Merge pull request #515 from Rocketknight1/master

Fix --reduce_memory in finetune_on_pregenerated

examples/lm_finetuning/finetune_on_pregenerated.py

@@ -74,7 +74,7 @@ class PregeneratedDataset(Dataset):
                                   mode='w+', dtype=np.int32, shape=(num_samples, seq_len))
             input_masks = np.memmap(filename=self.working_dir/'input_masks.memmap',
                                     shape=(num_samples, seq_len), mode='w+', dtype=np.bool)
-            segment_ids = np.memmap(filename=self.working_dir/'input_masks.memmap',
+            segment_ids = np.memmap(filename=self.working_dir/'segment_ids.memmap',
                                     shape=(num_samples, seq_len), mode='w+', dtype=np.bool)
             lm_label_ids = np.memmap(filename=self.working_dir/'lm_label_ids.memmap',
                                      shape=(num_samples, seq_len), mode='w+', dtype=np.int32)
@@ -283,7 +283,7 @@ def main():
     model.train()
     for epoch in range(args.epochs):
         epoch_dataset = PregeneratedDataset(epoch=epoch, training_path=args.pregenerated_data, tokenizer=tokenizer,
-                                            num_data_epochs=num_data_epochs)
+                                            num_data_epochs=num_data_epochs, reduce_memory=args.reduce_memory)
         if args.local_rank == -1:
             train_sampler = RandomSampler(epoch_dataset)
         else: