Use LossRecord to record and print loss for the training process

egs/librispeech/ASR/conformer_ctc/train.py

@@ -1,6 +1,7 @@
 #!/usr/bin/env python3
 # Copyright    2021  Xiaomi Corp.        (authors: Fangjun Kuang,
-#                                                  Wei Kang)
+#                                                  Wei Kang
+#                                                  Mingshuang Luo)
 #
 # See ../../../../LICENSE for clarification regarding multiple authors
 #
@@ -18,16 +19,21 @@
 
 
 import argparse
+import collections
+import copy
 import logging
 from pathlib import Path
 from shutil import copyfile
-from typing import Optional
+from typing import Optional, Tuple, List
+
 
 import k2
 import torch
 import torch.distributed as dist
 import torch.multiprocessing as mp
 import torch.nn as nn
+from torch import Tensor
+
 from asr_datamodule import LibriSpeechAsrDataModule
 from conformer import Conformer
 from lhotse.utils import fix_random_seed
@@ -281,13 +287,80 @@ def save_checkpoint(
         copyfile(src=filename, dst=best_valid_filename)
 
 
+class LossRecord(collections.defaultdict):
+    def __init__(self):
+        # Passing the type 'int' to the base-class constructor
+        # makes undefined items default to int() which is zero.
+        super(LossRecord, self).__init__(int)
+
+    def __add__(self, other: 'LossRecord') -> 'LossRecord':
+        ans = LossRecord()
+        for k, v in self.items():
+            ans[k] = v
+        for k, v in other.items():
+            ans[k] = ans[k] + v
+        return ans
+
+    def __mul__(self, alpha: float) -> 'LossRecord':
+        ans = LossRecord()
+        for k, v in self.items():
+            ans[k] = v * alpha
+        return ans
+
+
+    def __str__(self) -> str:
+        ans = ''
+        for k, v in self.norm_items():
+            norm_value = '%.4g' % v
+            ans += (str(k) + '=' + str(norm_value) + ', ')
+        frames = str(self['frames'])
+        ans += 'over ' + frames + ' frames.'
+        return ans
+
+    def norm_items(self) -> List[Tuple[str, float]]:
+        """
+        Returns a list of pairs, like:
+          [('ctc_loss', 0.1), ('att_loss', 0.07)]
+        """
+        num_frames = self['frames'] if 'frames' in self else 1
+        ans = []
+        for k, v in self.items():
+            if k != 'frames':
+                norm_value = float(v) / num_frames
+                ans.append((k, norm_value))
+        return ans
+
+    def reduce(self, device):
+        """
+        Reduce using torch.distributed, which I believe ensures that
+        all processes get the total.
+        """
+        keys = sorted(self.keys())
+        s = torch.tensor([ float(self[k]) for k in keys ],
+                         device=device)
+        dist.all_reduce(s, op=dist.ReduceOp.SUM)
+        for k, v in zip(keys, s.cpu().tolist()):
+            self[k] = v
+
+    def write_summary(self, tb_writer: SummaryWriter, prefix: str, batch_idx: int) -> None:
+        """
+        Add logging information to a TensorBoard writer.
+                tb_writer: a TensorBoard writer
+                  prefix: a prefix for the name of the loss, e.g. "train/valid_",
+                              or "train/current_"
+               batch_idx: The current batch index, used as the x-axis of the plot.
+        """
+        for k, v in self.norm_items():
+            tb_writer.add_scalar(prefix + k, v, batch_idx)
+
+
 def compute_loss(
     params: AttributeDict,
     model: nn.Module,
     batch: dict,
     graph_compiler: BpeCtcTrainingGraphCompiler,
     is_training: bool,
-):
+) -> Tuple[Tensor, LossRecord]:
     """
     Compute CTC loss given the model and its inputs.
 
