AiVibeFine-Tuning
Fine-tuning continues training a pre-trained model on a smaller, task-specific dataset so that it adapts to a particular domain, style or task, such as classifying defect images or following an organisation's response format.
Full fine-tuning updates all model weights, which for large models requires substantial GPU memory. Parameter-efficient fine-tuning (PEFT) methods such as low-rank adaptation (LoRA) train small additional matrices while freezing the original weights, and QLoRA combines this with a quantised base model to cut memory further. For LLMs, supervised fine-tuning on instruction-response pairs is often followed by preference tuning such as RLHF or direct preference optimisation.
Fine-tuning suits tasks that need a consistent output format, domain-specific classification or specialised vocabulary, and it can produce smaller, cheaper models that perform well on a narrow task. In vision, fine-tuning pre-trained networks on plant images is the usual way to build inspection models.
For adding factual or frequently changing knowledge, retrieval-augmented generation is usually more practical, because knowledge can be updated without retraining. Fine-tuning needs clean, representative examples, held-out evaluation and checks for regressions in general capability and safety behaviour. The licence terms of the base model govern whether and how a fine-tuned model may be used.
Key points
- Continues training a pre-trained model on task-specific data
- LoRA and other PEFT methods train only small added parameters
- RAG is usually better for frequently changing knowledge
- Requires held-out evaluation and regression checks
Where AiVibe comes in
AiVibe delivers AI and machine learning services, chatbots and virtual assistants with RAG, MCP tools and voice, AI quality management including bias detection and model validation, and the AIMURUGA AI agent, and builds Intel-based edge AI devices using the Intel Distribution of OpenVINO toolkit.
Related terms
- Transfer LearningAI & Machine Learning
- Reinforcement Learning from Human Feedback (RLHF)AI & Machine Learning
- Large Language Model (LLM)AI & Machine Learning
- Retrieval-Augmented Generation (RAG)AI & Machine Learning
- Model QuantisationAI & Machine Learning
- Foundation ModelAI & Machine Learning