Fix E266 flake8 warning (x90).

examples/contrib/run_swag.py

@@ -487,7 +487,7 @@ def evaluate(args, model, tokenizer, prefix=""):
 def main():
     parser = argparse.ArgumentParser()
 
-    ## Required parameters
+    # Required parameters
     parser.add_argument(
         "--train_file", default=None, type=str, required=True, help="SWAG csv for training. E.g., train.csv"
     )
@@ -520,7 +520,7 @@ def main():
         help="The output directory where the model checkpoints and predictions will be written.",
     )
 
-    ## Other parameters
+    # Other parameters
     parser.add_argument(
         "--config_name", default="", type=str, help="Pretrained config name or path if not the same as model_name"
     )

examples/distillation/run_squad_w_distillation.py

@@ -430,7 +430,7 @@ def load_and_cache_examples(args, tokenizer, evaluate=False, output_examples=Fal
 def main():
     parser = argparse.ArgumentParser()
 
-    ## Required parameters
+    # Required parameters
     parser.add_argument(
         "--train_file", default=None, type=str, required=True, help="SQuAD json for training. E.g., train-v1.1.json"
     )
@@ -486,7 +486,7 @@ def main():
         "--temperature", default=2.0, type=float, help="Distillation temperature. Only for distillation."
     )
 
-    ## Other parameters
+    # Other parameters
     parser.add_argument(
         "--config_name", default="", type=str, help="Pretrained config name or path if not the same as model_name"
     )

examples/distillation/scripts/extract.py

@@ -43,7 +43,7 @@ if __name__ == "__main__":
     state_dict = model.state_dict()
     compressed_sd = {}
 
-    ### Embeddings ###
+    # Embeddings #
     if args.model_type == "gpt2":
         for param_name in ["wte.weight", "wpe.weight"]:
             compressed_sd[f"{prefix}.{param_name}"] = state_dict[f"{prefix}.{param_name}"]
@@ -55,7 +55,7 @@ if __name__ == "__main__":
             param_name = f"{prefix}.embeddings.LayerNorm.{w}"
             compressed_sd[param_name] = state_dict[param_name]
 
-    ### Transformer Blocks ###
+    # Transformer Blocks #
     std_idx = 0
     for teacher_idx in [0, 2, 4, 7, 9, 11]:
         if args.model_type == "gpt2":
@@ -82,7 +82,7 @@ if __name__ == "__main__":
                     ]
         std_idx += 1
 
-    ### Language Modeling Head ###s
+    # Language Modeling Head ###s
     if args.model_type == "roberta":
         for layer in ["lm_head.decoder.weight", "lm_head.bias"]:
             compressed_sd[f"{layer}"] = state_dict[f"{layer}"]

examples/distillation/train.py

@@ -219,7 +219,7 @@ def main():
     args = parser.parse_args()
     sanity_checks(args)
 
-    ## ARGS ##
+    # ARGS #
     init_gpu_params(args)
     set_seed(args)
     if args.is_master:
@@ -236,7 +236,7 @@ def main():
             os.makedirs(args.dump_path)
         logger.info(f"Experiment will be dumped and logged in {args.dump_path}")
 
-        ### SAVE PARAMS ###
+        # SAVE PARAMS #
         logger.info(f"Param: {args}")
         with open(os.path.join(args.dump_path, "parameters.json"), "w") as f:
             json.dump(vars(args), f, indent=4)
@@ -245,7 +245,7 @@ def main():
     student_config_class, student_model_class, _ = MODEL_CLASSES[args.student_type]
     teacher_config_class, teacher_model_class, teacher_tokenizer_class = MODEL_CLASSES[args.teacher_type]
 
-    ### TOKENIZER ###
+    # TOKENIZER #
     tokenizer = teacher_tokenizer_class.from_pretrained(args.teacher_name)
     special_tok_ids = {}
     for tok_name, tok_symbol in tokenizer.special_tokens_map.items():
@@ -255,7 +255,7 @@ def main():
     args.special_tok_ids = special_tok_ids
     args.max_model_input_size = tokenizer.max_model_input_sizes[args.teacher_name]
 
-    ## DATA LOADER ##
+    # DATA LOADER #
     logger.info(f"Loading data from {args.data_file}")
     with open(args.data_file, "rb") as fp:
         data = pickle.load(fp)
@@ -275,7 +275,7 @@ def main():
     train_lm_seq_dataset = LmSeqsDataset(params=args, data=data)
     logger.info(f"Data loader created.")
 
