TF multiple choice loss fix (#13513)

Fix issues with `TFMultipleChoiceLoss` if the choices dimension is None when `build()` is called.

src/transformers/modeling_tf_utils.py

@@ -220,9 +220,15 @@ class TFSequenceClassificationLoss:
         return loss_fn(labels, logits)
 
 
-class TFMultipleChoiceLoss(TFSequenceClassificationLoss):
+class TFMultipleChoiceLoss:
     """Loss function suitable for multiple choice tasks."""
 
+    def compute_loss(self, labels, logits):
+        loss_fn = tf.keras.losses.SparseCategoricalCrossentropy(
+            from_logits=True, reduction=tf.keras.losses.Reduction.NONE
+        )
+        return loss_fn(labels, logits)
+
 
 class TFMaskedLanguageModelingLoss(TFCausalLanguageModelingLoss):
     """