Delete lots of transformers.AdamW references

Author: Rocketknight1

examples/legacy/question-answering/run_squad.py

@@ -32,7 +32,6 @@
 from transformers import (
     MODEL_FOR_QUESTION_ANSWERING_MAPPING,
     WEIGHTS_NAME,
-    AdamW,
     AutoConfig,
     AutoModelForQuestionAnswering,
     AutoTokenizer,
@@ -96,7 +95,7 @@ def train(args, train_dataset, model, tokenizer):
         },
         {"params": [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)], "weight_decay": 0.0},
     ]
-    optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
+    optimizer = torch.optim.AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
     scheduler = get_linear_schedule_with_warmup(
         optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total
     )

examples/legacy/run_openai_gpt.py

@@ -43,7 +43,6 @@
 from transformers import (
     CONFIG_NAME,
     WEIGHTS_NAME,
-    AdamW,
     OpenAIGPTDoubleHeadsModel,
     OpenAIGPTTokenizer,
     get_linear_schedule_with_warmup,
@@ -236,7 +235,7 @@ def tokenize_and_encode(obj):
             },
             {"params": [p for n, p in param_optimizer if any(nd in n for nd in no_decay)], "weight_decay": 0.0},
         ]
-        optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
+        optimizer = torch.optim.AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
         scheduler = get_linear_schedule_with_warmup(
             optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total
         )

examples/legacy/run_swag.py

@@ -34,7 +34,6 @@
 import transformers
 from transformers import (
     WEIGHTS_NAME,
-    AdamW,
     AutoConfig,
     AutoModelForMultipleChoice,
     AutoTokenizer,
@@ -298,7 +297,7 @@ def train(args, train_dataset, model, tokenizer):
         },
         {"params": [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)], "weight_decay": 0.0},
     ]
-    optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
+    optimizer = torch.optim.AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
     scheduler = get_linear_schedule_with_warmup(
         optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total
     )

examples/pytorch/question-answering/run_qa_beam_search_no_trainer.py

@@ -41,7 +41,6 @@
 
 import transformers
 from transformers import (
-    AdamW,
     DataCollatorWithPadding,
     EvalPrediction,
     SchedulerType,
@@ -767,7 +766,7 @@ def create_and_fill_np_array(start_or_end_logits, dataset, max_len):
             "weight_decay": 0.0,
         },
     ]
-    optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate)
+    optimizer = torch.optim.AdamW(optimizer_grouped_parameters, lr=args.learning_rate)
 
     # Scheduler and math around the number of training steps.
     overrode_max_train_steps = False

examples/pytorch/speech-pretraining/run_wav2vec2_pretraining_no_trainer.py

@@ -33,7 +33,6 @@
 
 import transformers
 from transformers import (
-    AdamW,
     SchedulerType,
     Wav2Vec2Config,
     Wav2Vec2FeatureExtractor,
@@ -583,7 +582,7 @@ def prepare_dataset(batch):
     )
 
     # Optimizer
-    optimizer = AdamW(
+    optimizer = torch.optim.AdamW(
         list(model.parameters()),
         lr=args.learning_rate,
         betas=[args.adam_beta1, args.adam_beta2],

examples/research_projects/bert-loses-patience/run_glue_with_pabee.py

@@ -34,7 +34,6 @@
 import transformers
 from transformers import (
     WEIGHTS_NAME,
-    AdamW,
     AlbertConfig,
     AlbertTokenizer,
     BertConfig,
@@ -95,7 +94,7 @@ def train(args, train_dataset, model, tokenizer):
         {"params": [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)], "weight_decay": 0.0},
     ]
 
-    optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
+    optimizer = torch.optim.AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
     scheduler = get_linear_schedule_with_warmup(
         optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total
     )

examples/research_projects/deebert/run_glue_deebert.py

@@ -19,7 +19,6 @@
 from src.modeling_highway_roberta import DeeRobertaForSequenceClassification
 from transformers import (
     WEIGHTS_NAME,
-    AdamW,
     BertConfig,
     BertTokenizer,
     RobertaConfig,
@@ -123,7 +122,7 @@ def train(args, train_dataset, model, tokenizer, train_highway=False):
                 "weight_decay": 0.0,
             },
         ]
-    optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
+    optimizer = torch.optim.AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
     scheduler = get_linear_schedule_with_warmup(
         optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total
     )

examples/research_projects/distillation/run_squad_w_distillation.py

@@ -32,7 +32,6 @@
 import transformers
 from transformers import (
     WEIGHTS_NAME,
-    AdamW,
     BertConfig,
     BertForQuestionAnswering,
     BertTokenizer,
@@ -114,7 +113,7 @@ def train(args, train_dataset, model, tokenizer, teacher=None):
         },
         {"params": [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)], "weight_decay": 0.0},
     ]
-    optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
+    optimizer = torch.optim.AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
     scheduler = get_linear_schedule_with_warmup(
         optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total
     )

examples/research_projects/information-gain-filtration/igf/igf.py

@@ -12,7 +12,7 @@
 from torch.utils.data import DataLoader
 from tqdm import tqdm
 
