Black 20 release

examples/movement-pruning/masked_run_glue.py

@@ -173,7 +173,10 @@ def train(args, train_dataset, model, tokenizer, teacher=None):
     # Distributed training (should be after apex fp16 initialization)
     if args.local_rank != -1:
         model = torch.nn.parallel.DistributedDataParallel(
-            model, device_ids=[args.local_rank], output_device=args.local_rank, find_unused_parameters=True,
+            model,
+            device_ids=[args.local_rank],
+            output_device=args.local_rank,
+            find_unused_parameters=True,
         )
 
     # Train!
@@ -217,7 +220,10 @@ def train(args, train_dataset, model, tokenizer, teacher=None):
     tr_loss, logging_loss = 0.0, 0.0
     model.zero_grad()
     train_iterator = trange(
-        epochs_trained, int(args.num_train_epochs), desc="Epoch", disable=args.local_rank not in [-1, 0],
+        epochs_trained,
+        int(args.num_train_epochs),
+        desc="Epoch",
+        disable=args.local_rank not in [-1, 0],
     )
     set_seed(args)  # Added here for reproductibility
     for _ in train_iterator:
@@ -280,11 +286,14 @@ def train(args, train_dataset, model, tokenizer, teacher=None):
                         attention_mask=inputs["attention_mask"],
                     )
 
-                loss_logits = F.kl_div(
+                loss_logits = (
+                    F.kl_div(
                         input=F.log_softmax(logits_stu / args.temperature, dim=-1),
                         target=F.softmax(logits_tea / args.temperature, dim=-1),
                         reduction="batchmean",
-                ) * (args.temperature ** 2)
+                    )
+                    * (args.temperature ** 2)
+                )
 
                 loss = args.alpha_distil * loss_logits + args.alpha_ce * loss
 
@@ -529,7 +538,11 @@ def load_and_cache_examples(args, task, tokenizer, evaluate=False):
             processor.get_dev_examples(args.data_dir) if evaluate else processor.get_train_examples(args.data_dir)
         )
         features = convert_examples_to_features(
-            examples, tokenizer, max_length=args.max_seq_length, label_list=label_list, output_mode=output_mode,
+            examples,
+            tokenizer,
+            max_length=args.max_seq_length,
+            label_list=label_list,
+            output_mode=output_mode,
         )
         if args.local_rank in [-1, 0]:
             logger.info("Saving features into cached file %s", cached_features_file)
@@ -592,7 +605,10 @@ def main():
     )
     # Other parameters
     parser.add_argument(
-        "--config_name", default="", type=str, help="Pretrained config name or path if not the same as model_name",
+        "--config_name",
+        default="",
+        type=str,
+        help="Pretrained config name or path if not the same as model_name",
     )
     parser.add_argument(
         "--tokenizer_name",
@@ -616,17 +632,27 @@ def main():
     parser.add_argument("--do_train", action="store_true", help="Whether to run training.")
     parser.add_argument("--do_eval", action="store_true", help="Whether to run eval on the dev set.")
     parser.add_argument(
-        "--evaluate_during_training", action="store_true", help="Run evaluation during training at each logging step.",
+        "--evaluate_during_training",
+        action="store_true",
+        help="Run evaluation during training at each logging step.",
     )
     parser.add_argument(
-        "--do_lower_case", action="store_true", help="Set this flag if you are using an uncased model.",
+        "--do_lower_case",
+        action="store_true",
+        help="Set this flag if you are using an uncased model.",
     )
 
     parser.add_argument(
-        "--per_gpu_train_batch_size", default=8, type=int, help="Batch size per GPU/CPU for training.",
+        "--per_gpu_train_batch_size",
+        default=8,
+        type=int,
+        help="Batch size per GPU/CPU for training.",
     )
     parser.add_argument(
-        "--per_gpu_eval_batch_size", default=8, type=int, help="Batch size per GPU/CPU for evaluation.",
+        "--per_gpu_eval_batch_size",
+        default=8,
+        type=int,
+        help="Batch size per GPU/CPU for evaluation.",
     )
     parser.add_argument("--learning_rate", default=5e-5, type=float, help="The initial learning rate for Adam.")
 
