def run_step()

in cm/train_util.py [0:0]

• 13 lines of code
• 4 McCabe index (conditional complexity)

    def run_step(self, batch, cond):
        self.forward_backward(batch, cond)
        took_step = self.mp_trainer.optimize(self.opt)
        if took_step:
            self._update_ema()
            if self.target_model:
                self._update_target_ema()
            if self.training_mode == "progdist":
                self.reset_training_for_progdist()
            self.step += 1
            self.global_step += 1

        self._anneal_lr()
        self.log_step()