Small features + lint

Summary: Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/588

Differential Revision: D15389638

Pulled By: myleott

fbshipit-source-id: 4632ce22d51dc2c74d250bae999630095d849701

fairseq/checkpoint_utils.py

@@ -103,6 +103,7 @@ def load_checkpoint(args, trainer):
         args.reset_optimizer,
         args.reset_lr_scheduler,
         eval(args.optimizer_overrides),
+        reset_meters=args.reset_meters,
     )
 
     if extra_state is not None and 'best' in extra_state and not args.reset_optimizer:

fairseq/data/concat_dataset.py

@@ -50,11 +50,12 @@ class ConcatDataset(FairseqDataset):
 
     @property
     def supports_prefetch(self):
-        return all([d.supports_prefetch for d in self.datasets])
+        return any(getattr(d, 'supports_prefetch', False) for d in self.datasets)
 
     def prefetch(self, indices):
         frm = 0
         for to, ds in zip(self.cumulative_sizes, self.datasets):
             real_size = len(ds)
+            if getattr(ds, 'supports_prefetch', False):
                 ds.prefetch([(i - frm) % real_size for i in indices if frm <= i < to])
             frm = to

fairseq/data/data_utils.py

@@ -21,7 +21,7 @@ def infer_language_pair(path):
     return src, dst
 
 
-def collate_tokens(values, pad_idx, eos_idx, left_pad, move_eos_to_beginning=False):
+def collate_tokens(values, pad_idx, eos_idx=None, left_pad=False, move_eos_to_beginning=False):
     """Convert a list of 1d tensors into a padded 2d tensor."""
     size = max(v.size(0) for v in values)
     res = values[0].new(len(values), size).fill_(pad_idx)

fairseq/data/fairseq_dataset.py

@@ -21,7 +21,7 @@ class FairseqDataset(torch.utils.data.Dataset):
         """Merge a list of samples to form a mini-batch.
 
         Args:
-            samples (List[int]): sample indices to collate
+            samples (List[dict]): samples to collate
 
         Returns:
             dict: a mini-batch suitable for forwarding with a Model

fairseq/data/indexed_dataset.py

@@ -4,6 +4,7 @@
 # This source code is licensed under the license found in the LICENSE file in
 # the root directory of this source tree. An additional grant of patent rights
 # can be found in the PATENTS file in the same directory.
+
 import os
 import shutil
 import struct
@@ -11,6 +12,8 @@ import struct
 import numpy as np
 import torch
 
+from . import FairseqDataset
+
 
 def make_builder(out_file, impl):
     if impl == 'mmap':
@@ -78,7 +81,7 @@ def data_file_path(prefix_path):
     return prefix_path + '.bin'
 
 
-class IndexedDataset(torch.utils.data.Dataset):
+class IndexedDataset(FairseqDataset):
     """Loader for TorchNet IndexedDataset"""
 
     def __init__(self, path, fix_lua_indexing=False):
@@ -99,16 +102,16 @@ class IndexedDataset(torch.utils.data.Dataset):
             assert struct.unpack('<Q', version) == (1,)
             code, self.element_size = struct.unpack('<QQ', f.read(16))
             self.dtype = dtypes[code]
-            self.size, self.s = struct.unpack('<QQ', f.read(16))
-            self.dim_offsets = read_longs(f, self.size + 1)
-            self.data_offsets = read_longs(f, self.size + 1)
+            self._len, self.s = struct.unpack('<QQ', f.read(16))
+            self.dim_offsets = read_longs(f, self._len + 1)
+            self.data_offsets = read_longs(f, self._len + 1)
             self.sizes = read_longs(f, self.s)
 
     def read_data(self, path):
         self.data_file = open(data_file_path(path), 'rb', buffering=0)
 
     def check_index(self, i):
-        if i < 0 or i >= self.size:
+        if i < 0 or i >= self._len:
             raise IndexError('index out of range')
 
     def __del__(self):
@@ -129,7 +132,13 @@ class IndexedDataset(torch.utils.data.Dataset):
         return item
 
     def __len__(self):
-        return self.size
+        return self._len
+
+    def num_tokens(self, index):
+        return self.sizes[index]
+
+    def size(self, index):
+        return self.sizes[index]
 
     @staticmethod
     def exists(path):
@@ -189,7 +198,7 @@ class IndexedCachedDataset(IndexedDataset):
         return item
 
 
-class IndexedRawTextDataset(torch.utils.data.Dataset):
+class IndexedRawTextDataset(FairseqDataset):
     """Takes a text file as input and binarizes it in memory at instantiation.
     Original lines are also kept in memory"""
 
@@ -232,6 +241,12 @@ class IndexedRawTextDataset(torch.utils.data.Dataset):
     def __len__(self):
         return self.size
 
+    def num_tokens(self, index):
+        return self.sizes[index]
+
+    def size(self, index):
+        return self.sizes[index]
+
     @staticmethod
     def exists(path):
         return os.path.exists(path)

fairseq/models/fairseq_model.py

@@ -68,7 +68,7 @@ class BaseFairseqModel(nn.Module):
         this additionally "upgrades" *state_dicts* from old checkpoints.
         """
         self.upgrade_state_dict(state_dict)
-        super().load_state_dict(state_dict, strict)
+        return super().load_state_dict(state_dict, strict)
 
     def upgrade_state_dict(self, state_dict):
         """Upgrade old state dicts to work with newer code."""

fairseq/options.py

@@ -330,6 +330,8 @@ def add_checkpoint_args(parser):
                        help='if set, does not load lr scheduler state from the checkpoint')
     group.add_argument('--optimizer-overrides', default="{}", type=str, metavar='DICT',
                        help='a dictionary used to override optimizer args when loading a checkpoint')
+    group.add_argument('--reset-meters', action='store_true',
+                       help='if set, does not load meters from the checkpoint')
     group.add_argument('--save-interval', type=int, default=1, metavar='N',
                        help='save a checkpoint every N epochs')
     group.add_argument('--save-interval-updates', type=int, default=0, metavar='N',

fairseq/trainer.py

@@ -129,7 +129,14 @@ class Trainer(object):
                 self._optim_history, extra_state,
             )
 
-    def load_checkpoint(self, filename, reset_optimizer=False, reset_lr_scheduler=False, optimizer_overrides=None):
+    def load_checkpoint(
+        self,
+        filename,
+        reset_optimizer=False,
+        reset_lr_scheduler=False,
+        optimizer_overrides=None,
+        reset_meters=False,
+    ):
         """Load all training state from a checkpoint file."""
         extra_state, self._optim_history, last_optim_state = None, [], None