@@ -723,7 +749,10 @@ def main():
     parser.add_argument("--adam_epsilon", default=1e-8, type=float, help="Epsilon for Adam optimizer.")
     parser.add_argument("--max_grad_norm", default=1.0, type=float, help="Max gradient norm.")
     parser.add_argument(
-        "--num_train_epochs", default=3.0, type=float, help="Total number of training epochs to perform.",
+        "--num_train_epochs",
+        default=3.0,
+        type=float,
+        help="Total number of training epochs to perform.",
     )
     parser.add_argument(
         "--max_steps",
@@ -742,10 +771,14 @@ def main():
     )
     parser.add_argument("--no_cuda", action="store_true", help="Avoid using CUDA when available")
     parser.add_argument(
-        "--overwrite_output_dir", action="store_true", help="Overwrite the content of the output directory",
+        "--overwrite_output_dir",
+        action="store_true",
+        help="Overwrite the content of the output directory",
     )
     parser.add_argument(
-        "--overwrite_cache", action="store_true", help="Overwrite the cached training and evaluation sets",
+        "--overwrite_cache",
+        action="store_true",
+        help="Overwrite the cached training and evaluation sets",
     )
     parser.add_argument("--seed", type=int, default=42, help="random seed for initialization")
 

examples/movement-pruning/masked_run_squad.py

@@ -181,7 +181,10 @@ def train(args, train_dataset, model, tokenizer, teacher=None):
     # Distributed training (should be after apex fp16 initialization)
     if args.local_rank != -1:
         model = torch.nn.parallel.DistributedDataParallel(
-            model, device_ids=[args.local_rank], output_device=args.local_rank, find_unused_parameters=True,
+            model,
+            device_ids=[args.local_rank],
+            output_device=args.local_rank,
+            find_unused_parameters=True,
         )
 
     # Train!
@@ -304,16 +307,22 @@ def train(args, train_dataset, model, tokenizer, teacher=None):
                         attention_mask=inputs["attention_mask"],
                     )
 
-                loss_start = F.kl_div(
+                loss_start = (
+                    F.kl_div(
                         input=F.log_softmax(start_logits_stu / args.temperature, dim=-1),
                         target=F.softmax(start_logits_tea / args.temperature, dim=-1),
                         reduction="batchmean",
-                ) * (args.temperature ** 2)
-                loss_end = F.kl_div(
+                    )
+                    * (args.temperature ** 2)
+                )
+                loss_end = (
+                    F.kl_div(
                         input=F.log_softmax(end_logits_stu / args.temperature, dim=-1),
                         target=F.softmax(end_logits_tea / args.temperature, dim=-1),
                         reduction="batchmean",
-                ) * (args.temperature ** 2)
+                    )
+                    * (args.temperature ** 2)
+                )
                 loss_logits = (loss_start + loss_end) / 2.0
 
                 loss = args.alpha_distil * loss_logits + args.alpha_ce * loss
@@ -859,7 +868,10 @@ def main():
     parser.add_argument("--adam_epsilon", default=1e-8, type=float, help="Epsilon for Adam optimizer.")
     parser.add_argument("--max_grad_norm", default=1.0, type=float, help="Max gradient norm.")
     parser.add_argument(
-        "--num_train_epochs", default=3.0, type=float, help="Total number of training epochs to perform.",
+        "--num_train_epochs",
+        default=3.0,
+        type=float,
+        help="Total number of training epochs to perform.",
     )
     parser.add_argument(
         "--max_steps",

examples/multiple-choice/utils_multiple_choice.py

@@ -100,7 +100,12 @@ if is_torch_available():
 
             cached_features_file = os.path.join(
                 data_dir,
-                "cached_{}_{}_{}_{}".format(mode.value, tokenizer.__class__.__name__, str(max_seq_length), task,),
+                "cached_{}_{}_{}_{}".format(
+                    mode.value,
+                    tokenizer.__class__.__name__,
+                    str(max_seq_length),
+                    task,
+                ),
             )
 
             # Make sure only the first process in distributed training processes the dataset,
@@ -121,7 +126,12 @@ if is_torch_available():
                     else:
                         examples = processor.get_train_examples(data_dir)
                     logger.info("Training examples: %s", len(examples))
-                    self.features = convert_examples_to_features(examples, label_list, max_seq_length, tokenizer,)
+                    self.features = convert_examples_to_features(
+                        examples,
+                        label_list,
+                        max_seq_length,
+                        tokenizer,
+                    )
                     logger.info("Saving features into cached file %s", cached_features_file)
                     torch.save(self.features, cached_features_file)
 
@@ -164,7 +174,12 @@ if is_tf_available():
                 examples = processor.get_train_examples(data_dir)
             logger.info("Training examples: %s", len(examples))
 
