Unsloth AI makes fine-tuning easier
Was reading about Unsloth AI and how it can be used to fine-tune open-source models like qwen3, llama 4, gemma 3, phi-4, etc. faster with 70% less memory. It's open-source and supports LoRA and QLoRA fine-tuning methods.
I got to know about this from this Andrea Volpini post about a fine-tuned reasoning model that thinks like an SEO. They created SEOcrate_4B_grpo_new_01 by using Unsloth AI for fine-tuning the Gemma 3 4B model via Unsloth.
Unsloth AI has a free plan that supports fine-tuning Mistal, Gemma, and LLaMA models that you can run on your computer. And they also have multiple beginner-friendly Google Colab notebooks available for different models that you want to train.
I have yet to try this, but will be going through their documentation and try to fine-tune for some data.
Comment via email