MIPRO (Multiprompt Instruction Proposal Optimizer) is a prompt optimization method published in 2024 by researchers at Stanford University and other institutions. For programs that chain several language model calls, it jointly optimizes the instructions and the few-shot examples of each stage. It collects execution traces that pass the metric as candidate examples, generates candidate instructions grounded in summaries of the data and the program, and then uses Bayesian optimization to search for strong combinations.
MIPROv2 is the version implemented as an optimizer in the prompt programming framework DSPy.