Supervised fine-tuning (SFT) is a foundational technique where a pre-trained language model is further trained on a dataset of high-quality, human-curated input-output pairs. This process aligns the model's behavior with desired instructions and response formats, serving as a crucial step before advanced alignment methods like RLHF.
Supervised fine-tuning (SFT) trains a pre-trained AI model on specific examples to teach it how to follow instructions and generate desired responses. It's a fundamental step that makes large language models useful for tasks like answering questions or writing text, often preceding more advanced training methods.
Instruction Fine-Tuning, Instruction Tuning, Supervised Instruction Tuning
Was this definition helpful?