Merge branch 'master' into fix-ctrl-past

examples/contrib/run_swag.py

@@ -42,7 +42,7 @@ from tqdm import tqdm, trange
 from transformers import (WEIGHTS_NAME, BertConfig,
                                   BertForMultipleChoice, BertTokenizer)
 
-from transformers import AdamW, WarmupLinearSchedule
+from transformers import AdamW, get_linear_schedule_with_warmup
 
 logger = logging.getLogger(__name__)
 
@@ -322,7 +322,7 @@ 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)
-    scheduler = WarmupLinearSchedule(optimizer, warmup_steps=args.warmup_steps, t_total=t_total)
+    scheduler = get_linear_schedule_with_warmup(optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total)
     if args.fp16:
         try:
             from apex import amp

examples/distillation/distiller.py

@@ -35,7 +35,7 @@ try:
 except:
     from tensorboardX import SummaryWriter
 
-from transformers import WarmupLinearSchedule
+from transformers import get_linear_schedule_with_warmup
 
 from utils import logger
 from lm_seqs_dataset import LmSeqsDataset
@@ -137,9 +137,9 @@ class Distiller:
                                betas=(0.9, 0.98))
 
         warmup_steps = math.ceil(num_train_optimization_steps * params.warmup_prop)
-        self.scheduler = WarmupLinearSchedule(self.optimizer,
-                                                warmup_steps=warmup_steps,
-                                                t_total=num_train_optimization_steps)
+        self.scheduler = get_linear_schedule_with_warmup(self.optimizer,
+                                                num_warmup_steps=warmup_steps,
+                                                num_training_steps=num_train_optimization_steps)
 
         if self.fp16:
             try:

examples/distillation/run_squad_w_distillation.py

@@ -46,7 +46,7 @@ from transformers import (WEIGHTS_NAME, BertConfig,
                                   XLNetTokenizer,
                                   DistilBertConfig, DistilBertForQuestionAnswering, DistilBertTokenizer)
 
-from transformers import AdamW, WarmupLinearSchedule
+from transformers import AdamW, get_linear_schedule_with_warmup
 
 from ..utils_squad import (read_squad_examples, convert_examples_to_features,
                          RawResult, write_predictions,
@@ -101,7 +101,7 @@ 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)
-    scheduler = WarmupLinearSchedule(optimizer, warmup_steps=args.warmup_steps, t_total=t_total)
+    scheduler = get_linear_schedule_with_warmup(optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total)
     if args.fp16:
         try:
             from apex import amp

examples/run_bertology.py

@@ -39,8 +39,9 @@ from transformers import (WEIGHTS_NAME,
 
 from run_glue import set_seed, load_and_cache_examples, ALL_MODELS, MODEL_CLASSES
 
-from utils_glue import (compute_metrics, convert_examples_to_features,
-                        output_modes, processors)
+from transformers import glue_compute_metrics as compute_metrics
+from transformers import glue_output_modes as output_modes
+from transformers import glue_processors as processors
 
 logger = logging.getLogger(__name__)
 
@@ -233,6 +234,8 @@ def main():
                         help="If > 0: limit the data to a subset of data_subset instances.")
     parser.add_argument("--overwrite_output_dir", action='store_true',
                         help="Whether to overwrite data in output directory")
+    parser.add_argument('--overwrite_cache', action='store_true',
+                        help="Overwrite the cached training and evaluation sets")
 
     parser.add_argument("--dont_normalize_importance_by_layer", action='store_true',
                         help="Don't normalize importance score by layers")

examples/run_glue.py

@@ -47,9 +47,13 @@ from transformers import (WEIGHTS_NAME, BertConfig,
                                   XLNetTokenizer,
                                   DistilBertConfig,
                                   DistilBertForSequenceClassification,
-                                  DistilBertTokenizer)
+                                  DistilBertTokenizer,
+                                  AlbertConfig,
+                                  AlbertForSequenceClassification, 
+                                  AlbertTokenizer,
+                                )
 
-from transformers import AdamW, WarmupLinearSchedule
+from transformers import AdamW, get_linear_schedule_with_warmup
 
 from transformers import glue_compute_metrics as compute_metrics
 from transformers import glue_output_modes as output_modes
@@ -66,7 +70,8 @@ MODEL_CLASSES = {
     'xlnet': (XLNetConfig, XLNetForSequenceClassification, XLNetTokenizer),
     'xlm': (XLMConfig, XLMForSequenceClassification, XLMTokenizer),
     'roberta': (RobertaConfig, RobertaForSequenceClassification, RobertaTokenizer),
-    'distilbert': (DistilBertConfig, DistilBertForSequenceClassification, DistilBertTokenizer)
+    'distilbert': (DistilBertConfig, DistilBertForSequenceClassification, DistilBertTokenizer),
+    'albert': (AlbertConfig, AlbertForSequenceClassification, AlbertTokenizer)
 }
 
