6aad1de658a933d3fa376f7fe9abf419da8a8bd2,python/baseline/tf/lm/train.py,LanguageModelTrainerTf,test,#LanguageModelTrainerTf#Any#Any#Any#,79

Before Change

            if xfer_state:
                state = vals["final_state"]
            total_loss += loss
            iters += ts.nctx
            step += 1

        self.log.debug({"phase": phase, "time": time.time() - start})
        metrics["avg_loss"] = total_loss / iters
        metrics["perplexity"] = np.exp(total_loss / iters)

        for reporting in reporting_fns:
            reporting(metrics, epochs, phase)
        return metrics


@register_training_func("lm")

After Change

            vals = self.model.sess.run(fetches, feed_dict)
            loss = vals["loss"]
            toks = self._num_toks(batch_dict)
            if xfer_state:
                state = vals["final_state"]
            total_loss += loss * toks
            total_toks += toks

        metrics = self.calc_metrics(total_loss, total_toks)
        self.report(
            epochs, metrics, start,
            phase, "EPOCH", reporting_fns
        )
        return metrics


@register_training_func("lm")

Instances