Please sign in to access this page
(note: Setup instructions and readme.md are written with ai as english is my second language and its a bit hard for me to write long paragraphs with good grammar)
An AI model maker tool.
It is a fully featured synthetic data generator that allows you to make QLora-trainable data for LLMs via other AI models. It has many adjustable parameters, such as the models to use, topics to generate about, amount of data, etc etc.
This is useful as it allows you to make hyper-specialized AI models. As an example - I am using this to make a Python - and specifically Python - coder model, that is specialized in the use of the AI coding tool Aider.
It supports paid models via a variety of providers e.g OpenAI, Claude, and free models via other providers like Chutes, OpenRouter, Groq, Targon, Google AI Studio, etc.
It supports retrieving logits as well, which allows for distillation (which is when a small AI model learns from a larger AI model, improving its performance.)
Also has qlora.py and distill.py scripts which are a 1 click method to train or distill a model on your data.
No followers yet
Once you ship this you can't edit the description of the project, but you'll be able to add more devlogs and re-ship it as you add new features!
Fixed some bugs with qlora.py and distill.py and made them match Unsloth examples
Made the actual main.py script with multithreading and fixed qlora.py based off the unsloth notebook for Qwen 2.5 coder