 
@@ -99,8 +104,9 @@ def train(args, train_dataset, model, tokenizer):
         {'params': [p for n, p in model.named_parameters() if not any(nd in n for nd in no_decay)], 'weight_decay': args.weight_decay},
         {'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)
-    scheduler = WarmupLinearSchedule(optimizer, warmup_steps=args.warmup_steps, t_total=t_total)
+    scheduler = get_linear_schedule_with_warmup(optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total)
     if args.fp16:
         try:
             from apex import amp
@@ -158,7 +164,7 @@ def train(args, train_dataset, model, tokenizer):
                 loss.backward()
 
             tr_loss += loss.item()
-            if (step + 1) % args.gradient_accumulation_steps == 0 and not args.tpu:
+            if (step + 1) % args.gradient_accumulation_steps == 0:
                 if args.fp16:
                     torch.nn.utils.clip_grad_norm_(amp.master_params(optimizer), args.max_grad_norm)
                 else:
@@ -189,11 +195,6 @@ def train(args, train_dataset, model, tokenizer):
                     torch.save(args, os.path.join(output_dir, 'training_args.bin'))
                     logger.info("Saving model checkpoint to %s", output_dir)
 
-            if args.tpu:
-                args.xla_model.optimizer_step(optimizer, barrier=True)
-                model.zero_grad()
-                global_step += 1
-
             if args.max_steps > 0 and global_step > args.max_steps:
                 epoch_iterator.close()
                 break
@@ -224,6 +225,10 @@ def evaluate(args, model, tokenizer, prefix=""):
         eval_sampler = SequentialSampler(eval_dataset) if args.local_rank == -1 else DistributedSampler(eval_dataset)
         eval_dataloader = DataLoader(eval_dataset, sampler=eval_sampler, batch_size=args.eval_batch_size)
 
+        # multi-gpu eval
+        if args.n_gpu > 1:
+            model = torch.nn.DataParallel(model)
+
         # Eval!
         logger.info("***** Running evaluation {} *****".format(prefix))
         logger.info("  Num examples = %d", len(eval_dataset))
@@ -318,7 +323,7 @@ def load_and_cache_examples(args, task, tokenizer, evaluate=False):
         all_labels = torch.tensor([f.label for f in features], dtype=torch.long)
     elif output_mode == "regression":
         all_labels = torch.tensor([f.label for f in features], dtype=torch.float)
-
+ 
     dataset = TensorDataset(all_input_ids, all_attention_mask, all_token_type_ids, all_labels)
     return dataset
 
@@ -362,7 +367,7 @@ def main():
     parser.add_argument("--per_gpu_eval_batch_size", default=8, type=int,
                         help="Batch size per GPU/CPU for evaluation.")
     parser.add_argument('--gradient_accumulation_steps', type=int, default=1,
-                        help="Number of updates steps to accumulate before performing a backward/update pass.")
+                        help="Number of updates steps to accumulate before performing a backward/update pass.")     
     parser.add_argument("--learning_rate", default=5e-5, type=float,
                         help="The initial learning rate for Adam.")
     parser.add_argument("--weight_decay", default=0.0, type=float,
@@ -393,15 +398,6 @@ def main():
     parser.add_argument('--seed', type=int, default=42,
                         help="random seed for initialization")
 
-    parser.add_argument('--tpu', action='store_true',
-                        help="Whether to run on the TPU defined in the environment variables")
-    parser.add_argument('--tpu_ip_address', type=str, default='',
-                        help="TPU IP address if none are set in the environment variables")
-    parser.add_argument('--tpu_name', type=str, default='',
-                        help="TPU name if none are set in the environment variables")
-    parser.add_argument('--xrt_tpu_config', type=str, default='',
-                        help="XRT TPU config if none are set in the environment variables")
-
     parser.add_argument('--fp16', action='store_true',
                         help="Whether to use 16-bit (mixed) precision (through NVIDIA apex) instead of 32-bit")
     parser.add_argument('--fp16_opt_level', type=str, default='O1',
@@ -435,23 +431,6 @@ def main():
         args.n_gpu = 1
     args.device = device
 
-    if args.tpu:
-        if args.tpu_ip_address:
-            os.environ["TPU_IP_ADDRESS"] = args.tpu_ip_address
-        if args.tpu_name:
-            os.environ["TPU_NAME"] = args.tpu_name
-        if args.xrt_tpu_config:
-            os.environ["XRT_TPU_CONFIG"] = args.xrt_tpu_config
-
-        assert "TPU_IP_ADDRESS" in os.environ
-        assert "TPU_NAME" in os.environ
-        assert "XRT_TPU_CONFIG" in os.environ
-
-        import torch_xla
-        import torch_xla.core.xla_model as xm
-        args.device = xm.xla_device()
-        args.xla_model = xm
-
     # Setup logging
     logging.basicConfig(format = '%(asctime)s - %(levelname)s - %(name)s -   %(message)s',
                         datefmt = '%m/%d/%Y %H:%M:%S',
@@ -505,7 +484,7 @@ def main():
 
 
     # Saving best-practices: if you use defaults names for the model, you can reload it using from_pretrained()
-    if args.do_train and (args.local_rank == -1 or torch.distributed.get_rank() == 0) and not args.tpu:
+    if args.do_train and (args.local_rank == -1 or torch.distributed.get_rank() == 0):
         # Create output directory if needed
         if not os.path.exists(args.output_dir) and args.local_rank in [-1, 0]:
             os.makedirs(args.output_dir)

examples/run_lm_finetuning.py

@@ -42,7 +42,7 @@ except:
 
 from tqdm import tqdm, trange
 
-from transformers import (WEIGHTS_NAME, AdamW, WarmupLinearSchedule,
+from transformers import (WEIGHTS_NAME, AdamW, get_linear_schedule_with_warmup,
                                   BertConfig, BertForMaskedLM, BertTokenizer,
                                   GPT2Config, GPT2LMHeadModel, GPT2Tokenizer,
                                   OpenAIGPTConfig, OpenAIGPTLMHeadModel, OpenAIGPTTokenizer,
@@ -63,12 +63,12 @@ MODEL_CLASSES = {
 
 
 class TextDataset(Dataset):
-    def __init__(self, tokenizer, file_path='train', block_size=512):
+    def __init__(self, tokenizer, args, file_path='train', block_size=512):
         assert os.path.isfile(file_path)
         directory, filename = os.path.split(file_path)
-        cached_features_file = os.path.join(directory, 'cached_lm_' + str(block_size) + '_' + filename)
+        cached_features_file = os.path.join(directory, args.model_name_or_path + '_cached_lm_' + str(block_size) + '_' + filename)
 