-            self.features = convert_examples_to_features(examples, label_list, max_seq_length, tokenizer,)
+            self.features = convert_examples_to_features(
+                examples,
+                label_list,
+                max_seq_length,
+                tokenizer,
+            )
 
             def gen():
                 for (ex_index, ex) in tqdm.tqdm(enumerate(self.features), desc="convert examples to features"):
@@ -491,7 +506,10 @@ class ArcProcessor(DataProcessor):
 
 
 def convert_examples_to_features(
-    examples: List[InputExample], label_list: List[str], max_length: int, tokenizer: PreTrainedTokenizer,
+    examples: List[InputExample],
+    label_list: List[str],
+    max_length: int,
+    tokenizer: PreTrainedTokenizer,
 ) -> List[InputFeatures]:
     """
     Loads a data file into a list of `InputFeatures`

examples/question-answering/run_squad_trainer.py

@@ -137,7 +137,12 @@ def main():
     )
 
     # Initialize our Trainer
-    trainer = Trainer(model=model, args=training_args, train_dataset=train_dataset, eval_dataset=eval_dataset,)
+    trainer = Trainer(
+        model=model,
+        args=training_args,
+        train_dataset=train_dataset,
+        eval_dataset=eval_dataset,
+    )
 
     # Training
     if training_args.do_train:

examples/question-answering/run_tf_squad.py

@@ -231,7 +231,12 @@ def main():
     eval_dataset = eval_dataset.apply(tf.data.experimental.assert_cardinality(len(eval_examples)))
 
     # Initialize our Trainer
-    trainer = TFTrainer(model=model, args=training_args, train_dataset=train_dataset, eval_dataset=eval_dataset,)
+    trainer = TFTrainer(
+        model=model,
+        args=training_args,
+        train_dataset=train_dataset,
+        eval_dataset=eval_dataset,
+    )
 
     # Training
     if training_args.do_train:

examples/seq2seq/bertabs/configuration_bertabs.py

@@ -28,7 +28,7 @@ BERTABS_FINETUNED_CONFIG_MAP = {
 
 
 class BertAbsConfig(PretrainedConfig):
-    r""" Class to store the configuration of the BertAbs model.
+    r"""Class to store the configuration of the BertAbs model.
 
     Arguments:
         vocab_size: int

examples/seq2seq/bertabs/convert_bertabs_original_pytorch_checkpoint.py

@@ -62,7 +62,7 @@ BertAbsConfig = namedtuple(
 
 
 def convert_bertabs_checkpoints(path_to_checkpoints, dump_path):
-    """ Copy/paste and tweak the pre-trained weights provided by the creators
+    """Copy/paste and tweak the pre-trained weights provided by the creators
     of BertAbs for the internal architecture.
     """
 
@@ -164,13 +164,22 @@ def convert_bertabs_checkpoints(path_to_checkpoints, dump_path):
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
     parser.add_argument(
-        "--bertabs_checkpoint_path", default=None, type=str, required=True, help="Path the official PyTorch dump.",
+        "--bertabs_checkpoint_path",
+        default=None,
+        type=str,
+        required=True,
+        help="Path the official PyTorch dump.",
     )
     parser.add_argument(
-        "--pytorch_dump_folder_path", default=None, type=str, required=True, help="Path to the output PyTorch model.",
+        "--pytorch_dump_folder_path",
+        default=None,
+        type=str,
+        required=True,
+        help="Path to the output PyTorch model.",
     )
     args = parser.parse_args()
 
     convert_bertabs_checkpoints(
-        args.bertabs_checkpoint_path, args.pytorch_dump_folder_path,
+        args.bertabs_checkpoint_path,
+        args.pytorch_dump_folder_path,
     )

examples/seq2seq/bertabs/modeling_bertabs.py

@@ -105,10 +105,17 @@ class BertAbs(BertAbsPreTrainedModel):
                 p.data.zero_()
 
     def forward(
-        self, encoder_input_ids, decoder_input_ids, token_type_ids, encoder_attention_mask, decoder_attention_mask,
+        self,
+        encoder_input_ids,
+        decoder_input_ids,
+        token_type_ids,
+        encoder_attention_mask,
+        decoder_attention_mask,
     ):
         encoder_output = self.bert(
-            input_ids=encoder_input_ids, token_type_ids=token_type_ids, attention_mask=encoder_attention_mask,
+            input_ids=encoder_input_ids,
+            token_type_ids=token_type_ids,
+            attention_mask=encoder_attention_mask,
         )
         encoder_hidden_states = encoder_output[0]
         dec_state = self.decoder.init_decoder_state(encoder_input_ids, encoder_hidden_states)
@@ -117,8 +124,7 @@ class BertAbs(BertAbsPreTrainedModel):
 
 
 class Bert(nn.Module):
-    """ This class is not really necessary and should probably disappear.
-    """
+    """This class is not really necessary and should probably disappear."""
 
