diff --git a/configs/mmtune/mmseg_asynchb_nevergrad_pso.py b/configs/mmtune/mmseg_asynchb_nevergrad_pso.py
index 335d1307..a9990181 100644
--- a/configs/mmtune/mmseg_asynchb_nevergrad_pso.py
+++ b/configs/mmtune/mmseg_asynchb_nevergrad_pso.py
@@ -7,7 +7,7 @@
 space = {
     'model': {{_base_.model}},
     'optimizer': {{_base_.optimizer}},
-    'data.samples_per_gpus': {{_base_.batch_size}}
+    'data.samples_per_gpu': {{_base_.batch_size}}
 }
 
 metric = 'val/mIoU'
diff --git a/mmtune/mm/context/rewriters/register.py b/mmtune/mm/context/rewriters/register.py
index 7fafe53b..4c158131 100644
--- a/mmtune/mm/context/rewriters/register.py
+++ b/mmtune/mm/context/rewriters/register.py
@@ -10,7 +10,8 @@ def __init__(self, post_custom_hooks: List[str]) -> None:
         self.post_custom_hooks = post_custom_hooks
 
     def __call__(self, context: dict) -> dict:
-        custom_hooks = getattr(context['cfg'], 'custom_hooks', [])
+        custom_hooks = getattr(context['cfg'], 'custom_hooks', []).copy()
         for custom_hook in self.post_custom_hooks:
             custom_hooks.append(custom_hook)
+        context['cfg'].custom_hooks = custom_hooks
         return context
diff --git a/mmtune/mm/tasks/mmtrainbase.py b/mmtune/mm/tasks/mmtrainbase.py
index 5fcc76ec..f969010a 100644
--- a/mmtune/mm/tasks/mmtrainbase.py
+++ b/mmtune/mm/tasks/mmtrainbase.py
@@ -35,6 +35,7 @@ def train_model(cls, model: torch.nn.Module,
 
     @classmethod
     def contextaware_run(cls, status, backend, *args, **kwargs) -> None:
+        from mmtune.mm.tasks import hooks  # noqa F401
         if backend == 'nccl' and os.getenv('NCCL_BLOCKING_WAIT') is None:
             os.environ['NCCL_BLOCKING_WAIT'] = '0'
         context_manager = ContextManager(**status)
@@ -53,5 +54,5 @@ def create_trainable(cls, backend: str = 'nccl') -> ray.tune.trainable:
                     rewriters=cls.REWRITERS), backend),
             backend=backend,
             num_workers=cls.ARGS.num_workers,
-            num_gpus_per_worker=cls.ARGS.num_cpus_per_worker,
+            num_gpus_per_worker=cls.ARGS.num_gpus_per_worker,
             num_cpus_per_worker=cls.ARGS.num_cpus_per_worker)