-        if os.path.exists(cached_features_file):
+        if os.path.exists(cached_features_file) and not args.overwrite_cache:
             logger.info("Loading features from cached file %s", cached_features_file)
             with open(cached_features_file, 'rb') as handle:
                 self.examples = pickle.load(handle)
@@ -99,7 +99,7 @@ class TextDataset(Dataset):
 
 
 def load_and_cache_examples(args, tokenizer, evaluate=False):
-    dataset = TextDataset(tokenizer, file_path=args.eval_data_file if evaluate else args.train_data_file, block_size=args.block_size)
+    dataset = TextDataset(tokenizer, args, file_path=args.eval_data_file if evaluate else args.train_data_file, block_size=args.block_size)
     return dataset
 
 
@@ -185,7 +185,7 @@ 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)
-    scheduler = WarmupLinearSchedule(optimizer, warmup_steps=args.warmup_steps, t_total=t_total)
+    scheduler = get_linear_schedule_with_warmup(optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total)
     if args.fp16:
         try:
             from apex import amp
@@ -215,6 +215,7 @@ def train(args, train_dataset, model, tokenizer):
 
     global_step = 0
     tr_loss, logging_loss = 0.0, 0.0
+    model.resize_token_embeddings(len(tokenizer))
     model.zero_grad()
     train_iterator = trange(int(args.num_train_epochs), desc="Epoch", disable=args.local_rank not in [-1, 0])
     set_seed(args)  # Added here for reproducibility (even between python 2 and 3)
@@ -300,6 +301,10 @@ def evaluate(args, model, tokenizer, prefix=""):
     eval_sampler = SequentialSampler(eval_dataset) if args.local_rank == -1 else DistributedSampler(eval_dataset)
     eval_dataloader = DataLoader(eval_dataset, sampler=eval_sampler, batch_size=args.eval_batch_size)
 
+    # multi-gpu evaluate
+    if args.n_gpu > 1:
+        model = torch.nn.DataParallel(model)
+
     # Eval!
     logger.info("***** Running evaluation {} *****".format(prefix))
     logger.info("  Num examples = %d", len(eval_dataset))

examples/run_multiple_choice.py

@@ -43,7 +43,7 @@ from transformers import (WEIGHTS_NAME, BertConfig,
                                   XLNetTokenizer, RobertaConfig,
                                   RobertaForMultipleChoice, RobertaTokenizer)
 
-from transformers import AdamW, WarmupLinearSchedule
+from transformers import AdamW, get_linear_schedule_with_warmup
 
 from utils_multiple_choice import (convert_examples_to_features, processors)
 
@@ -101,7 +101,7 @@ 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)
-    scheduler = WarmupLinearSchedule(optimizer, warmup_steps=args.warmup_steps, t_total=t_total)
+    scheduler = get_linear_schedule_with_warmup(optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total)
     if args.fp16:
         try:
             from apex import amp
@@ -229,6 +229,10 @@ def evaluate(args, model, tokenizer, prefix="", test=False):
         eval_sampler = SequentialSampler(eval_dataset) if args.local_rank == -1 else DistributedSampler(eval_dataset)
         eval_dataloader = DataLoader(eval_dataset, sampler=eval_sampler, batch_size=args.eval_batch_size)
 
+        # multi-gpu evaluate
+        if args.n_gpu > 1:
+            model = torch.nn.DataParallel(model)
+
         # Eval!
         logger.info("***** Running evaluation {} *****".format(prefix))
         logger.info("  Num examples = %d", len(eval_dataset))

examples/run_ner.py

@@ -33,19 +33,23 @@ from torch.utils.data.distributed import DistributedSampler
 from tqdm import tqdm, trange
 from utils_ner import convert_examples_to_features, get_labels, read_examples_from_file
 
-from transformers import AdamW, WarmupLinearSchedule
+from transformers import AdamW, get_linear_schedule_with_warmup
 from transformers import WEIGHTS_NAME, BertConfig, BertForTokenClassification, BertTokenizer
 from transformers import RobertaConfig, RobertaForTokenClassification, RobertaTokenizer
+from transformers import DistilBertConfig, DistilBertForTokenClassification, DistilBertTokenizer
+from transformers import CamembertConfig, CamembertForTokenClassification, CamembertTokenizer
 
 logger = logging.getLogger(__name__)
 
 ALL_MODELS = sum(
-    (tuple(conf.pretrained_config_archive_map.keys()) for conf in (BertConfig, RobertaConfig)),
+    (tuple(conf.pretrained_config_archive_map.keys()) for conf in (BertConfig, RobertaConfig, DistilBertConfig)),
     ())
 