     def __init__(self):
         super().__init__()
@@ -307,7 +313,14 @@ class TransformerDecoderLayer(nn.Module):
         self.register_buffer("mask", mask)
 
     def forward(
-        self, inputs, memory_bank, src_pad_mask, tgt_pad_mask, previous_input=None, layer_cache=None, step=None,
+        self,
+        inputs,
+        memory_bank,
+        src_pad_mask,
+        tgt_pad_mask,
+        previous_input=None,
+        layer_cache=None,
+        step=None,
     ):
         """
         Args:
@@ -331,13 +344,25 @@ class TransformerDecoderLayer(nn.Module):
             all_input = torch.cat((previous_input, input_norm), dim=1)
             dec_mask = None
 
-        query = self.self_attn(all_input, all_input, input_norm, mask=dec_mask, layer_cache=layer_cache, type="self",)
+        query = self.self_attn(
+            all_input,
+            all_input,
+            input_norm,
+            mask=dec_mask,
+            layer_cache=layer_cache,
+            type="self",
+        )
 
         query = self.drop(query) + inputs
 
         query_norm = self.layer_norm_2(query)
         mid = self.context_attn(
-            memory_bank, memory_bank, query_norm, mask=src_pad_mask, layer_cache=layer_cache, type="context",
+            memory_bank,
+            memory_bank,
+            query_norm,
+            mask=src_pad_mask,
+            layer_cache=layer_cache,
+            type="context",
         )
         output = self.feed_forward(self.drop(mid) + query)
 
@@ -422,7 +447,14 @@ class MultiHeadedAttention(nn.Module):
             self.final_linear = nn.Linear(model_dim, model_dim)
 
     def forward(
-        self, key, value, query, mask=None, layer_cache=None, type=None, predefined_graph_1=None,
+        self,
+        key,
+        value,
+        query,
+        mask=None,
+        layer_cache=None,
+        type=None,
+        predefined_graph_1=None,
     ):
         """
         Compute the context vector and the attention vectors.
@@ -628,7 +660,7 @@ def gelu(x):
 
 
 class PositionwiseFeedForward(nn.Module):
-    """ A two-layer Feed-Forward-Network with residual layer norm.
+    """A two-layer Feed-Forward-Network with residual layer norm.
 
     Args:
         d_model (int): the size of input for the first-layer of the FFN.
@@ -770,8 +802,7 @@ class Translator(object):
         self.max_length = args.max_length
 
     def translate(self, batch, step, attn_debug=False):
-        """ Generates summaries from one batch of data.
-        """
+        """Generates summaries from one batch of data."""
         self.model.eval()
         with torch.no_grad():
             batch_data = self.translate_batch(batch)
@@ -798,8 +829,7 @@ class Translator(object):
     # Where the beam search lives
     # I have no idea why it is being called from the method above
     def _fast_translate_batch(self, batch, max_length, min_length=0):
-        """ Beam Search using the encoder inputs contained in `batch`.
-        """
+        """Beam Search using the encoder inputs contained in `batch`."""
 
         # The batch object is funny
         # Instead of just looking at the size of the arguments we encapsulate
@@ -981,7 +1011,7 @@ def tile(x, count, dim=0):
 
 
 class BertSumOptimizer(object):
-    """ Specific optimizer for BertSum.
+    """Specific optimizer for BertSum.
 
     As described in [1], the authors fine-tune BertSum for abstractive
     summarization using two Adam Optimizers with different warm-up steps and
@@ -999,10 +1029,16 @@ class BertSumOptimizer(object):
 
         self.optimizers = {
             "encoder": torch.optim.Adam(
-                model.encoder.parameters(), lr=lr["encoder"], betas=(beta_1, beta_2), eps=eps,
+                model.encoder.parameters(),
+                lr=lr["encoder"],
+                betas=(beta_1, beta_2),
+                eps=eps,
             ),
             "decoder": torch.optim.Adam(
-                model.decoder.parameters(), lr=lr["decoder"], betas=(beta_1, beta_2), eps=eps,
+                model.decoder.parameters(),
+                lr=lr["decoder"],
+                betas=(beta_1, beta_2),
+                eps=eps,
             ),
         }
 

examples/seq2seq/bertabs/run_summarization.py

@@ -99,7 +99,7 @@ def evaluate(args):
 
 
 def save_summaries(summaries, path, original_document_name):
-    """ Write the summaries in fies that are prefixed by the original
+    """Write the summaries in fies that are prefixed by the original
     files' name with the `_summary` appended.
 