-    ## STUDENT ##
+    # STUDENT #
     logger.info(f"Loading student config from {args.student_config}")
     stu_architecture_config = student_config_class.from_pretrained(args.student_config)
     stu_architecture_config.output_hidden_states = True
@@ -290,26 +290,26 @@ def main():
         student.to(f"cuda:{args.local_rank}")
     logger.info(f"Student loaded.")
 
-    ## TEACHER ##
+    # TEACHER #
     teacher = teacher_model_class.from_pretrained(args.teacher_name, output_hidden_states=True)
     if args.n_gpu > 0:
         teacher.to(f"cuda:{args.local_rank}")
     logger.info(f"Teacher loaded from {args.teacher_name}.")
 
-    ## FREEZING ##
+    # FREEZING #
     if args.freeze_pos_embs:
         freeze_pos_embeddings(student, args)
     if args.freeze_token_type_embds:
         freeze_token_type_embeddings(student, args)
 
-    ## SANITY CHECKS ##
+    # SANITY CHECKS #
     assert student.config.vocab_size == teacher.config.vocab_size
     assert student.config.hidden_size == teacher.config.hidden_size
     assert student.config.max_position_embeddings == teacher.config.max_position_embeddings
     if args.mlm:
         assert token_probs.size(0) == stu_architecture_config.vocab_size
 
-    ## DISTILLER ##
+    # DISTILLER #
     torch.cuda.empty_cache()
     distiller = Distiller(
         params=args, dataset=train_lm_seq_dataset, token_probs=token_probs, student=student, teacher=teacher

examples/mm-imdb/run_mmimdb.py

@@ -344,7 +344,7 @@ def load_examples(args, tokenizer, evaluate=False):
 def main():
     parser = argparse.ArgumentParser()
 
-    ## Required parameters
+    # Required parameters
     parser.add_argument(
         "--data_dir",
         default=None,
@@ -374,7 +374,7 @@ def main():
         help="The output directory where the model predictions and checkpoints will be written.",
     )
 
-    ## Other parameters
+    # Other parameters
     parser.add_argument(
         "--config_name", default="", type=str, help="Pretrained config name or path if not the same as model_name"
     )

examples/run_bertology.py

@@ -242,7 +242,7 @@ def prune_heads(args, model, eval_dataloader, head_mask):
 
 def main():
     parser = argparse.ArgumentParser()
-    ## Required parameters
+    # Required parameters
     parser.add_argument(
         "--data_dir",
         default=None,
@@ -272,7 +272,7 @@ def main():
         help="The output directory where the model predictions and checkpoints will be written.",
     )
 
-    ## Other parameters
+    # Other parameters
     parser.add_argument(
         "--config_name",
         default="",

examples/run_glue.py

@@ -410,7 +410,7 @@ def load_and_cache_examples(args, task, tokenizer, evaluate=False):
 def main():
     parser = argparse.ArgumentParser()
 
-    ## Required parameters
+    # Required parameters
     parser.add_argument(
         "--data_dir",
         default=None,
@@ -447,7 +447,7 @@ def main():
         help="The output directory where the model predictions and checkpoints will be written.",
     )
 
-    ## Other parameters
+    # Other parameters
     parser.add_argument(
         "--config_name", default="", type=str, help="Pretrained config name or path if not the same as model_name"
     )

examples/run_lm_finetuning.py

@@ -422,7 +422,7 @@ def evaluate(args, model, tokenizer, prefix=""):
 def main():
     parser = argparse.ArgumentParser()
 
-    ## Required parameters
+    # Required parameters
     parser.add_argument(
         "--train_data_file", default=None, type=str, required=True, help="The input training data file (a text file)."
     )
@@ -434,7 +434,7 @@ def main():
         help="The output directory where the model predictions and checkpoints will be written.",
     )
 