 MODEL_CLASSES = {
     "bert": (BertConfig, BertForTokenClassification, BertTokenizer),
-    "roberta": (RobertaConfig, RobertaForTokenClassification, RobertaTokenizer)
+    "roberta": (RobertaConfig, RobertaForTokenClassification, RobertaTokenizer),
+    "distilbert": (DistilBertConfig, DistilBertForTokenClassification, DistilBertTokenizer),
+    "camembert": (CamembertConfig, CamembertForTokenClassification, CamembertTokenizer),
 }
 
 
@@ -80,7 +84,7 @@ def train(args, train_dataset, model, tokenizer, labels, pad_token_label_id):
         {"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)
-    scheduler = WarmupLinearSchedule(optimizer, warmup_steps=args.warmup_steps, t_total=t_total)
+    scheduler = get_linear_schedule_with_warmup(optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total)
     if args.fp16:
         try:
             from apex import amp
@@ -121,9 +125,10 @@ def train(args, train_dataset, model, tokenizer, labels, pad_token_label_id):
             batch = tuple(t.to(args.device) for t in batch)
             inputs = {"input_ids": batch[0],
                       "attention_mask": batch[1],
-                      "token_type_ids": batch[2] if args.model_type in ["bert", "xlnet"] else None,
-                      # XLM and RoBERTa don"t use segment_ids
                       "labels": batch[3]}
+            if args.model_type != "distilbert":
+                inputs["token_type_ids"] = batch[2] if args.model_type in ["bert", "xlnet"] else None  # XLM and RoBERTa don"t use segment_ids
+
             outputs = model(**inputs)
             loss = outputs[0]  # model outputs are always tuple in pytorch-transformers (see doc)
 
@@ -191,6 +196,10 @@ def evaluate(args, model, tokenizer, labels, pad_token_label_id, mode, prefix=""
     eval_sampler = SequentialSampler(eval_dataset) if args.local_rank == -1 else DistributedSampler(eval_dataset)
     eval_dataloader = DataLoader(eval_dataset, sampler=eval_sampler, batch_size=args.eval_batch_size)
 
+    # multi-gpu evaluate
+    if args.n_gpu > 1:
+        model = torch.nn.DataParallel(model)
+
     # Eval!
     logger.info("***** Running evaluation %s *****", prefix)
     logger.info("  Num examples = %d", len(eval_dataset))
@@ -206,9 +215,9 @@ def evaluate(args, model, tokenizer, labels, pad_token_label_id, mode, prefix=""
         with torch.no_grad():
             inputs = {"input_ids": batch[0],
                       "attention_mask": batch[1],
-                      "token_type_ids": batch[2] if args.model_type in ["bert", "xlnet"] else None,
-                      # XLM and RoBERTa don"t use segment_ids
                       "labels": batch[3]}
+            if args.model_type != "distilbert":
+                inputs["token_type_ids"] = batch[2] if args.model_type in ["bert", "xlnet"] else None  # XLM and RoBERTa don"t use segment_ids
             outputs = model(**inputs)
             tmp_eval_loss, logits = outputs[:2]
 
@@ -520,3 +529,4 @@ def main():
 
 if __name__ == "__main__":
     main()
+

examples/run_squad.py

@@ -43,9 +43,10 @@ from transformers import (WEIGHTS_NAME, BertConfig,
                                   XLMTokenizer, XLNetConfig,
                                   XLNetForQuestionAnswering,
                                   XLNetTokenizer,
-                                  DistilBertConfig, DistilBertForQuestionAnswering, DistilBertTokenizer)
+                                  DistilBertConfig, DistilBertForQuestionAnswering, DistilBertTokenizer,
+                                  AlbertConfig, AlbertForQuestionAnswering, AlbertTokenizer)
 
-from transformers import AdamW, WarmupLinearSchedule
+from transformers import AdamW, get_linear_schedule_with_warmup
 
 from utils_squad import (read_squad_examples, convert_examples_to_features,
                          RawResult, write_predictions,
@@ -65,7 +66,8 @@ MODEL_CLASSES = {
     'bert': (BertConfig, BertForQuestionAnswering, BertTokenizer),
     'xlnet': (XLNetConfig, XLNetForQuestionAnswering, XLNetTokenizer),
     'xlm': (XLMConfig, XLMForQuestionAnswering, XLMTokenizer),
-    'distilbert': (DistilBertConfig, DistilBertForQuestionAnswering, DistilBertTokenizer)
+    'distilbert': (DistilBertConfig, DistilBertForQuestionAnswering, DistilBertTokenizer),
+    'albert': (AlbertConfig, AlbertForQuestionAnswering, AlbertTokenizer)
 }
 
 def set_seed(args):
@@ -100,7 +102,7 @@ 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)
-    scheduler = WarmupLinearSchedule(optimizer, warmup_steps=args.warmup_steps, t_total=t_total)
+    scheduler = get_linear_schedule_with_warmup(optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total)
     if args.fp16:
         try:
             from apex import amp
@@ -128,7 +130,7 @@ def train(args, train_dataset, model, tokenizer):
     logger.info("  Gradient Accumulation steps = %d", args.gradient_accumulation_steps)
     logger.info("  Total optimization steps = %d", t_total)
 