     Attributes:
@@ -125,7 +125,7 @@ def save_summaries(summaries, path, original_document_name):
 
 
 def format_summary(translation):
-    """ Transforms the output of the `from_batch` function
+    """Transforms the output of the `from_batch` function
     into nicely formatted summaries.
     """
     raw_summary, _, _ = translation
@@ -190,7 +190,12 @@ def build_data_iterator(args, tokenizer):
     def collate_fn(data):
         return collate(data, tokenizer, block_size=512, device=args.device)
 
-    iterator = DataLoader(dataset, sampler=sampler, batch_size=args.batch_size, collate_fn=collate_fn,)
+    iterator = DataLoader(
+        dataset,
+        sampler=sampler,
+        batch_size=args.batch_size,
+        collate_fn=collate_fn,
+    )
 
     return iterator
 
@@ -201,7 +206,7 @@ def load_and_cache_examples(args, tokenizer):
 
 
 def collate(data, tokenizer, block_size, device):
-    """ Collate formats the data passed to the data loader.
+    """Collate formats the data passed to the data loader.
 
     In particular we tokenize the data batch after batch to avoid keeping them
     all in memory. We output the data as a namedtuple to fit the original BertAbs's
@@ -231,7 +236,7 @@ def collate(data, tokenizer, block_size, device):
 
 
 def decode_summary(summary_tokens, tokenizer):
-    """ Decode the summary and return it in a format
+    """Decode the summary and return it in a format
     suitable for evaluation.
     """
     summary_tokens = summary_tokens.to("cpu").numpy()
@@ -242,8 +247,7 @@ def decode_summary(summary_tokens, tokenizer):
 
 
 def main():
-    """ The main function defines the interface with the users.
-    """
+    """The main function defines the interface with the users."""
     parser = argparse.ArgumentParser()
     parser.add_argument(
         "--documents_dir",
@@ -268,23 +272,41 @@ def main():
     )
     # EVALUATION options
     parser.add_argument(
-        "--no_cuda", default=False, type=bool, help="Whether to force the execution on CPU.",
+        "--no_cuda",
+        default=False,
+        type=bool,
+        help="Whether to force the execution on CPU.",
     )
     parser.add_argument(
-        "--batch_size", default=4, type=int, help="Batch size per GPU/CPU for training.",
+        "--batch_size",
+        default=4,
+        type=int,
+        help="Batch size per GPU/CPU for training.",
     )
     # BEAM SEARCH arguments
     parser.add_argument(
-        "--min_length", default=50, type=int, help="Minimum number of tokens for the summaries.",
+        "--min_length",
+        default=50,
+        type=int,
+        help="Minimum number of tokens for the summaries.",
     )
     parser.add_argument(
-        "--max_length", default=200, type=int, help="Maixmum number of tokens for the summaries.",
+        "--max_length",
+        default=200,
+        type=int,
+        help="Maixmum number of tokens for the summaries.",
     )
     parser.add_argument(
-        "--beam_size", default=5, type=int, help="The number of beams to start with for each example.",
+        "--beam_size",
+        default=5,
+        type=int,
+        help="The number of beams to start with for each example.",
     )
     parser.add_argument(
-        "--alpha", default=0.95, type=float, help="The value of alpha for the length penalty in the beam search.",
+        "--alpha",
+        default=0.95,
+        type=float,
+        help="The value of alpha for the length penalty in the beam search.",
     )
     parser.add_argument(
         "--block_trigram",

examples/seq2seq/bertabs/test_utils_summarization.py

@@ -43,8 +43,7 @@ class SummarizationDataProcessingTest(unittest.TestCase):
         self.assertEqual(truncate_or_pad(sequence, self.block_size, 0), expected_output)
 
     def test_process_story_no_highlights(self):
-        """ Processing a story with no highlights returns an empty list for the summary.
-        """
+        """Processing a story with no highlights returns an empty list for the summary."""
         raw_story = """It was the year of Our Lord one thousand seven hundred and
         seventy-five.\n\nSpiritual revelations were conceded to England at that
         favoured period, as at this."""
@@ -52,8 +51,7 @@ class SummarizationDataProcessingTest(unittest.TestCase):
         self.assertEqual(summary_lines, [])
 
