Adding files from hf_modeling_btm_log_prob_mixing

modeling.py

@@ -185,7 +185,7 @@ class MoLM(PreTrainedModel):
             loss_to_log = None
 
         return Output(
-            logits=torch.Tensor([expert_output for expert_output in expert_outputs]),
+            logits=expert_outputs,
             loss=loss,
             combined_log_probs=combined_log_probs,
             loss_to_log=loss_to_log,