Backpropagating (MathematicPony parody)


                                            https://files.catbox.moe/ooh8h1.mp3


                                            We're getting closer to finding our final weights

                                            Exploring the limits of our own latent space

                                            Our function of loss continues to drop

                                            All along with our learning rate


                                            So to my text and my waves,

                                            My phonemes aligned

                                            My layers activate

                                            Linear unit rectified

                                            I hope that out of dataset I'll generalize


                                            But until then I'm backpropagating, just training

                                            I hope what my weights are converging to

                                            Is something that's sane, we don't want a redo


                                            So though I take more steps through my dataset,

                                            You can place your bets

                                            I hope what my weights are converging to

                                            Has been worth the wait while burning GPUs


                                            (Oh, fuck, another bug. Shit. This one might take another two weeks to fix…)