     def test_process_empty_story(self):
-        """ An empty story returns an empty collection of lines.
-        """
+        """An empty story returns an empty collection of lines."""
         raw_story = ""
         story_lines, summary_lines = process_story(raw_story)
         self.assertEqual(story_lines, [])

examples/seq2seq/bertabs/utils_summarization.py

@@ -11,7 +11,7 @@ from torch.utils.data import Dataset
 
 
 class CNNDMDataset(Dataset):
-    """ Abstracts the dataset used to train seq2seq models.
+    """Abstracts the dataset used to train seq2seq models.
 
     The class will process the documents that are located in the specified
     folder. The preprocessing will work on any document that is reasonably
@@ -31,7 +31,7 @@ class CNNDMDataset(Dataset):
     """
 
     def __init__(self, path="", prefix="train"):
-        """ We initialize the class by listing all the documents to summarize.
+        """We initialize the class by listing all the documents to summarize.
         Files are not read in memory due to the size of some datasets (like CNN/DailyMail).
         """
         assert os.path.isdir(path)
@@ -60,7 +60,7 @@ class CNNDMDataset(Dataset):
 
 
 def process_story(raw_story):
-    """ Extract the story and summary from a story file.
+    """Extract the story and summary from a story file.
 
     Arguments:
         raw_story (str): content of the story file as an utf-8 encoded string.
@@ -108,7 +108,7 @@ def _add_missing_period(line):
 
 
 def truncate_or_pad(sequence, block_size, pad_token_id):
-    """ Adapt the source and target sequences' lengths to the block size.
+    """Adapt the source and target sequences' lengths to the block size.
     If the sequence is shorter we append padding token to the right of the sequence.
     """
     if len(sequence) > block_size:
@@ -119,8 +119,8 @@ def truncate_or_pad(sequence, block_size, pad_token_id):
 
 
 def build_mask(sequence, pad_token_id):
-    """ Builds the mask. The attention mechanism will only attend to positions
-    with value 1. """
+    """Builds the mask. The attention mechanism will only attend to positions
+    with value 1."""
     mask = torch.ones_like(sequence)
     idx_pad_tokens = sequence == pad_token_id
     mask[idx_pad_tokens] = 0
@@ -128,7 +128,7 @@ def build_mask(sequence, pad_token_id):
 
 
 def encode_for_summarization(story_lines, summary_lines, tokenizer):
-    """ Encode the story and summary lines, and join them
+    """Encode the story and summary lines, and join them
     as specified in [1] by using `[SEP] [CLS]` tokens to separate
     sentences.
     """
@@ -141,7 +141,7 @@ def encode_for_summarization(story_lines, summary_lines, tokenizer):
 
 
 def compute_token_type_ids(batch, separator_token_id):
-    """ Segment embeddings as described in [1]
+    """Segment embeddings as described in [1]
 
     The values {0,1} were found in the repository [2].
 

examples/seq2seq/callbacks.py

@@ -97,4 +97,9 @@ def get_checkpoint_callback(output_dir, metric):
 
 
 def get_early_stopping_callback(metric, patience):
-    return EarlyStopping(monitor=f"val_{metric}", mode="max", patience=patience, verbose=True,)
+    return EarlyStopping(
+        monitor=f"val_{metric}",
+        mode="max",
+        patience=patience,
+        verbose=True,
+    )

examples/seq2seq/distillation.py

@@ -348,7 +348,10 @@ class T5SummarizationDistiller(BartSummarizationDistiller):
         if self.different_encoder:
             with torch.no_grad():
                 teacher_enc_outputs, teacher_enc_hid = self.teacher.encoder(
-                    source_ids, attention_mask=source_mask, output_hidden_states=True, use_cache=False,
+                    source_ids,
+                    attention_mask=source_mask,
+                    output_hidden_states=True,
+                    use_cache=False,
                 )
             if self.hparams.alpha_encoder_loss > 0:
                 loss_encoder = self.calc_mse_loss(enc_outputs, teacher_enc_outputs, source_mask)

examples/seq2seq/test_seq2seq_examples.py

@@ -117,7 +117,12 @@ class TestSummarizationDistiller(unittest.TestCase):
 
