fix: set model_accepts_loss_kwargs=False

src/axolotl/integrations/lolcats/trainer/distill_attention_xent_mse.py

@@ -33,6 +33,8 @@ class DistillAttentionXentMSETrainer(AxolotlTrainer):
         self.xent_factor = xent_factor
         # self.compute_loss_backprop = False  # Whether we backprop in self.compute_loss # NOTE: this config seems unnecessary
 
+        self.model_accepts_loss_kwargs = False  # added to combat explosive loss
+
     def compute_loss(
         self,
         model: nn.Module,