-from transformers import AdamW, GPT2LMHeadModel, get_linear_schedule_with_warmup
+from transformers import GPT2LMHeadModel, get_linear_schedule_with_warmup
 
 
 logger = logging.getLogger(__name__)
@@ -112,7 +112,7 @@ def recopy_gpt2(orig_model, device, max_steps):
         },
         {"params": [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)], "weight_decay": 0.0},
     ]
-    lm_optimizer = AdamW(optimizer_grouped_parameters, lr=5e-5, eps=1e-8)
+    lm_optimizer = torch.optim.AdamW(optimizer_grouped_parameters, lr=5e-5, eps=1e-8)
     lm_scheduler = get_linear_schedule_with_warmup(lm_optimizer, 0, max_steps)
     torch.cuda.empty_cache()
     return model, lm_optimizer, lm_scheduler

examples/research_projects/longform-qa/eli5_utils.py

@@ -16,7 +16,7 @@
 from torch.utils.data import DataLoader, Dataset, RandomSampler, SequentialSampler
 from tqdm import tqdm
 
-from transformers import AdamW, AutoModel, AutoModelForSeq2SeqLM, AutoTokenizer, get_linear_schedule_with_warmup
+from transformers import AutoModel, AutoModelForSeq2SeqLM, AutoTokenizer, get_linear_schedule_with_warmup
 
 
 pd.set_option("display.max_colwidth", None)
@@ -316,7 +316,7 @@ def evaluate_qa_retriever(model, dataset, tokenizer, args):
 
 
 def train_qa_retriever(qar_model, qar_tokenizer, qar_train_dset, qar_valid_dset, qar_args):
-    qar_optimizer = AdamW(qar_model.parameters(), lr=qar_args.learning_rate, eps=1e-8)
+    qar_optimizer = torch.optim.AdamW(qar_model.parameters(), lr=qar_args.learning_rate, eps=1e-8)
     qar_scheduler = get_linear_schedule_with_warmup(
         qar_optimizer,
         num_warmup_steps=100,
@@ -493,7 +493,7 @@ def eval_qa_s2s_epoch(model, dataset, tokenizer, args):
 
 
 def train_qa_s2s(qa_s2s_model, qa_s2s_tokenizer, s2s_train_dset, s2s_valid_dset, s2s_args):
-    s2s_optimizer = AdamW(qa_s2s_model.parameters(), lr=s2s_args.learning_rate, eps=1e-8)
+    s2s_optimizer = torch.optim.AdamW(qa_s2s_model.parameters(), lr=s2s_args.learning_rate, eps=1e-8)
     s2s_scheduler = get_linear_schedule_with_warmup(
         s2s_optimizer,
         num_warmup_steps=400,

examples/research_projects/luke/run_luke_ner_no_trainer.py

@@ -36,7 +36,6 @@
 
 import transformers
 from transformers import (
-    AdamW,
     LukeConfig,
     LukeForEntitySpanClassification,
     LukeTokenizer,
@@ -569,7 +568,7 @@ def tokenize_and_align_labels(examples):
             "weight_decay": 0.0,
         },
     ]
-    optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate)
+    optimizer = torch.optim.AdamW(optimizer_grouped_parameters, lr=args.learning_rate)
 
     # Use the device given by the `accelerator` object.
     device = accelerator.device

examples/research_projects/mm-imdb/run_mmimdb.py

@@ -34,7 +34,6 @@
 import transformers
 from transformers import (
     WEIGHTS_NAME,
-    AdamW,
     AutoConfig,
     AutoModel,
     AutoTokenizer,
@@ -93,7 +92,7 @@ def train(args, train_dataset, model, tokenizer, criterion):
         {"params": [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)], "weight_decay": 0.0},
     ]
 