     @require_multigpu
     def test_multigpu(self):
-        updates = dict(no_teacher=True, freeze_encoder=True, gpus=2, sortish_sampler=False,)
+        updates = dict(
+            no_teacher=True,
+            freeze_encoder=True,
+            gpus=2,
+            sortish_sampler=False,
+        )
         self._test_distiller_cli(updates)
 
     def test_distill_no_teacher(self):
@@ -261,7 +266,8 @@ def test_run_eval_bart(model):
 
 
 @pytest.mark.parametrize(
-    ["model"], [pytest.param(T5_TINY), pytest.param(BART_TINY), pytest.param(MBART_TINY), pytest.param(MARIAN_TINY)],
+    ["model"],
+    [pytest.param(T5_TINY), pytest.param(BART_TINY), pytest.param(MBART_TINY), pytest.param(MARIAN_TINY)],
 )
 def test_finetune(model):
     args_d: dict = CHEAP_ARGS.copy()
@@ -329,7 +335,8 @@ def test_finetune_extra_model_args():
     output_dir = tempfile.mkdtemp(prefix="output_1_")
     args_d1 = args_d.copy()
     args_d1.update(
-        model_name_or_path=model, output_dir=output_dir,
+        model_name_or_path=model,
+        output_dir=output_dir,
     )
     extra_model_params = ("encoder_layerdrop", "decoder_layerdrop", "dropout", "attention_dropout")
     for p in extra_model_params:
@@ -344,7 +351,8 @@ def test_finetune_extra_model_args():
     output_dir = tempfile.mkdtemp(prefix="output_2_")
     args_d2 = args_d.copy()
     args_d2.update(
-        model_name_or_path=model, output_dir=output_dir,
+        model_name_or_path=model,
+        output_dir=output_dir,
     )
     unsupported_param = "encoder_layerdrop"
     args_d2[unsupported_param] = 0.5
@@ -478,7 +486,11 @@ def test_summarization_dataset_truncation(tok):
     max_len_target = max(len(tokenizer.encode(a)) for a in SUMMARIES)
     trunc_target = 4
     train_dataset = Seq2SeqDataset(
-        tokenizer, data_dir=tmp_dir, type_path="train", max_source_length=20, max_target_length=trunc_target,
+        tokenizer,
+        data_dir=tmp_dir,
+        type_path="train",
+        max_source_length=20,
+        max_target_length=trunc_target,
     )
     dataloader = DataLoader(train_dataset, batch_size=2, collate_fn=train_dataset.collate_fn)
     for batch in dataloader:

examples/seq2seq/utils.py

@@ -63,7 +63,9 @@ def calculate_bleu(output_lns, refs_lns, **kwargs) -> dict:
 
 
 def trim_batch(
-    input_ids, pad_token_id, attention_mask=None,
+    input_ids,
+    pad_token_id,
+    attention_mask=None,
 ):
     """Remove columns that are populated exclusively by pad_token_id"""
     keep_column_mask = input_ids.ne(pad_token_id).any(dim=0)

examples/text-classification/run_pl_glue.py

@@ -153,7 +153,11 @@ class GLUETransformer(BaseTransformer):
         )
 
         parser.add_argument(
-            "--task", default="", type=str, required=True, help="The GLUE task to run",
+            "--task",
+            default="",
+            type=str,
+            required=True,
+            help="The GLUE task to run",
         )
         parser.add_argument(
             "--gpus",
@@ -177,7 +181,10 @@ def main():
 
     # If output_dir not provided, a folder will be generated in pwd
     if args.output_dir is None:
-        args.output_dir = os.path.join("./results", f"{args.task}_{time.strftime('%Y%m%d_%H%M%S')}",)
+        args.output_dir = os.path.join(
+            "./results",
+            f"{args.task}_{time.strftime('%Y%m%d_%H%M%S')}",
+        )
         os.makedirs(args.output_dir)
 
     model = GLUETransformer(args)

examples/text-classification/run_xnli.py

@@ -328,7 +328,11 @@ def load_and_cache_examples(args, task, tokenizer, evaluate=False):
             processor.get_test_examples(args.data_dir) if evaluate else processor.get_train_examples(args.data_dir)
         )
         features = convert_examples_to_features(
-            examples, tokenizer, max_length=args.max_seq_length, label_list=label_list, output_mode=output_mode,
+            examples,
+            tokenizer,
+            max_length=args.max_seq_length,
+            label_list=label_list,
+            output_mode=output_mode,
         )
         if args.local_rank in [-1, 0]:
             logger.info("Saving features into cached file %s", cached_features_file)