@@ -367,15 +440,18 @@ def compute_loss(
         loss = ctc_loss
         att_loss = torch.tensor([0])
 
-    # train_frames and valid_frames are used for printing.
-    if is_training:
-        params.train_frames = supervision_segments[:, 2].sum().item()
-    else:
-        params.valid_frames = supervision_segments[:, 2].sum().item()
-
     assert loss.requires_grad == is_training
 
-    return loss, ctc_loss.detach(), att_loss.detach()
+    info = LossRecord()
+    # TODO: there are many GPU->CPU transfers here, maybe combine them into one.
+    info['frames'] = supervision_segments[:, 2].sum().item()
+    info['ctc_loss'] = ctc_loss.detach().cpu().item()
+    if params.att_rate != 0.0:
+        info['att_loss'] = att_loss.detach().cpu().item()
+
+    info['loss'] = loss.detach().cpu().item()
+
+    return loss, info
 
 
 def compute_validation_loss(
@@ -384,18 +460,14 @@ def compute_validation_loss(
     graph_compiler: BpeCtcTrainingGraphCompiler,
     valid_dl: torch.utils.data.DataLoader,
     world_size: int = 1,
-) -> None:
+) -> LossRecord:
-    """Run the validation process. The validation loss
+    """Run the validation process."""
-    is saved in `params.valid_loss`.
-    """
     model.eval()
 
-    tot_loss = 0.0
+    tot_loss = LossRecord()
-    tot_ctc_loss = 0.0
+    
-    tot_att_loss = 0.0
-    tot_frames = 0.0
     for batch_idx, batch in enumerate(valid_dl):
-        loss, ctc_loss, att_loss = compute_loss(
+        loss, loss_info = compute_loss(
             params=params,
             model=model,
             batch=batch,
@@ -403,36 +475,17 @@ def compute_validation_loss(
             is_training=False,
         )
         assert loss.requires_grad is False
-        assert ctc_loss.requires_grad is False
+        tot_loss = tot_loss + loss_info
-        assert att_loss.requires_grad is False
-
-        loss_cpu = loss.detach().cpu().item()
-        tot_loss += loss_cpu
-
-        tot_ctc_loss += ctc_loss.detach().cpu().item()
-        tot_att_loss += att_loss.detach().cpu().item()
-
-        tot_frames += params.valid_frames
 
     if world_size > 1:
-        s = torch.tensor(
+        tot_loss.reduce(loss.device)
-            [tot_loss, tot_ctc_loss, tot_att_loss, tot_frames],
-            device=loss.device,
-        )
-        dist.all_reduce(s, op=dist.ReduceOp.SUM)
-        s = s.cpu().tolist()
-        tot_loss = s[0]
-        tot_ctc_loss = s[1]
-        tot_att_loss = s[2]
-        tot_frames = s[3]
 
-    params.valid_loss = tot_loss / tot_frames
+    loss_value = tot_loss['loss'] / tot_loss['frames']
-    params.valid_ctc_loss = tot_ctc_loss / tot_frames
+    if loss_value < params.best_valid_loss:
-    params.valid_att_loss = tot_att_loss / tot_frames
-
-    if params.valid_loss < params.best_valid_loss:
         params.best_valid_epoch = params.cur_epoch
-        params.best_valid_loss = params.valid_loss
+        params.best_valid_loss = loss_value
+
+    return tot_loss
 
 
 def train_one_epoch(
@@ -471,24 +524,21 @@ def train_one_epoch(
     """
     model.train()
 
-    tot_loss = 0.0  # sum of losses over all batches
+    tot_loss = LossRecord()
-    tot_ctc_loss = 0.0
-    tot_att_loss = 0.0
 
-    tot_frames = 0.0  # sum of frames over all batches
-    params.tot_loss = 0.0
-    params.tot_frames = 0.0
     for batch_idx, batch in enumerate(train_dl):
         params.batch_idx_train += 1
         batch_size = len(batch["supervisions"]["text"])
 
-        loss, ctc_loss, att_loss = compute_loss(
+        loss, loss_info = compute_loss(
             params=params,
             model=model,
             batch=batch,
             graph_compiler=graph_compiler,
             is_training=True,
         )
+        # summary stats
+        tot_loss = (tot_loss * (1 - 1 / params.reset_interval)) + loss_info
 
         # NOTE: We use reduction==sum and loss is computed over utterances
         # in the batch and there is no normalization to it so far.
@@ -498,75 +548,21 @@ def train_one_epoch(
         clip_grad_norm_(model.parameters(), 5.0, 2.0)
         optimizer.step()
 
