minimal LLM scripts for 24GB VRAM GPUs. training, inference, whatever - View it on GitHub