Large Language Models do not become specialized, aligned, or reasoning-focused through a single training technique. When working with models such as Qwen, Llama, Mistral, or other open-weight LLMs, we encounter terms such as: SFT, DPO, PPO, GRPO, PEFT, LoRA, and QLoRA. At first, these can look like competing fine-tuning methods. They are not. Some define […]