-        loss_cpu = loss.detach().cpu().item()
-        ctc_loss_cpu = ctc_loss.detach().cpu().item()
-        att_loss_cpu = att_loss.detach().cpu().item()
-
-        tot_frames += params.train_frames
-        tot_loss += loss_cpu
-        tot_ctc_loss += ctc_loss_cpu
-        tot_att_loss += att_loss_cpu
-
-        params.tot_frames += params.train_frames
-        params.tot_loss += loss_cpu
-
-        tot_avg_loss = tot_loss / tot_frames
-        tot_avg_ctc_loss = tot_ctc_loss / tot_frames
-        tot_avg_att_loss = tot_att_loss / tot_frames
-
         if batch_idx % params.log_interval == 0:
             logging.info(
-                f"Epoch {params.cur_epoch}, batch {batch_idx}, "
+                f"Epoch {params.cur_epoch}, batch {batch_idx}, loss[{loss_info}]"
-                f"batch avg ctc loss {ctc_loss_cpu/params.train_frames:.4f}, "
+                f"tot_loss[{tot_loss}], batch size: {batch_size}"
-                f"batch avg att loss {att_loss_cpu/params.train_frames:.4f}, "
-                f"batch avg loss {loss_cpu/params.train_frames:.4f}, "
-                f"total avg ctc loss: {tot_avg_ctc_loss:.4f}, "
-                f"total avg att loss: {tot_avg_att_loss:.4f}, "
-                f"total avg loss: {tot_avg_loss:.4f}, "
-                f"batch size: {batch_size}"
             )
 
+        if batch_idx % 10 == 0:
+
             if tb_writer is not None:
-                tb_writer.add_scalar(
+                loss_info.write_summary(tb_writer, "train/current_", params.batch_idx_train)
-                    "train/current_ctc_loss",
+                tot_loss.write_summary(tb_writer, "train/tot_", params.batch_idx_train)
-                    ctc_loss_cpu / params.train_frames,
-                    params.batch_idx_train,
-                )
-                tb_writer.add_scalar(
-                    "train/current_att_loss",
-                    att_loss_cpu / params.train_frames,
-                    params.batch_idx_train,
-                )
-                tb_writer.add_scalar(
-                    "train/current_loss",
-                    loss_cpu / params.train_frames,
-                    params.batch_idx_train,
-                )
-                tb_writer.add_scalar(
-                    "train/tot_avg_ctc_loss",
-                    tot_avg_ctc_loss,
-                    params.batch_idx_train,
-                )
-
-                tb_writer.add_scalar(
-                    "train/tot_avg_att_loss",
-                    tot_avg_att_loss,
-                    params.batch_idx_train,
-                )
-                tb_writer.add_scalar(
-                    "train/tot_avg_loss",
-                    tot_avg_loss,
-                    params.batch_idx_train,
-                )
-        if batch_idx > 0 and batch_idx % params.reset_interval == 0:
-            tot_loss = 0.0  # sum of losses over all batches
-            tot_ctc_loss = 0.0
-            tot_att_loss = 0.0
-
-            tot_frames = 0.0  # sum of frames over all batches
 
         if batch_idx > 0 and batch_idx % params.valid_interval == 0:
-            compute_validation_loss(
+            logging.info("Computing validation loss")
+            valid_info = compute_validation_loss(
                 params=params,
                 model=model,
                 graph_compiler=graph_compiler,
@@ -575,32 +571,13 @@ def train_one_epoch(
             )
             model.train()
             logging.info(
-                f"Epoch {params.cur_epoch}, "
+                f"Epoch {params.cur_epoch}, validation: {valid_info}"
-                f"valid ctc loss {params.valid_ctc_loss:.4f},"
-                f"valid att loss {params.valid_att_loss:.4f},"
-                f"valid loss {params.valid_loss:.4f},"
-                f" best valid loss: {params.best_valid_loss:.4f} "
-                f"best valid epoch: {params.best_valid_epoch}"
                 )
             if tb_writer is not None:
-                tb_writer.add_scalar(
+                valid_info.write_summary(tb_writer, "train/valid_", params.batch_idx_train)
-                    "train/valid_ctc_loss",
-                    params.valid_ctc_loss,
-                    params.batch_idx_train,
-                )
-                tb_writer.add_scalar(
-                    "train/valid_att_loss",
-                    params.valid_att_loss,
-                    params.batch_idx_train,
-                )
-                tb_writer.add_scalar(
-                    "train/valid_loss",
-                    params.valid_loss,
-                    params.batch_idx_train,
-                )
-
-    params.train_loss = params.tot_loss / params.tot_frames
 
