Automates fine-tuning configs and streams Llama-3.1-8B on 4GB GPUs. Built for ML engineers who need memory-efficient training.
