max_length

data.py

@@ -61,7 +61,7 @@ class HateSpeechCollater:
 
         try:
             inputs = self.tokenizer(
-                sentences, return_tensors="pt", padding=True, truncation=True)
+                sentences, return_tensors="pt", padding=True, truncation=True, max_length=128)
         except TypeError:
             pad_id = self.tokenizer.token_to_id("<pad>")
             self.tokenizer.enable_padding(pad_id=pad_id)

train_roberta.py

@@ -25,9 +25,9 @@ if __name__ == "__main__":
 
         for fold, (train, test) in enumerate(dataset.iter_folds(args.folds, True), 1):
             train_loader = DataLoader(
-                train, collate_fn=collater, shuffle=True, batch_size=64, pin_memory=pm)
+                train, collate_fn=collater, shuffle=True, batch_size=128, pin_memory=pm)
             test_loader = DataLoader(
-                test, collate_fn=collater, shuffle=False, batch_size=64, pin_memory=pm)
+                test, collate_fn=collater, shuffle=False, batch_size=128, pin_memory=pm)
 
             if args.jobname is not None:
                 writer = SummaryWriter(