+    loss_value = tot_loss['loss'] / tot_loss['frames']
+    params.train_loss = loss_value
     if params.train_loss < params.best_train_loss:
         params.best_train_epoch = params.cur_epoch
         params.best_train_loss = params.train_loss

egs/librispeech/ASR/tdnn_lstm_ctc/train.py

@@ -1,5 +1,6 @@
 #!/usr/bin/env python3
-# Copyright      2021  Xiaomi Corp.        (authors: Fangjun Kuang)
+# Copyright      2021  Xiaomi Corp.        (authors: Fangjun Kuang
+#                                                    Mingshuang Luo)
 #
 # See ../../../../LICENSE for clarification regarding multiple authors
 #
@@ -18,9 +19,10 @@
 
 import argparse
 import logging
+import collections
 from pathlib import Path
 from shutil import copyfile
-from typing import Optional
+from typing import Optional, Tuple, List
 
 import k2
 import torch
@@ -28,6 +30,8 @@ import torch.distributed as dist
 import torch.multiprocessing as mp
 import torch.nn as nn
 import torch.optim as optim
+from torch import Tensor
+
 from asr_datamodule import LibriSpeechAsrDataModule
 from lhotse.utils import fix_random_seed
 from model import TdnnLstm
@@ -260,6 +264,71 @@ def save_checkpoint(
         best_valid_filename = params.exp_dir / "best-valid-loss.pt"
         copyfile(src=filename, dst=best_valid_filename)
 
+class LossRecord(collections.defaultdict):
+    def __init__(self):
+        # Passing the type 'int' to the base-class constructor
+        # makes undefined items default to int() which is zero.
+        super(LossRecord, self).__init__(int)
+
+    def __add__(self, other: 'LossRecord') -> 'LossRecord':
+        ans = LossRecord()
+        for k, v in self.items():
+            ans[k] = v
+        for k, v in other.items():
+            ans[k] = ans[k] + v
+        return ans
+
+    def __mul__(self, alpha: float) -> 'LossRecord':
+        ans = LossRecord()
+        for k, v in self.items():
+            ans[k] = v * alpha
+        return ans
+
+    def __str__(self) -> str:
+        ans = ''
+        for k, v in self.norm_items():
+            norm_value = '%.4g' % v
+            ans += (str(k) + '=' + str(norm_value) + ', ')
+        frames = str(self['frames'])
+        ans += 'over ' + frames + ' frames.'
+        return ans
+
+    def norm_items(self) -> List[Tuple[str, float]]:
+        """
+        Returns a list of pairs, like:
+          [('ctc_loss', 0.1), ('att_loss', 0.07)]
+        """
+        num_frames = self['frames'] if 'frames' in self else 1
+        ans = []
+        for k, v in self.items():
+            if k != 'frames':
+                norm_value = float(v) / num_frames
+                ans.append((k, norm_value))
+        return ans
+
+    def reduce(self, device):
+        """
+        Reduce using torch.distributed, which I believe ensures that
+        all processes get the total.
+        """
+        keys = sorted(self.keys())
+        s = torch.tensor([ float(self[k]) for k in keys ],
+                         device=device)
+        dist.all_reduce(s, op=dist.ReduceOp.SUM)
+        for k, v in zip(keys, s.cpu().tolist()):
+            self[k] = v
+
+    def write_summary(self, tb_writer: SummaryWriter, prefix: str, batch_idx: int) -> None:
+        """
+        Add logging information to a TensorBoard writer.
+                tb_writer: a TensorBoard writer
+                  prefix: a prefix for the name of the loss, e.g. "train/valid_",
+                              or "train/current_"
+               batch_idx: The current batch index, used as the x-axis of the plot.
+        """
+        for k, v in self.norm_items():
+            tb_writer.add_scalar(prefix + k, v, batch_idx)
+
 
 def compute_loss(
     params: AttributeDict,
@@ -267,7 +336,7 @@ def compute_loss(
     batch: dict,
     graph_compiler: CtcTrainingGraphCompiler,
     is_training: bool,
-):
+) -> Tuple[Tensor, LossRecord]:
     """
     Compute CTC loss given the model and its inputs.
 
