Supervised Learning
Training a model on input-output pairs where the correct output is provided as a label.
Full Definition
In supervised learning, the training data consists of (input, label) pairs and the model learns a function mapping inputs to outputs by minimising the difference between its predictions and the ground-truth labels. For language models, supervised fine-tuning uses (prompt, response) pairs where a human or expert has written the ideal response. This contrasts with the unsupervised pretraining phase where labels are derived from the data itself (predicting the next token). Supervised learning is highly data-efficient when high-quality labelled examples are available, but labelling at scale is expensive. Most alignment and instruction tuning pipelines begin with a supervised fine-tuning (SFT) phase.
Examples
Training a text classifier on 10,000 emails labelled as spam or not-spam using logistic regression.
The SFT phase of InstructGPT: fine-tuning GPT-3 on 13,000 human-written prompt-response pairs before RLHF.
Apply this in your prompts
PromptITIN automatically uses techniques like Supervised Learning to build better prompts for you.
Related Terms
Fine-Tuning
Continuing training of a pretrained model on a smaller, task-specific dataset to…
View →Instruction Tuning
Supervised fine-tuning on diverse instruction-response pairs to improve a model'…
View →Unsupervised Learning
Training a model on raw data without human-provided labels, discovering structur…
View →