-    ## Other parameters
+    # Other parameters
     parser.add_argument(
         "--eval_data_file",
         default=None,

examples/run_multiple_choice.py

@@ -385,7 +385,7 @@ def load_and_cache_examples(args, task, tokenizer, evaluate=False, test=False):
 def main():
     parser = argparse.ArgumentParser()
 
-    ## Required parameters
+    # Required parameters
     parser.add_argument(
         "--data_dir",
         default=None,
@@ -422,7 +422,7 @@ def main():
         help="The output directory where the model predictions and checkpoints will be written.",
     )
 
-    ## Other parameters
+    # Other parameters
     parser.add_argument(
         "--config_name", default="", type=str, help="Pretrained config name or path if not the same as model_name"
     )

examples/run_ner.py

@@ -385,7 +385,7 @@ def load_and_cache_examples(args, tokenizer, labels, pad_token_label_id, mode):
 def main():
     parser = argparse.ArgumentParser()
 
-    ## Required parameters
+    # Required parameters
     parser.add_argument(
         "--data_dir",
         default=None,
@@ -415,7 +415,7 @@ def main():
         help="The output directory where the model predictions and checkpoints will be written.",
     )
 
-    ## Other parameters
+    # Other parameters
     parser.add_argument(
         "--labels",
         default="",

examples/run_xnli.py

@@ -377,7 +377,7 @@ def load_and_cache_examples(args, task, tokenizer, evaluate=False):
 def main():
     parser = argparse.ArgumentParser()
 
-    ## Required parameters
+    # Required parameters
     parser.add_argument(
         "--data_dir",
         default=None,
@@ -417,7 +417,7 @@ def main():
         help="The output directory where the model predictions and checkpoints will be written.",
     )
 
-    ## Other parameters
+    # Other parameters
     parser.add_argument(
         "--config_name", default="", type=str, help="Pretrained config name or path if not the same as model_name"
     )

templates/adding_a_new_example_script/run_xxx.py

@@ -401,7 +401,7 @@ def load_and_cache_examples(args, tokenizer, evaluate=False, output_examples=Fal
 def main():
     parser = argparse.ArgumentParser()
 
-    ## Required parameters
+    # Required parameters
     parser.add_argument(
         "--train_file", default=None, type=str, required=True, help="SQuAD json for training. E.g., train-v1.1.json"
     )
@@ -434,7 +434,7 @@ def main():
         help="The output directory where the model checkpoints and predictions will be written.",
     )
 
-    ## Other parameters
+    # Other parameters
     parser.add_argument(
         "--config_name", default="", type=str, help="Pretrained config name or path if not the same as model_name"
     )

templates/adding_a_new_model/convert_xxx_original_tf_checkpoint_to_pytorch.py

@@ -43,7 +43,7 @@ def convert_tf_checkpoint_to_pytorch(tf_checkpoint_path, config_file, pytorch_du
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
-    ## Required parameters
+    # Required parameters
     parser.add_argument(
         "--tf_checkpoint_path", default=None, type=str, required=True, help="Path to the TensorFlow checkpoint path."
     )

transformers/convert_albert_original_tf_checkpoint_to_pytorch.py

@@ -43,7 +43,7 @@ def convert_tf_checkpoint_to_pytorch(tf_checkpoint_path, albert_config_file, pyt
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
-    ## Required parameters
+    # Required parameters
     parser.add_argument(
         "--tf_checkpoint_path", default=None, type=str, required=True, help="Path to the TensorFlow checkpoint path."
     )

transformers/convert_bert_original_tf_checkpoint_to_pytorch.py

@@ -43,7 +43,7 @@ def convert_tf_checkpoint_to_pytorch(tf_checkpoint_path, bert_config_file, pytor
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
-    ## Required parameters
+    # Required parameters
     parser.add_argument(
         "--tf_checkpoint_path", default=None, type=str, required=True, help="Path to the TensorFlow checkpoint path."
     )