@@ -324,13 +393,12 @@ def compute_loss(
 
     assert loss.requires_grad == is_training
 
-    # train_frames and valid_frames are used for printing.
+    info = LossRecord()
-    if is_training:
+    # TODO: there are many GPU->CPU transfers here, maybe combine them into one.
-        params.train_frames = supervision_segments[:, 2].sum().item()
+    info['frames'] = supervision_segments[:, 2].sum().item()
-    else:
+    info['loss'] = loss.detach().cpu().item()
-        params.valid_frames = supervision_segments[:, 2].sum().item()
 
-    return loss
+    return loss, info
 
 
 def compute_validation_loss(
@@ -339,16 +407,16 @@ def compute_validation_loss(
     graph_compiler: CtcTrainingGraphCompiler,
     valid_dl: torch.utils.data.DataLoader,
     world_size: int = 1,
-) -> None:
+) -> LossRecord:
     """Run the validation process. The validation loss
     is saved in `params.valid_loss`.
     """
     model.eval()
 
-    tot_loss = 0.0
+    tot_loss = LossRecord()
-    tot_frames = 0.0
+
     for batch_idx, batch in enumerate(valid_dl):
-        loss = compute_loss(
+        loss, loss_info = compute_loss(
             params=params,
             model=model,
             batch=batch,
@@ -357,22 +425,18 @@ def compute_validation_loss(
         )
         assert loss.requires_grad is False
 
-        loss_cpu = loss.detach().cpu().item()
+        tot_loss = tot_loss + loss_info
-        tot_loss += loss_cpu
-        tot_frames += params.valid_frames
         
     if world_size > 1:
-        s = torch.tensor([tot_loss, tot_frames], device=loss.device)
+        tot_loss.reduce(loss.device)
-        dist.all_reduce(s, op=dist.ReduceOp.SUM)
-        s = s.cpu().tolist()
-        tot_loss = s[0]
-        tot_frames = s[1]
 
-    params.valid_loss = tot_loss / tot_frames
+    loss_value = tot_loss['loss'] / tot_loss['frames']
 
-    if params.valid_loss < params.best_valid_loss:
+    if loss_value < params.best_valid_loss:
         params.best_valid_epoch = params.cur_epoch
-        params.best_valid_loss = params.valid_loss
+        params.best_valid_loss = loss_value
+
+    return tot_loss
 
 
 def train_one_epoch(
@@ -411,23 +475,21 @@ def train_one_epoch(
     """
     model.train()
 
-    tot_loss = 0.0  # reset after params.reset_interval of batches
+    tot_loss = LossRecord()
-    tot_frames = 0.0  # reset after params.reset_interval of batches
-
-    params.tot_loss = 0.0
-    params.tot_frames = 0.0
 
     for batch_idx, batch in enumerate(train_dl):
         params.batch_idx_train += 1
         batch_size = len(batch["supervisions"]["text"])
 
-        loss = compute_loss(
+        loss, loss_info = compute_loss(
             params=params,
             model=model,
             batch=batch,
             graph_compiler=graph_compiler,
             is_training=True,
         )
+        # summary stats.
+        tot_loss = (tot_loss * (1 - 1 / params.reset_interval)) + loss_info 
 
         # NOTE: We use reduction==sum and loss is computed over utterances
         # in the batch and there is no normalization to it so far.
@@ -437,41 +499,19 @@ def train_one_epoch(
         clip_grad_norm_(model.parameters(), 5.0, 2.0)
         optimizer.step()
 
