Fine-Tune and Optimise Open LLMs with LoRA and QLoRA

For AI and ML engineers, data scientists and MLOps engineers who write Python. Decide when fine-tuning beats prompting and retrieval, then do it on open weights you control: build a clean, rights-checked training set, run supervised fine-tuning with Hugging Face TRL, train LoRA and QLoRA adapters with PEFT on a free GPU, and prove the tuned model beats its baselines without forgetting. Then align it with DPO and checkable rewards, quantize and serve it with llama.cpp behind a local API, monitor it, and gate every release in CI. Learn what changed in 2026, including the close of OpenAI self-serve fine-tuning, and finish with a free lab where you tune, score and serve a clause summariser yourself.