Remove shared memory from MPFuture, fix minor bugs

hivemind/moe/server/task_pool.py

@@ -100,7 +100,7 @@ def __init__(
 
     def submit_task(self, *args: torch.Tensor) -> Future:
         """Add task to this pool's queue, return Future for its output"""
-        task = Task(MPFuture(), args)
+        task = Task(MPFuture(synchronize=False), args)
         if self.get_task_size(task) > self.max_batch_size:
             exc = ValueError(f"Task size greater than max_batch_size ({self.max_batch_size}), it can't be processed")
             task.future.set_exception(exc)