examples/text-generation/pplm/run_pplm.py

@@ -698,7 +698,9 @@ def run_pplm_example(
                 for word_id in pert_gen_tok_text.tolist()[0]:
                     if word_id in bow_word_ids:
                         pert_gen_text += "{}{}{}".format(
-                            colorama.Fore.RED, tokenizer.decode([word_id]), colorama.Style.RESET_ALL,
+                            colorama.Fore.RED,
+                            tokenizer.decode([word_id]),
+                            colorama.Style.RESET_ALL,
                         )
                     else:
                         pert_gen_text += tokenizer.decode([word_id])
@@ -729,7 +731,10 @@ if __name__ == "__main__":
     parser.add_argument("--cond_text", type=str, default="The lake", help="Prefix texts to condition on")
     parser.add_argument("--uncond", action="store_true", help="Generate from end-of-text as prefix")
     parser.add_argument(
-        "--num_samples", type=int, default=1, help="Number of samples to generate from the modified latents",
+        "--num_samples",
+        type=int,
+        default=1,
+        help="Number of samples to generate from the modified latents",
     )
     parser.add_argument(
         "--bag_of_words",
@@ -751,13 +756,22 @@ if __name__ == "__main__":
         help="Discriminator to use",
     )
     parser.add_argument(
-        "--discrim_weights", type=str, default=None, help="Weights for the generic discriminator",
+        "--discrim_weights",
+        type=str,
+        default=None,
+        help="Weights for the generic discriminator",
     )
     parser.add_argument(
-        "--discrim_meta", type=str, default=None, help="Meta information for the generic discriminator",
+        "--discrim_meta",
+        type=str,
+        default=None,
+        help="Meta information for the generic discriminator",
     )
     parser.add_argument(
-        "--class_label", type=int, default=-1, help="Class label used for the discriminator",
+        "--class_label",
+        type=int,
+        default=-1,
+        help="Class label used for the discriminator",
     )
     parser.add_argument("--length", type=int, default=100)
     parser.add_argument("--stepsize", type=float, default=0.02)
@@ -773,7 +787,10 @@ if __name__ == "__main__":
         help="Length of past which is being optimized; 0 corresponds to infinite window length",
     )
     parser.add_argument(
-        "--horizon_length", type=int, default=1, help="Length of future to optimize over",
+        "--horizon_length",
+        type=int,
+        default=1,
+        help="Length of future to optimize over",
     )
     parser.add_argument("--decay", action="store_true", help="whether to decay or not")
     parser.add_argument("--gamma", type=float, default=1.5)
@@ -783,7 +800,10 @@ if __name__ == "__main__":
     parser.add_argument("--no_cuda", action="store_true", help="no cuda")
     parser.add_argument("--colorama", action="store_true", help="colors keywords")
     parser.add_argument(
-        "--repetition_penalty", type=float, default=1.0, help="Penalize repetition. More than 1.0 -> less repetition",
+        "--repetition_penalty",
+        type=float,
+        default=1.0,
+        help="Penalize repetition. More than 1.0 -> less repetition",
     )
 
     args = parser.parse_args()

examples/text-generation/pplm/run_pplm_discrim_train.py

@@ -242,7 +242,12 @@ def train_discriminator(
 
         text = torchtext_data.Field()
         label = torchtext_data.Field(sequential=False)
-        train_data, val_data, test_data = datasets.SST.splits(text, label, fine_grained=True, train_subtrees=True,)
+        train_data, val_data, test_data = datasets.SST.splits(
+            text,
+            label,
+            fine_grained=True,
+            train_subtrees=True,
+        )
 
         x = []
         y = []

examples/text-generation/run_generation.py

@@ -41,7 +41,9 @@ from transformers import (
 
 
 logging.basicConfig(
-    format="%(asctime)s - %(levelname)s - %(name)s -   %(message)s", datefmt="%m/%d/%Y %H:%M:%S", level=logging.INFO,
+    format="%(asctime)s - %(levelname)s - %(name)s -   %(message)s",
+    datefmt="%m/%d/%Y %H:%M:%S",
+    level=logging.INFO,
 )
 logger = logging.getLogger(__name__)
 
@@ -197,7 +199,10 @@ def main():
     args.n_gpu = 0 if args.no_cuda else torch.cuda.device_count()
 
     logger.warning(
-        "device: %s, n_gpu: %s, 16-bits training: %s", args.device, args.n_gpu, args.fp16,
+        "device: %s, n_gpu: %s, 16-bits training: %s",
+        args.device,
+        args.n_gpu,
+        args.fp16,
     )
 
     set_seed(args)