-    optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
+    optimizer = torch.optim.AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
     scheduler = get_linear_schedule_with_warmup(
         optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total
     )

examples/research_projects/movement-pruning/masked_run_glue.py

@@ -32,7 +32,6 @@
 
 from transformers import (
     WEIGHTS_NAME,
-    AdamW,
     BertConfig,
     BertForSequenceClassification,
     BertTokenizer,
@@ -145,7 +144,7 @@ def train(args, train_dataset, model, tokenizer, teacher=None):
         },
     ]
 
-    optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
+    optimizer = torch.optim.AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
     scheduler = get_linear_schedule_with_warmup(
         optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total
     )

examples/research_projects/movement-pruning/masked_run_squad.py

@@ -32,7 +32,6 @@
 
 from transformers import (
     WEIGHTS_NAME,
-    AdamW,
     BertConfig,
     BertForQuestionAnswering,
     BertTokenizer,
@@ -152,7 +151,7 @@ def train(args, train_dataset, model, tokenizer, teacher=None):
         },
     ]
 
-    optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
+    optimizer = torch.optim.AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
     scheduler = get_linear_schedule_with_warmup(
         optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total
     )

examples/research_projects/rag-end2end-retriever/lightning_base.py

@@ -5,10 +5,10 @@
 from typing import Any, Dict
 
 import pytorch_lightning as pl
+import torch
 from pytorch_lightning.utilities import rank_zero_info
 
 from transformers import (
-    AdamW,
     AutoConfig,
     AutoModel,
     AutoModelForPreTraining,
@@ -148,7 +148,7 @@ def configure_optimizers(self):
             )
 
         else:
-            optimizer = AdamW(
+            optimizer = torch.optim.AdamW(
                 optimizer_grouped_parameters, lr=self.hparams.learning_rate, eps=self.hparams.adam_epsilon
             )
         self.opt = optimizer

examples/research_projects/rag/lightning_base.py

@@ -5,10 +5,10 @@
 from typing import Any, Dict
 
 import pytorch_lightning as pl
+import torch
 from pytorch_lightning.utilities import rank_zero_info
 
 from transformers import (
-    AdamW,
     AutoConfig,
     AutoModel,
     AutoModelForPreTraining,
@@ -146,7 +146,7 @@ def configure_optimizers(self):
             )
 
         else:
-            optimizer = AdamW(
+            optimizer = torch.optim.AdamW(
                 optimizer_grouped_parameters, lr=self.hparams.learning_rate, eps=self.hparams.adam_epsilon
             )
         self.opt = optimizer

examples/research_projects/self-training-text-classification/finetuning.py

@@ -33,7 +33,6 @@
 from tqdm.auto import tqdm
 
 from transformers import (
-    AdamW,
     AutoConfig,
     AutoModelForSequenceClassification,
     AutoTokenizer,
@@ -749,7 +748,7 @@ def preprocess_function(examples):
             "weight_decay": 0.0,
         },
     ]
-    optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate)
+    optimizer = torch.optim.AdamW(optimizer_grouped_parameters, lr=args.learning_rate)
 
     # Prepare everything with our `accelerator`.
     model, optimizer, train_dataloader, eval_dataloader, test_dataloader, infer_dataloader = accelerator.prepare(

examples/research_projects/seq2seq-distillation/lightning_base.py

@@ -5,10 +5,10 @@
 from typing import Any, Dict
 
 import pytorch_lightning as pl
+import torch
 from pytorch_lightning.utilities import rank_zero_info
 
 from transformers import (
-    AdamW,
     AutoConfig,
     AutoModel,
     AutoModelForPreTraining,
@@ -146,7 +146,7 @@ def configure_optimizers(self):
             )
 
         else:
-            optimizer = AdamW(
+            optimizer = torch.optim.AdamW(
                 optimizer_grouped_parameters, lr=self.hparams.learning_rate, eps=self.hparams.adam_epsilon
             )
         self.opt = optimizer

templates/adding_a_new_example_script/{{cookiecutter.directory_name}}/run_{{cookiecutter.example_shortcut}}.py

@@ -535,7 +535,6 @@ def _mp_fn(index):
 from transformers import (
     CONFIG_MAPPING,
     MODEL_MAPPING,
-    AdamW,
     AutoConfig,
     {{cookiecutter.model_class}},
     AutoTokenizer,
@@ -863,7 +862,7 @@ def tokenize_function(examples):
             "weight_decay": 0.0,
         },
     ]
-    optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate)
+    optimizer = torch.optim.AdamW(optimizer_grouped_parameters, lr=args.learning_rate)
 
     # Prepare everything with our `accelerator`.
     model, optimizer, train_dataloader, eval_dataloader = accelerator.prepare(