Unified fine-tuning framework for open LLMs. Supports QLoRA, full-parameter, multi-node training and all major model architectures.