fix: num_items_in_batch wrong type in kd trainer loss

src/axolotl/integrations/kd/trainer.py

@@ -74,6 +74,9 @@ class AxolotlKDTrainer(AxolotlTrainer):
         target_token_ids_for_loss = target_token_ids[..., 1:, :].contiguous()
         target_mask_for_loss = target_mask[..., 1:, :].contiguous()
 
+        if num_items_in_batch is None:
+            num_items_in_batch = -1
+
         if self.args.kd_zscore_base_temp:
             loss_kd = topk_kd_loss_with_zscore(
                 shift_logits,