LLM Fine-Tuning Course – From Supervised FT to RLHF, LoRA, and Multimodal

A comprehensive LLM fine-tuning course covering the full training pipeline from unsupervised pre-training through supervised fine-tuning (SFT) to preference-based alignment. Topics include full vs. partial fine-tuning, parameter-efficient methods (LoRA, QLoRA, DoRA, adapter layers, IA3, BitFit, prefix tuning), instruction vs. non-instruction fine-tuning, RLHF, DPO, and multimodal fine-tuning. Practical implementations use the Hugging Face ecosystem, Unsloth, and LLaMA Factory. The course walks through real dataset examples for each training stage and demonstrates hands-on fine-tuning on domain-specific data using Google Colab.