add update (#36017)

python/paddle/distributed/fleet/meta_parallel/pipeline_parallel.py

@@ -329,6 +329,7 @@ class PipelineParallel(MetaParallelBase):
     def _optimizer_step(self):
         if self.scaler:
             self.scaler.step(self.optimizer)
+            self.scaler.update()
         else:
             self.optimizer.step()
 

python/paddle/fluid/tests/unittests/hybrid_parallel_mp_amp.py

@@ -48,6 +48,7 @@ class TestMPClipGrad(TestDistMPTraning):
         scaled.backward()  # do backward
 
         scaler.step(optimizer)  # update parameters
+        scaler.update()
         optimizer.clear_grad()
         return scaled