Update modeling_super_linear.py

modeling_super_linear.py

@@ -525,7 +525,10 @@ class SuperLinearForCausalLM(PreTrainedModel, GenerationMixin):
        
 
         # backbone returns (B, pred_len, C)
-        preds = self.backbone(x_enc)[0]
+        preds = self.backbone(x_enc)
+        print(F"preds shape: {preds.shape}")
+        preds = preds[0]
+        print(F"preds shape: {preds.shape}")
 
         # if we keep continuous values, treat them as logits directly
         logits = (preds if self.vocab_size is None else self.lm_head(preds).transpose(1, 2))