git pull

hivemind/moe/client/balanced_expert.py

@@ -123,7 +123,6 @@ class _BalancedRemoteModuleCall(torch.autograd.Function):
                 break
             except BaseException as e:
                 logger.error(f"Tried to call forward for expert {chosen_expert} but caught {e}")
-                raise
 
         deserialized_outputs = [deserialize_torch_tensor(tensor) for tensor in outputs.tensors]
         return tuple(deserialized_outputs)
@@ -146,6 +145,5 @@ class _BalancedRemoteModuleCall(torch.autograd.Function):
                 break
             except BaseException as e:
                 logger.error(f"Tried to call backward for expert {chosen_expert} but caught {e}")
-                raise
         deserialized_grad_inputs = [deserialize_torch_tensor(tensor) for tensor in grad_inputs.tensors]
         return (DUMMY, None, None, None, None, None, None, *deserialized_grad_inputs)