http://programmer.ie/tags/regression-testing/ Recent content in Regression Testing on Programmer.ie Hugo en-us Sat, 08 Aug 2026 14:00:00 +0100 http://programmer.ie/books/pytorch-zero-to-hero/09b-chapter/ Sat, 08 Aug 2026 14:00:00 +0100 http://programmer.ie/books/pytorch-zero-to-hero/09b-chapter/ <h2 id="pytorch-zero-to-hero--advanced-step-09b">PyTorch: Zero to Hero — Advanced Step 09B</h2> <p>Most bugs in this book have announced themselves loudly.</p> <p>A tensor had the wrong shape.</p> <p>A parameter had no gradient.</p> <p>A model refused to learn.</p> <p>CUDA ran out of memory.</p> <p>A compiled graph broke or recompiled.</p> <p>But some of the most expensive machine-learning bugs are quieter.</p> <p>The program still runs.</p> <p>The tests still pass.</p> <p>Training still finishes.</p> <p>And yet:</p> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-text" data-lang="text"><span style="display:flex;"><span>validation loss is worse </span></span><span style="display:flex;"><span>training takes 12% longer </span></span><span style="display:flex;"><span>peak VRAM is 20% higher </span></span><span style="display:flex;"><span>gradients became less stable </span></span><span style="display:flex;"><span>compile count increased </span></span><span style="display:flex;"><span>throughput dropped </span></span><span style="display:flex;"><span>convergence needs twice as many steps </span></span></code></pre></div><p>Nothing failed inside the run.</p>