-    global_step = 0
+    global_step = 1
     tr_loss, logging_loss = 0.0, 0.0
     model.zero_grad()
     train_iterator = trange(int(args.num_train_epochs), desc="Epoch", disable=args.local_rank not in [-1, 0])
@@ -217,6 +219,10 @@ def evaluate(args, model, tokenizer, prefix=""):
     eval_sampler = SequentialSampler(dataset) if args.local_rank == -1 else DistributedSampler(dataset)
     eval_dataloader = DataLoader(dataset, sampler=eval_sampler, batch_size=args.eval_batch_size)
 
+    # multi-gpu evaluate
+    if args.n_gpu > 1:
+        model = torch.nn.DataParallel(model)
+
     # Eval!
     logger.info("***** Running evaluation {} *****".format(prefix))
     logger.info("  Num examples = %d", len(dataset))
@@ -398,7 +404,7 @@ def main():
     parser.add_argument('--gradient_accumulation_steps', type=int, default=1,
                         help="Number of updates steps to accumulate before performing a backward/update pass.")
     parser.add_argument("--weight_decay", default=0.0, type=float,
-                        help="Weight deay if we apply some.")
+                        help="Weight decay if we apply some.")
     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,
@@ -533,7 +539,7 @@ def main():
         torch.save(args, os.path.join(args.output_dir, 'training_args.bin'))
 
         # Load a trained model and vocabulary that you have fine-tuned
-        model = model_class.from_pretrained(args.output_dir)
+        model = model_class.from_pretrained(args.output_dir, force_download=True)
         tokenizer = tokenizer_class.from_pretrained(args.output_dir, do_lower_case=args.do_lower_case)
         model.to(args.device)
 
@@ -551,7 +557,7 @@ def main():
         for checkpoint in checkpoints:
             # Reload the model
             global_step = checkpoint.split('-')[-1] if len(checkpoints) > 1 else ""
-            model = model_class.from_pretrained(checkpoint)
+            model = model_class.from_pretrained(checkpoint, force_download=True)
             model.to(args.device)
 
             # Evaluate

examples/run_summarization_finetuning.py

@@ -275,6 +275,10 @@ def evaluate(args, model, tokenizer, prefix=""):
         eval_dataset, sampler=eval_sampler, batch_size=args.eval_batch_size
     )
 
+    # multi-gpu evaluate
+    if args.n_gpu > 1:
+        model = torch.nn.DataParallel(model)
+        
     logger.info("***** Running evaluation {} *****".format(prefix))
     logger.info("  Num examples = %d", len(eval_dataset))
     logger.info("  Batch size = %d", args.eval_batch_size)

examples/run_tf_glue.py

@@ -73,6 +73,8 @@ model.save_pretrained('./save/')
 
 if TASK == "mrpc":
     # Load the TensorFlow model in PyTorch for inspection
+    # This is to demo the interoperability between the two frameworks, you don't have to 
+    # do this in real life (you can run the inference on the TF model).
     pytorch_model = BertForSequenceClassification.from_pretrained('./save/', from_tf=True)
 
     # Quickly test a few predictions - MRPC is a paraphrasing task, let's see if our model learned the task

examples/utils_squad.py

@@ -240,6 +240,7 @@ def convert_examples_to_features(examples, tokenizer, max_seq_length,
 
         # The -3 accounts for [CLS], [SEP] and [SEP]
         max_tokens_for_doc = max_seq_length - len(query_tokens) - 3
+        assert max_tokens_for_doc > 0
 
         # We can have documents that are longer than the maximum sequence length.
         # To deal with this we do a sliding window approach, where we take chunks

setup.py

@@ -38,7 +38,7 @@ from setuptools import find_packages, setup
 
 setup(
     name="transformers",
-    version="2.1.1",
+    version="2.2.0",
     author="Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Google AI Language Team Authors, Open AI team Authors, Facebook AI Authors, Carnegie Mellon University Authors",
     author_email="thomas@huggingface.co",
     description="State-of-the-art Natural Language Processing for TensorFlow 2.0 and PyTorch",

templates/adding_a_new_example_script/run_xxx.py

@@ -43,7 +43,7 @@ from transformers import (WEIGHTS_NAME, BertConfig,
                                   XLNetTokenizer,
                                   DistilBertConfig, DistilBertForQuestionAnswering, DistilBertTokenizer)
 
-from transformers import AdamW, WarmupLinearSchedule
+from transformers import AdamW, get_linear_schedule_with_warmup
 
 from utils_squad import (read_squad_examples, convert_examples_to_features,
                          RawResult, write_predictions,
@@ -98,7 +98,7 @@ 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)
-    scheduler = WarmupLinearSchedule(optimizer, warmup_steps=args.warmup_steps, t_total=t_total)
+    scheduler = get_linear_schedule_with_warmup(optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total)
     if args.fp16:
         try:
             from apex import amp

templates/adding_a_new_model/README.md

@@ -7,7 +7,7 @@ The library is designed to incorporate a variety of models and code bases. As su
 One important point though is that the library has the following goals impacting the way models are incorporated:
 
 - one specific feature of the API is the capability to run the model and tokenizer inline. The tokenization code thus often have to be slightly adapted to allow for running in the python interpreter.
-- the package is also designed to be as self-consistent and with a small and reliable set of packages dependencies. In consequence, additional dependencies are usually not allowed when adding a model but can be allowed for the inclusion of a new tokenizer (recent examples of dependencies added for tokenizer specificites includes `sentencepiece` and `sacremoses`). Please make sure to check the existing dependencies when possible before adding a new one.
+- the package is also designed to be as self-consistent and with a small and reliable set of packages dependencies. In consequence, additional dependencies are usually not allowed when adding a model but can be allowed for the inclusion of a new tokenizer (recent examples of dependencies added for tokenizer specificities include `sentencepiece` and `sacremoses`). Please make sure to check the existing dependencies when possible before adding a new one.
 