transformers/convert_gpt2_original_tf_checkpoint_to_pytorch.py

@@ -51,7 +51,7 @@ def convert_gpt2_checkpoint_to_pytorch(gpt2_checkpoint_path, gpt2_config_file, p
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
-    ## Required parameters
+    # Required parameters
     parser.add_argument(
         "--gpt2_checkpoint_path", default=None, type=str, required=True, help="Path to the TensorFlow checkpoint path."
     )

transformers/convert_openai_original_tf_checkpoint_to_pytorch.py

@@ -51,7 +51,7 @@ def convert_openai_checkpoint_to_pytorch(openai_checkpoint_folder_path, openai_c
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
-    ## Required parameters
+    # Required parameters
     parser.add_argument(
         "--openai_checkpoint_folder_path",
         default=None,

transformers/convert_pytorch_checkpoint_to_tf2.py

@@ -410,7 +410,7 @@ def convert_all_pt_checkpoints_to_tf(
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
-    ## Required parameters
+    # Required parameters
     parser.add_argument(
         "--tf_dump_path", default=None, type=str, required=True, help="Path to the output Tensorflow dump file."
     )

transformers/convert_roberta_original_pytorch_checkpoint_to_pytorch.py

@@ -94,7 +94,7 @@ def convert_roberta_checkpoint_to_pytorch(roberta_checkpoint_path, pytorch_dump_
         layer: BertLayer = model.roberta.encoder.layer[i]
         roberta_layer: TransformerSentenceEncoderLayer = roberta_sent_encoder.layers[i]
 
-        ### self attention
+        # self attention
         self_attn: BertSelfAttention = layer.attention.self
         assert (
             roberta_layer.self_attn.k_proj.weight.data.shape
@@ -110,7 +110,7 @@ def convert_roberta_checkpoint_to_pytorch(roberta_checkpoint_path, pytorch_dump_
         self_attn.value.weight.data = roberta_layer.self_attn.v_proj.weight
         self_attn.value.bias.data = roberta_layer.self_attn.v_proj.bias
 
-        ### self-attention output
+        # self-attention output
         self_output: BertSelfOutput = layer.attention.output
         assert self_output.dense.weight.shape == roberta_layer.self_attn.out_proj.weight.shape
         self_output.dense.weight = roberta_layer.self_attn.out_proj.weight
@@ -118,20 +118,20 @@ def convert_roberta_checkpoint_to_pytorch(roberta_checkpoint_path, pytorch_dump_
         self_output.LayerNorm.weight = roberta_layer.self_attn_layer_norm.weight
         self_output.LayerNorm.bias = roberta_layer.self_attn_layer_norm.bias
 
-        ### intermediate
+        # intermediate
         intermediate: BertIntermediate = layer.intermediate
         assert intermediate.dense.weight.shape == roberta_layer.fc1.weight.shape
         intermediate.dense.weight = roberta_layer.fc1.weight
         intermediate.dense.bias = roberta_layer.fc1.bias
 
-        ### output
+        # output
         bert_output: BertOutput = layer.output
         assert bert_output.dense.weight.shape == roberta_layer.fc2.weight.shape
         bert_output.dense.weight = roberta_layer.fc2.weight
         bert_output.dense.bias = roberta_layer.fc2.bias
         bert_output.LayerNorm.weight = roberta_layer.final_layer_norm.weight
         bert_output.LayerNorm.bias = roberta_layer.final_layer_norm.bias
-        #### end of layer
+        # end of layer
 
     if classification_head:
         model.classifier.dense.weight = roberta.model.classification_heads["mnli"].dense.weight
@@ -170,7 +170,7 @@ def convert_roberta_checkpoint_to_pytorch(roberta_checkpoint_path, pytorch_dump_
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
-    ## Required parameters
+    # Required parameters
     parser.add_argument(
         "--roberta_checkpoint_path", default=None, type=str, required=True, help="Path the official PyTorch dump."
     )

transformers/convert_t5_original_tf_checkpoint_to_pytorch.py

@@ -43,7 +43,7 @@ def convert_tf_checkpoint_to_pytorch(tf_checkpoint_path, config_file, pytorch_du
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
-    ## Required parameters
+    # Required parameters
     parser.add_argument(
         "--tf_checkpoint_path", default=None, type=str, required=True, help="Path to the TensorFlow checkpoint path."
     )