-        loss_cpu = loss.detach().cpu().item()
-
-        tot_frames += params.train_frames
-        tot_loss += loss_cpu
-        tot_avg_loss = tot_loss / tot_frames
-
-        params.tot_frames += params.train_frames
-        params.tot_loss += loss_cpu
-
         if batch_idx % params.log_interval == 0:
             logging.info(
-                f"Epoch {params.cur_epoch}, batch {batch_idx}, "
+                f"Epoch {params.cur_epoch}, batch {batch_idx}, loss[{loss_info}]"
-                f"batch avg loss {loss_cpu/params.train_frames:.4f}, "
+                f"tot_loss[{tot_loss}], batch size: {batch_size}"
-                f"total avg loss: {tot_avg_loss:.4f}, "
-                f"batch size: {batch_size}"
             )
+        if batch_idx % 10 == 0:
+
             if tb_writer is not None:
-                tb_writer.add_scalar(
+                loss_info.write_summary(tb_writer, "train/current_", params.batch_idx_train)
-                    "train/current_loss",
+                tot_loss.write_summary(tb_writer, "train/tot_", params.batch_idx_train)
-                    loss_cpu / params.train_frames,
-                    params.batch_idx_train,
-                )
-
-                tb_writer.add_scalar(
-                    "train/tot_avg_loss",
-                    tot_avg_loss,
-                    params.batch_idx_train,
-                )
-
-        if batch_idx > 0 and batch_idx % params.reset_interval == 0:
-            tot_loss = 0
-            tot_frames = 0
 
         if batch_idx > 0 and batch_idx % params.valid_interval == 0:
-            compute_validation_loss(
+            valid_info = compute_validation_loss(
                 params=params,
                 model=model,
                 graph_compiler=graph_compiler,
@@ -480,12 +520,17 @@ def train_one_epoch(
             )
             model.train()
             logging.info(
-                f"Epoch {params.cur_epoch}, valid loss {params.valid_loss:.4f},"
+                f"Epoch {params.cur_epoch}, validation {valid_info}"
-                f" best valid loss: {params.best_valid_loss:.4f} "
+                )
-                f"best valid epoch: {params.best_valid_epoch}"
+            if tb_writer is not None:
+                valid_info.write_summary(
+                    tb_writer,
+                    "train/valid_",
+                    params.batch_idx_train,
                 )
 
-    params.train_loss = params.tot_loss / params.tot_frames
+    loss_value = tot_loss['loss'] / tot_loss['frames']
+    params.train_loss = loss_value
 
     if params.train_loss < params.best_train_loss:
         params.best_train_epoch = params.cur_epoch

egs/yesno/ASR/tdnn/train.py

@@ -2,9 +2,10 @@
 
 import argparse
 import logging
+import collections
 from pathlib import Path
 from shutil import copyfile
-from typing import Optional
+from typing import Optional, Tuple, List
 
 import k2
 import torch
@@ -12,6 +13,8 @@ import torch.distributed as dist
 import torch.multiprocessing as mp
 import torch.nn as nn
 import torch.optim as optim
+from torch import Tensor
+
 from asr_datamodule import YesNoAsrDataModule
 from lhotse.utils import fix_random_seed
 from model import Tdnn
@@ -122,6 +125,8 @@ def get_params() -> AttributeDict:
 
         - valid_interval:  Run validation if batch_idx % valid_interval` is 0
 
+        - reset_interval: Reset statistics if batch_idx % reset_interval is 0
+
         - beam_size: It is used in k2.ctc_loss
 
         - reduction: It is used in k2.ctc_loss
@@ -142,6 +147,7 @@ def get_params() -> AttributeDict:
             "best_valid_epoch": -1,
             "batch_idx_train": 0,
             "log_interval": 10,
+            "reset_interval": 20,
             "valid_interval": 10,
             "beam_size": 10,
             "reduction": "sum",
@@ -239,13 +245,79 @@ def save_checkpoint(
         copyfile(src=filename, dst=best_valid_filename)
 