 For a quick overview of the library organization, please check the [QuickStart section of the documentation](https://huggingface.co/transformers/quickstart.html).
 
@@ -20,7 +20,7 @@ Here an overview of the general workflow:
 - [ ] add tests
 - [ ] finalize
 
-Let's details what should be done at each step
+Let's detail what should be done at each step
 
 ## Adding model/configuration/tokenization classes
 
@@ -28,16 +28,16 @@ Here is the workflow for adding model/configuration/tokenization classes:
 
 - [ ] copy the python files from the present folder to the main folder and rename them, replacing `xxx` with your model name,
 - [ ] edit the files to replace `XXX` (with various casing) with your model name
-- [ ] copy-past or create a simple configuration class for your model in the `configuration_...` file
-- [ ] copy-past or create the code for your model in the `modeling_...` files (PyTorch and TF 2.0)
-- [ ] copy-past or create a tokenizer class for your model in the `tokenization_...` file
+- [ ] copy-paste or create a simple configuration class for your model in the `configuration_...` file
+- [ ] copy-paste or create the code for your model in the `modeling_...` files (PyTorch and TF 2.0)
+- [ ] copy-paste or create a tokenizer class for your model in the `tokenization_...` file
 
 # Adding conversion scripts
 
 Here is the workflow for the conversion scripts:
 
 - [ ] copy the conversion script (`convert_...`) from the present folder to the main folder.
-- [ ] edit this scipt to convert your original checkpoint weights to the current pytorch ones.
+- [ ] edit this script to convert your original checkpoint weights to the current pytorch ones.
 
 # Adding tests:
 
@@ -58,5 +58,5 @@ You can then finish the addition step by adding imports for your classes in the
 - [ ] add your models and tokenizer to `pipeline.py`
 - [ ] add a link to your conversion script in the main conversion utility (currently in `__main__` but will be moved to the `commands` subfolder in the near future)
 - [ ] edit the PyTorch to TF 2.0 conversion script to add your model in the `convert_pytorch_checkpoint_to_tf2.py` file
-- [ ] add a mention of your model in the doc: `README.md` and the documentation it-self at `docs/source/pretrained_models.rst`.
+- [ ] add a mention of your model in the doc: `README.md` and the documentation itself at `docs/source/pretrained_models.rst`.
 - [ ] upload the pretrained weigths, configurations and vocabulary files.

templates/adding_a_new_model/modeling_tf_xxx.py

@@ -49,7 +49,7 @@ TF_XXX_PRETRAINED_MODEL_ARCHIVE_MAP = {
 ####################################################
 # TF 2.0 Models are constructed using Keras imperative API by sub-classing
 # - tf.keras.layers.Layer for the layers and
-# - TFPreTrainedModel for the models (it-self a sub-class of tf.keras.Model)
+# - TFPreTrainedModel for the models (itself a sub-class of tf.keras.Model)
 ####################################################
 
 ####################################################
@@ -255,6 +255,10 @@ XXX_INPUTS_DOCSTRING = r"""
             Mask to nullify selected heads of the self-attention modules.
             Mask values selected in ``[0, 1]``:
             ``1`` indicates the head is **not masked**, ``0`` indicates the head is **masked**.
+        **inputs_embeds**: (`optional`) ``Numpy array`` or ``tf.Tensor`` of shape ``(batch_size, sequence_length, embedding_dim)``:
+            Optionally, instead of passing ``input_ids`` you can choose to directly pass an embedded representation.
+            This is useful if you want more control over how to convert `input_ids` indices into associated vectors
+            than the model's internal embedding lookup matrix.
 """
 
 @add_start_docstrings("The bare Xxx Model transformer outputing raw hidden-states without any specific head on top.",

templates/adding_a_new_model/modeling_xxx.py

@@ -120,7 +120,7 @@ def load_tf_weights_in_xxx(model, config, tf_checkpoint_path):
 ####################################################
 # PyTorch Models are constructed by sub-classing
 # - torch.nn.Module for the layers and
-# - PreTrainedModel for the models (it-self a sub-class of torch.nn.Module)
+# - PreTrainedModel for the models (itself a sub-class of torch.nn.Module)
 ####################################################
 
 ####################################################
@@ -238,6 +238,10 @@ XXX_INPUTS_DOCSTRING = r"""
             Mask to nullify selected heads of the self-attention modules.
             Mask values selected in ``[0, 1]``:
             ``1`` indicates the head is **not masked**, ``0`` indicates the head is **masked**.
+        **inputs_embeds**: (`optional`) ``torch.FloatTensor`` of shape ``(batch_size, sequence_length, embedding_dim)``:
+            Optionally, instead of passing ``input_ids`` you can choose to directly pass an embedded representation.
+            This is useful if you want more control over how to convert `input_ids` indices into associated vectors
+            than the model's internal embedding lookup matrix.
 """
 
 @add_start_docstrings("The bare Xxx Model transformer outputting raw hidden-states without any specific head on top.",
@@ -280,7 +284,6 @@ class XxxModel(XxxPreTrainedModel):
 
         self.init_weights()
 
-    @property
     def get_input_embeddings(self):
         return self.embeddings.word_embeddings
 
