http://programmer.ie/tags/miprov2/ Recent content in MIPROv2 on Programmer.ie Hugo en-us Fri, 28 Aug 2026 10:55:00 +0100 http://programmer.ie/books/dspy-from-first-principles/12-chapter/ Fri, 28 Aug 2026 10:55:00 +0100 http://programmer.ie/books/dspy-from-first-principles/12-chapter/ <h1 id="optimize-the-instructions">Optimize the Instructions</h1> <p>Chapter 11 let the optimizer change predictor behavior through bootstrapped demonstrations. That changes the examples shown to the LM, but the predictor instructions remain mostly fixed.</p> <p>Now we expand the allowed mutation surface:</p> <hr> <h2 id="1-what-miprov2-searches">1. What MIPROv2 searches</h2> <p>Current MIPROv2 builds candidate few-shot sets, proposes instruction candidates for predictors, and searches combinations against validation evidence. Its current implementation uses an Optuna-backed search over prompt parameters.</p> <p>Conceptually:</p> <hr> <h2 id="2-a-miprov2-compile-shape">2. A MIPROv2 compile shape</h2> <p>Current docs show <code>MIPROv2(metric=..., auto=&quot;light&quot; | &quot;medium&quot; | &quot;heavy&quot;)</code>. Its <code>compile</code> method accepts <code>trainset</code> and <code>valset</code>; internally, the optimizer validates those datasets, builds an evaluator over the validation set, and searches prompt parameters against that development evidence.</p>