 
+class LossRecord(collections.defaultdict):
+    def __init__(self):
+        # Passing the type 'int' to the base-class constructor
+        # makes undefined items default to int() which is zero.
+        super(LossRecord, self).__init__(int)
+
+    def __add__(self, other: 'LossRecord') -> 'LossRecord':
+        ans = LossRecord()
+        for k, v in self.items():
+            ans[k] = v
+        for k, v in other.items():
+            ans[k] = ans[k] + v
+        return ans
+
+    def __mul__(self, alpha: float) -> 'LossRecord':
+        ans = LossRecord()
+        for k, v in self.items():
+            ans[k] = v * alpha
+        return ans
+
+    def __str__(self) -> str:
+        ans = ''
+        for k, v in self.norm_items():
+            norm_value = '%.4g' % v
+            ans += (str(k) + '=' + str(norm_value) + ', ')
+        frames = str(self['frames'])
+        ans += 'over ' + frames + ' frames.'
+        return ans
+
+    def norm_items(self) -> List[Tuple[str, float]]:
+        """
+        Returns a list of pairs, like:
+          [('ctc_loss', 0.1), ('att_loss', 0.07)]
+        """
+        num_frames = self['frames'] if 'frames' in self else 1
+        ans = []
+        for k, v in self.items():
+            if k != 'frames':
+                norm_value = float(v) / num_frames
+                ans.append((k, norm_value))
+        return ans
+
+    def reduce(self, device):
+        """
+        Reduce using torch.distributed, which I believe ensures that
+        all processes get the total.
+        """
+        keys = sorted(self.keys())
+        s = torch.tensor([ float(self[k]) for k in keys ],
+                         device=device)
+        dist.all_reduce(s, op=dist.ReduceOp.SUM)
+        for k, v in zip(keys, s.cpu().tolist()):
+            self[k] = v
+
+    def write_summary(self, tb_writer: SummaryWriter, prefix: str, batch_idx: int) -> None:
+        """
+        Add logging information to a TensorBoard writer.
+                tb_writer: a TensorBoard writer
+                  prefix: a prefix for the name of the loss, e.g. "train/valid_",
+                              or "train/current_"
+               batch_idx: The current batch index, used as the x-axis of the plot.
+        """
+        for k, v in self.norm_items():
+            tb_writer.add_scalar(prefix + k, v, batch_idx)
+
+
 def compute_loss(
     params: AttributeDict,
     model: nn.Module,
     batch: dict,
     graph_compiler: CtcTrainingGraphCompiler,
     is_training: bool,
-):
+) -> Tuple[Tensor, LossRecord]:
     """
     Compute CTC loss given the model and its inputs.
 
@@ -305,13 +377,12 @@ def compute_loss(
 
     assert loss.requires_grad == is_training
     
-    # train_frames and valid_frames are used for printing.
+    info = LossRecord()
-    if is_training:
+    # TODO: there are many GPU->CPU transfers here, maybe combine them into one.
-        params.train_frames = supervision_segments[:, 2].sum().item()
+    info['frames'] = supervision_segments[:, 2].sum().item()
-    else:
+    info['loss'] = loss.detach().cpu().item()
-        params.valid_frames = supervision_segments[:, 2].sum().item()
 
-    return loss
+    return loss, info
 
 
 def compute_validation_loss(
@@ -320,16 +391,16 @@ def compute_validation_loss(
     graph_compiler: CtcTrainingGraphCompiler,
     valid_dl: torch.utils.data.DataLoader,
     world_size: int = 1,
-) -> None:
+) -> LossRecord:
     """Run the validation process. The validation loss
     is saved in `params.valid_loss`.
     """
     model.eval()
 
-    tot_loss = 0.0
+    tot_loss = LossRecord()
-    tot_frames = 0.0
+    
     for batch_idx, batch in enumerate(valid_dl):
-        loss = compute_loss(
+        loss, loss_info = compute_loss(
             params=params,
             model=model,
             batch=batch,
@@ -338,22 +409,18 @@ def compute_validation_loss(
         )
         assert loss.requires_grad is False
 
-        loss_cpu = loss.detach().cpu().item()
+        tot_loss = tot_loss + loss_info
-        tot_loss += loss_cpu
-        tot_frames += params.valid_frames
         
     if world_size > 1:
-        s = torch.tensor([tot_loss, tot_frames], device=loss.device)
+        tot_loss.reduce(loss.device)
-        dist.all_reduce(s, op=dist.ReduceOp.SUM)
-        s = s.cpu().tolist()
-        tot_loss = s[0]
-        tot_frames = s[1]
 