@@ -295,11 +298,22 @@ class XxxModel(XxxPreTrainedModel):
         for layer, heads in heads_to_prune.items():
             self.encoder.layer[layer].attention.prune_heads(heads)
 
-    def forward(self, input_ids, attention_mask=None, token_type_ids=None, position_ids=None, head_mask=None):
+    def forward(self, input_ids=None, attention_mask=None, token_type_ids=None, position_ids=None, head_mask=None, inputs_embeds=None):
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
+        elif input_ids is not None:
+            input_shape = input_ids.size()
+        elif inputs_embeds is not None:
+            input_shape = inputs_embeds.size()[:-1]
+        else:
+            raise ValueError("You have to specify either input_ids or inputs_embeds")
+
+        device = input_ids.device if input_ids is not None else inputs_embeds.device
+
         if attention_mask is None:
-            attention_mask = torch.ones_like(input_ids)
+            attention_mask = torch.ones(input_shape, device=device)
         if token_type_ids is None:
-            token_type_ids = torch.zeros_like(input_ids)
+            token_type_ids = torch.zeros(input_shape, dtype=torch.long, device=device)
 
         # We create a 3D attention mask from a 2D tensor mask.
         # Sizes are [batch_size, 1, 1, to_seq_length]
@@ -333,7 +347,7 @@ class XxxModel(XxxPreTrainedModel):
 
         ##################################
         # Replace this with your model code
-        embedding_output = self.embeddings(input_ids, position_ids=position_ids, token_type_ids=token_type_ids)
+        embedding_output = self.embeddings(input_ids=input_ids, position_ids=position_ids, token_type_ids=token_type_ids, inputs_embeds=inputs_embeds)
         encoder_outputs = self.encoder(embedding_output, extended_attention_mask, head_mask=head_mask)
         sequence_output = encoder_outputs[0]
         outputs = (sequence_output,) + encoder_outputs[1:]  # add hidden_states and attentions if they are here
@@ -384,14 +398,15 @@ class XxxForMaskedLM(XxxPreTrainedModel):
     def get_output_embeddings(self):
         return self.lm_head
 
-    def forward(self, input_ids, attention_mask=None, token_type_ids=None, position_ids=None, head_mask=None,
+    def forward(self, input_ids=None, attention_mask=None, token_type_ids=None, position_ids=None, head_mask=None, inputs_embeds=None,
                 masked_lm_labels=None):
 
         outputs = self.transformer(input_ids,
                             attention_mask=attention_mask,
                             token_type_ids=token_type_ids,
                             position_ids=position_ids, 
-                            head_mask=head_mask)
+                            head_mask=head_mask,
+                            inputs_embeds=inputs_embeds)
 
         sequence_output = outputs[0]
         prediction_scores = self.cls(sequence_output)
@@ -449,14 +464,15 @@ class XxxForSequenceClassification(XxxPreTrainedModel):
 
         self.init_weights()
 
-    def forward(self, input_ids, attention_mask=None, token_type_ids=None,
-                position_ids=None, head_mask=None, labels=None):
+    def forward(self, input_ids=None, attention_mask=None, token_type_ids=None,
+                position_ids=None, head_mask=None, inputs_embeds=None, labels=None):
 
         outputs = self.transformer(input_ids,
                             attention_mask=attention_mask,
                             token_type_ids=token_type_ids,
                             position_ids=position_ids, 
-                            head_mask=head_mask)
+                            head_mask=head_mask,
+                            inputs_embeds=inputs_embeds)
 
         pooled_output = outputs[1]
 
@@ -520,14 +536,15 @@ class XxxForTokenClassification(XxxPreTrainedModel):
 
         self.init_weights()
 
-    def forward(self, input_ids, attention_mask=None, token_type_ids=None,
-                position_ids=None, head_mask=None, labels=None):
+    def forward(self, input_ids=None, attention_mask=None, token_type_ids=None,
+                position_ids=None, head_mask=None, inputs_embeds=None, labels=None):
 
         outputs = self.transformer(input_ids,
                             attention_mask=attention_mask,
                             token_type_ids=token_type_ids,
                             position_ids=position_ids, 
-                            head_mask=head_mask)
+                            head_mask=head_mask,
+                            inputs_embeds=inputs_embeds)
 
         sequence_output = outputs[0]
 
@@ -603,14 +620,15 @@ class XxxForQuestionAnswering(XxxPreTrainedModel):
 
         self.init_weights()
 
-    def forward(self, input_ids, attention_mask=None, token_type_ids=None, position_ids=None, head_mask=None,
+    def forward(self, input_ids=None, attention_mask=None, token_type_ids=None, position_ids=None, head_mask=None, inputs_embeds=None,
                 start_positions=None, end_positions=None):
 
         outputs = self.transformer(input_ids,
                             attention_mask=attention_mask,
                             token_type_ids=token_type_ids,
                             position_ids=position_ids, 
-                            head_mask=head_mask)
+                            head_mask=head_mask,
+                            inputs_embeds=inputs_embeds)
 
         sequence_output = outputs[0]
 

templates/adding_a_new_model/tokenization_xxx.py

@@ -172,7 +172,7 @@ class XxxTokenizer(PreTrainedTokenizer):
                 special tokens for the model
 
         Returns:
-            A list of integers in the range [0, 1]: 0 for a special token, 1 for a sequence token.
+            A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
         """
 
         if already_has_special_tokens:

transformers/__init__.py

-__version__ = "2.1.1"
+__version__ = "2.2.0"
 
