LLM without changing apps), and many more!

Thanks for reading my first Devlog! I hope I really stick with the project and don’t abandon it… I am eager to hear your suggestions or questions about my project!

0
130

Comments 4

@raviiiibro

so where you get gpu to train ai model if it was a local project and we can train a ai model on kaggle why we need this and there are many other free tools to create a clean dataset

@batow133

So I wanted to bring everything together in one single app. Also I want to focus on making fine tuning accessible for devices with no gpu and little computation power.

@Kiviee

Hii! i am too very passionate about Local LLMs and was interested in this project.
I have a question though, How is this different from Unsloth studio which promises to enable users to fine tune easily on their own hardware and lets them create datasets and run models inside one unified interface?
Thanks!

@batow133

Hi! Thanks for the question.

Unsloth Studio is a great project and there is definitely some overlap. My goal with Trainbox is to build a more complete local AI workspace that covers the whole workflow, including dataset creation, fine tuning, and running models locally in one place.

I’m also focusing heavily on simplicity for beginners, so users can go from raw data to a usable model without needing to jump between multiple tools. The project is still early, but that’s the direction I’m aiming for. Thanks for checking it out!