-    params.valid_loss = tot_loss / tot_frames
+    loss_value = tot_loss['loss'] / tot_loss['frames']
 
-    if params.valid_loss < params.best_valid_loss:
+    if loss_value < params.best_valid_loss:
         params.best_valid_epoch = params.cur_epoch
-        params.best_valid_loss = params.valid_loss
+        params.best_valid_loss = loss_value
+
+    return tot_loss
 
 
 def train_one_epoch(
@@ -392,19 +459,21 @@ def train_one_epoch(
     """
     model.train()
     
-    tot_loss = 0.0  # sum of losses over all batches
+    tot_loss = LossRecord()
-    tot_frames = 0.0  # sum of frames over all batches
+
     for batch_idx, batch in enumerate(train_dl):
         params.batch_idx_train += 1
         batch_size = len(batch["supervisions"]["text"])
 
-        loss = compute_loss(
+        loss, loss_info = compute_loss(
             params=params,
             model=model,
             batch=batch,
             graph_compiler=graph_compiler,
             is_training=True,
         )
+        # summary stats.
+        tot_loss = (tot_loss * (1 - 1 / params.reset_interval)) + loss_info 
 
         # NOTE: We use reduction==sum and loss is computed over utterances
         # in the batch and there is no normalization to it so far.
@@ -414,35 +483,19 @@ def train_one_epoch(
         clip_grad_norm_(model.parameters(), 5.0, 2.0)
         optimizer.step()
 
-        loss_cpu = loss.detach().cpu().item()
-
-        tot_frames += params.train_frames
-        tot_loss += loss_cpu
-        tot_avg_loss = tot_loss / tot_frames
-
         if batch_idx % params.log_interval == 0:
             logging.info(
-                f"Epoch {params.cur_epoch}, batch {batch_idx}, "
+                f"Epoch {params.cur_epoch}, batch {batch_idx}, loss[{loss_info}]"
-                f"batch avg loss {loss_cpu/params.train_frames:.4f}, "
+                f"tot_loss[{tot_loss}], batch size: {batch_size}"
-                f"total avg loss: {tot_avg_loss:.4f}, "
-                f"batch size: {batch_size}"
             )
+        if batch_idx % 10 == 0:
 
             if tb_writer is not None:
-                tb_writer.add_scalar(
+                loss_info.write_summary(tb_writer, "train/current_", params.batch_idx_train)
-                    "train/current_loss",
+                tot_loss.write_summary(tb_writer, "train/tot_", params.batch_idx_train)
-                    loss_cpu / params.train_frames,
-                    params.batch_idx_train,
-                )
-
-                tb_writer.add_scalar(
-                    "train/tot_avg_loss",
-                    tot_avg_loss,
-                    params.batch_idx_train,
-                )
 
         if batch_idx > 0 and batch_idx % params.valid_interval == 0:
-            compute_validation_loss(
+            valid_info = compute_validation_loss(
                 params=params,
                 model=model,
                 graph_compiler=graph_compiler,
@@ -451,18 +504,17 @@ def train_one_epoch(
             )
             model.train()
             logging.info(
-                f"Epoch {params.cur_epoch}, valid loss {params.valid_loss:.4f},"
+                f"Epoch {params.cur_epoch}, validation {valid_info}"
-                f" best valid loss: {params.best_valid_loss:.4f} "
-                f"best valid epoch: {params.best_valid_epoch}"
                 )
             if tb_writer is not None:
-                tb_writer.add_scalar(
+                valid_info.write_summary(
-                    "train/valid_loss",
+                    tb_writer,
-                    params.valid_loss,
+                    "train/valid_",
                     params.batch_idx_train,
                 )
 
-    params.train_loss = tot_loss / tot_frames
+    loss_value = tot_loss['loss'] / tot_loss['frames']
+    params.train_loss = loss_value
 
     if params.train_loss < params.best_train_loss:
         params.best_train_epoch = params.cur_epoch