diff --git a/python/paddle/distributed/fleet/meta_optimizers/dygraph_optimizer/hybrid_parallel_optimizer.py b/python/paddle/distributed/fleet/meta_optimizers/dygraph_optimizer/hybrid_parallel_optimizer.py index beb3fa0ff6875..a553a8a5d9cc2 100755 --- a/python/paddle/distributed/fleet/meta_optimizers/dygraph_optimizer/hybrid_parallel_optimizer.py +++ b/python/paddle/distributed/fleet/meta_optimizers/dygraph_optimizer/hybrid_parallel_optimizer.py @@ -432,7 +432,9 @@ def minimize( # minimize does not support parameters in the form of param_group, # so no need use _obtain_optimizer_parameters_list parameter_list = ( - parameters if parameters else self._inner_opt._parameter_list + parameters + if parameters + else _obtain_optimizer_parameters_list(self._inner_opt) ) # Here sharding should use global parameter list