 # Work around to update TensorFlow's absl.logging threshold which alters the
 # default Python logging output behavior when present.
@@ -25,10 +25,11 @@ from .file_utils import (TRANSFORMERS_CACHE, PYTORCH_TRANSFORMERS_CACHE, PYTORCH
 from .data import (is_sklearn_available,
                    InputExample, InputFeatures, DataProcessor,
                    glue_output_modes, glue_convert_examples_to_features,
-                   glue_processors, glue_tasks_num_labels)
+                   glue_processors, glue_tasks_num_labels,
+                   xnli_output_modes, xnli_processors, xnli_tasks_num_labels)
 
 if is_sklearn_available():
-    from .data import glue_compute_metrics
+    from .data import glue_compute_metrics, xnli_compute_metrics
 
 # Tokenizers
 from .tokenization_utils import (PreTrainedTokenizer)
@@ -42,6 +43,8 @@ from .tokenization_xlnet import XLNetTokenizer, SPIECE_UNDERLINE
 from .tokenization_xlm import XLMTokenizer
 from .tokenization_roberta import RobertaTokenizer
 from .tokenization_distilbert import DistilBertTokenizer
+from .tokenization_albert import AlbertTokenizer
+from .tokenization_camembert import CamembertTokenizer
 
 # Configurations
 from .configuration_utils import PretrainedConfig
@@ -56,6 +59,8 @@ from .configuration_ctrl import CTRLConfig, CTRL_PRETRAINED_CONFIG_ARCHIVE_MAP
 from .configuration_xlm import XLMConfig, XLM_PRETRAINED_CONFIG_ARCHIVE_MAP
 from .configuration_roberta import RobertaConfig, ROBERTA_PRETRAINED_CONFIG_ARCHIVE_MAP
 from .configuration_distilbert import DistilBertConfig, DISTILBERT_PRETRAINED_CONFIG_ARCHIVE_MAP
+from .configuration_albert import AlbertConfig, ALBERT_PRETRAINED_CONFIG_ARCHIVE_MAP
+from .configuration_camembert import CamembertConfig, CAMEMBERT_PRETRAINED_CONFIG_ARCHIVE_MAP
 
 # Modeling
 if is_torch_available():
@@ -72,6 +77,7 @@ if is_torch_available():
                                 OpenAIGPTLMHeadModel, OpenAIGPTDoubleHeadsModel,
                                 load_tf_weights_in_openai_gpt, OPENAI_GPT_PRETRAINED_MODEL_ARCHIVE_MAP)
     from .modeling_transfo_xl import (TransfoXLPreTrainedModel, TransfoXLModel, TransfoXLLMHeadModel,
+                                    AdaptiveEmbedding,
                                     load_tf_weights_in_transfo_xl, TRANSFO_XL_PRETRAINED_MODEL_ARCHIVE_MAP)
     from .modeling_gpt2 import (GPT2PreTrainedModel, GPT2Model,
                                 GPT2LMHeadModel, GPT2DoubleHeadsModel,
@@ -93,12 +99,21 @@ if is_torch_available():
                                 ROBERTA_PRETRAINED_MODEL_ARCHIVE_MAP)
     from .modeling_distilbert import (DistilBertForMaskedLM, DistilBertModel,
                                 DistilBertForSequenceClassification, DistilBertForQuestionAnswering,
+                                DistilBertForTokenClassification,
                                 DISTILBERT_PRETRAINED_MODEL_ARCHIVE_MAP)
+    from .modeling_camembert import (CamembertForMaskedLM, CamembertModel,
+                                CamembertForSequenceClassification, CamembertForMultipleChoice,
+                                CamembertForTokenClassification,
+                                CAMEMBERT_PRETRAINED_MODEL_ARCHIVE_MAP)
     from .modeling_encoder_decoder import PreTrainedEncoderDecoder, Model2Model
 
+    from .modeling_albert import (AlbertModel, AlbertForMaskedLM, AlbertForSequenceClassification,
+                                AlbertForQuestionAnswering,
+                                load_tf_weights_in_albert, ALBERT_PRETRAINED_MODEL_ARCHIVE_MAP)
+
     # Optimization
-    from .optimization import (AdamW, ConstantLRSchedule, WarmupConstantSchedule, WarmupCosineSchedule,
-                               WarmupCosineWithHardRestartsSchedule, WarmupLinearSchedule)
+    from .optimization import (AdamW, get_constant_schedule, get_constant_schedule_with_warmup, get_cosine_schedule_with_warmup,
+                               get_cosine_with_hard_restarts_schedule_with_warmup, get_linear_schedule_with_warmup)
 
 
 # TensorFlow
@@ -154,6 +169,10 @@ if is_tf_available():
                                     TFCTRLLMHeadModel,
                                     TF_CTRL_PRETRAINED_MODEL_ARCHIVE_MAP)
 
+    from .modeling_tf_albert import (TFAlbertPreTrainedModel, TFAlbertModel, TFAlbertForMaskedLM,
+                                     TFAlbertForSequenceClassification,
+                                    TF_ALBERT_PRETRAINED_MODEL_ARCHIVE_MAP)
+
 # TF 2.0 <=> PyTorch conversion utilities
 from .modeling_tf_pytorch_utils import (convert_tf_weight_name_to_pt_weight_name,
                                         load_pytorch_checkpoint_in_tf2_model,