Update codeparrot_training.py

codeparrot_training.py

@@ -15,7 +15,7 @@ import wandb
 
 class ConstantLengthDataset(IterableDataset):
     
-    def __init__(self, tokenizer, dataset, seq_length=1024,
+    def __init__(self, tokenizer, dataset, infinite=False, seq_length=1024,
                  num_of_sequences=1024, chars_per_token=3.6):
         self.tokenizer = tokenizer
         self.concat_token_id = tokenizer.bos_token_id
@@ -23,6 +23,7 @@ class ConstantLengthDataset(IterableDataset):
         self.seq_length = seq_length
         self.input_characters = seq_length * chars_per_token * num_of_sequences
         self.epoch = 0
+        self.infinite = infinite
     
     def __iter__(self):
         iterator = iter(self.dataset)
@@ -36,9 +37,13 @@ class ConstantLengthDataset(IterableDataset):
                     buffer.append(next(iterator)['content'])
                     buffer_len += len(buffer[-1])
                 except StopIteration:
-                    iterator = iter(self.dataset)
-                    self.epoch += 1
-                    logger.info(f"Dataset epoch: {self.epoch}")
+                    if self.infinite:
+                        iterator = iter(self.dataset)
+                        self.epoch += 1
+                        logger.info(f"Dataset epoch: {self.epoch}")
+                    else:
+                        more_examples = False
+                        break
             tokenized_inputs = tokenizer(buffer, truncation=False)['input_ids']
             all_token_ids = []
             for tokenized_input in tokenized_inputs:
@@ -77,9 +82,9 @@ def create_dataloaders(dataset_name, args):
     train_data = train_data.shuffle(buffer_size=args.shuffle_buffer,
                                     seed=args.seed)
     valid_data = load_dataset(dataset_name+'-valid', split="train", **ds_kwargs)
-    train_dataset = ConstantLengthDataset(tokenizer, train_data,
+    train_dataset = ConstantLengthDataset(tokenizer, train_data, infinite=True,
                                           seq_length=args.seq_length)
-    valid_dataset = ConstantLengthDataset(tokenizer, valid_data,
+    valid_dataset = ConstantLengthDataset(tokenizer, valid_data, infinite=False,
                                           seq_length=args.seq_length)
     train_dataloader=DataLoader(train_dataset, batch_size=args.train_batch_size)
     eval_dataloader=DataLoader(valid_dataset, batch_size=args.valid_batch_size)