From 281d46dad22763ba74446ae0e33f8cccf736d794 Mon Sep 17 00:00:00 2001 From: Hui Zhang Date: Fri, 16 Apr 2021 03:34:33 +0000 Subject: [PATCH] fix logger --- deepspeech/__init__.py | 2 -- deepspeech/exps/u2/model.py | 10 ++++++---- deepspeech/io/dataset.py | 8 ++++---- deepspeech/models/u2.py | 6 +++--- 4 files changed, 13 insertions(+), 13 deletions(-) diff --git a/deepspeech/__init__.py b/deepspeech/__init__.py index 39523d897..0257dbe55 100644 --- a/deepspeech/__init__.py +++ b/deepspeech/__init__.py @@ -410,13 +410,11 @@ def ctc_loss(logits, input_lengths, label_lengths) loss_out = paddle.fluid.layers.squeeze(loss_out, [-1]) - logger.debug(f"warpctc loss: {loss_out}/{loss_out.shape} ") assert reduction in ['mean', 'sum', 'none'] if reduction == 'mean': loss_out = paddle.mean(loss_out / label_lengths) elif reduction == 'sum': loss_out = paddle.sum(loss_out) - logger.debug(f"ctc loss: {loss_out}") return loss_out diff --git a/deepspeech/exps/u2/model.py b/deepspeech/exps/u2/model.py index b445a501b..4ae3a603e 100644 --- a/deepspeech/exps/u2/model.py +++ b/deepspeech/exps/u2/model.py @@ -89,8 +89,9 @@ class U2Trainer(Trainer): if (batch_index + 1) % train_conf.accum_grad == 0: if dist.get_rank() == 0 and self.visualizer: - losses_np.update({"lr": self.lr_scheduler()}) - self.visualizer.add_scalars("step", losses_np, self.iteration) + losses_np_v = losses_np.copy() + losses_np_v.update({"lr": self.lr_scheduler()}) + self.visualizer.add_scalars("step", losses_np_v, self.iteration) self.optimizer.step() self.optimizer.clear_grad() self.lr_scheduler.step() @@ -171,8 +172,9 @@ class U2Trainer(Trainer): logger.info(msg) if self.visualizer: - valid_losses.update({"lr": self.lr_scheduler()}) - self.visualizer.add_scalars('epoch', valid_losses, self.epoch) + valid_losses_v = valid_losses.copy() + valid_losses_v.update({"lr": self.lr_scheduler()}) + self.visualizer.add_scalars('epoch', valid_losses_v, self.epoch) return valid_losses def setup_dataloader(self): diff --git a/deepspeech/io/dataset.py b/deepspeech/io/dataset.py index e1ee7a01b..4f58d41c6 100644 --- a/deepspeech/io/dataset.py +++ b/deepspeech/io/dataset.py @@ -297,13 +297,13 @@ class ManifestDataset(Dataset): else: speech_segment = SpeechSegment.from_file(audio_file, transcript) load_wav_time = time.time() - start_time - logger.debug(f"load wav time: {load_wav_time}") + #logger.debug(f"load wav time: {load_wav_time}") # audio augment start_time = time.time() self._augmentation_pipeline.transform_audio(speech_segment) audio_aug_time = time.time() - start_time - logger.debug(f"audio augmentation time: {audio_aug_time}") + #logger.debug(f"audio augmentation time: {audio_aug_time}") start_time = time.time() specgram, transcript_part = self._speech_featurizer.featurize( @@ -311,13 +311,13 @@ class ManifestDataset(Dataset): if self._normalizer: specgram = self._normalizer.apply(specgram) feature_time = time.time() - start_time - logger.debug(f"audio & test feature time: {feature_time}") + #logger.debug(f"audio & test feature time: {feature_time}") # specgram augment start_time = time.time() specgram = self._augmentation_pipeline.transform_feature(specgram) feature_aug_time = time.time() - start_time - logger.debug(f"audio feature augmentation time: {feature_aug_time}") + #logger.debug(f"audio feature augmentation time: {feature_aug_time}") return specgram, transcript_part def _instance_reader_creator(self, manifest): diff --git a/deepspeech/models/u2.py b/deepspeech/models/u2.py index 16e5eaad6..f34aac771 100644 --- a/deepspeech/models/u2.py +++ b/deepspeech/models/u2.py @@ -159,7 +159,7 @@ class U2BaseModel(nn.Module): start = time.time() encoder_out, encoder_mask = self.encoder(speech, speech_lengths) encoder_time = time.time() - start - logger.debug(f"encoder time: {encoder_time}") + #logger.debug(f"encoder time: {encoder_time}") #TODO(Hui Zhang): sum not support bool type #encoder_out_lens = encoder_mask.squeeze(1).sum(1) #[B, 1, T] -> [B] encoder_out_lens = encoder_mask.squeeze(1).cast(paddle.int64).sum( @@ -172,7 +172,7 @@ class U2BaseModel(nn.Module): loss_att, acc_att = self._calc_att_loss(encoder_out, encoder_mask, text, text_lengths) decoder_time = time.time() - start - logger.debug(f"decoder time: {decoder_time}") + #logger.debug(f"decoder time: {decoder_time}") # 2b. CTC branch loss_ctc = None @@ -181,7 +181,7 @@ class U2BaseModel(nn.Module): loss_ctc = self.ctc(encoder_out, encoder_out_lens, text, text_lengths) ctc_time = time.time() - start - logger.debug(f"ctc time: {ctc_time}") + #logger.debug(f"ctc time: {ctc_time}") if loss_ctc is None: loss = loss_att