diff --git a/tests/unit/checkpoint/common.py b/tests/unit/checkpoint/common.py
index e891d9d90f8e..001c08f1a99f 100644
--- a/tests/unit/checkpoint/common.py
+++ b/tests/unit/checkpoint/common.py
@@ -218,7 +218,7 @@ def checkpoint_correctness_verification(config_dict,
     for root, _, files in os.walk(save_folder):
         for f in files:
             if "_expert_" in f and "_model_states" in f:
-                expert = torch.load(os.path.join(root, f), weights_only=True)
+                expert = torch.load(os.path.join(root, f), weights_only=False)
                 needed, storages = 0, {}
                 for name, tensor in expert.items():
                     needed += tensor.size().numel()
diff --git a/tests/unit/checkpoint/test_universal_checkpoint.py b/tests/unit/checkpoint/test_universal_checkpoint.py
index 386116aea29a..46d4294bdd0d 100644
--- a/tests/unit/checkpoint/test_universal_checkpoint.py
+++ b/tests/unit/checkpoint/test_universal_checkpoint.py
@@ -181,7 +181,7 @@ def _run_test(self, tmpdir, dtype, ds_config, load_optim, use_torch_adam):
             )
 
         hidden_dim = 10
-        loaded_model_state, loaded_optimizer_state = torch.load(f"{tmpdir}/baseline_state.pt", weights_only=True)
+        loaded_model_state, loaded_optimizer_state = torch.load(f"{tmpdir}/baseline_state.pt", weights_only=False)
 
         ds_config["checkpoint"] = {"load_universal": True}
         univ_model = SimpleModel(hidden_dim)
diff --git a/tests/unit/checkpoint/test_zero_optimizer.py b/tests/unit/checkpoint/test_zero_optimizer.py
index 074a96b5e081..44966b331d0f 100644
--- a/tests/unit/checkpoint/test_zero_optimizer.py
+++ b/tests/unit/checkpoint/test_zero_optimizer.py
@@ -523,7 +523,7 @@ def test_save_exclude_frozen_weights(self, tmpdir, zero_stage):
         all_ckpt_folder = os.path.join(tmpdir, 'all_params')
         ds_engine.save_checkpoint(all_ckpt_folder)
         all_params_ckpt_file = get_model_ckpt_name_for_rank(os.path.join(all_ckpt_folder, 'global_step0'), '00')
-        loaded_all_param_model = torch.load(all_params_ckpt_file, weights_only=True)['module']
+        loaded_all_param_model = torch.load(all_params_ckpt_file, weights_only=False)['module']
         all_param_names = set([n for n, p in model.named_parameters()])
         assert set(loaded_all_param_model.keys()) == all_param_names
 
@@ -536,7 +536,7 @@ def test_save_exclude_frozen_weights(self, tmpdir, zero_stage):
         # Excluding frozen parameters should reduce checkpoint size
         assert os.path.getsize(all_params_ckpt_file) > os.path.getsize(trainable_ckpt_file)
 
-        loaded_trainable_param_model = torch.load(trainable_ckpt_file, weights_only=True)['module']
+        loaded_trainable_param_model = torch.load(trainable_ckpt_file, weights_only=False)['module']
         frozen_param_names = set([n for n, p in model.named_parameters() if not p.requires_grad])
         loaded_trainable_param_names = set(loaded_trainable_param_model.keys())
         overlap_names = set.intersection(loaded_trainable_param_names, frozen_param_names)
@@ -575,7 +575,7 @@ def test_save_exclude_custom_frozen_weights(self, tmpdir, zero_stage):
 
         custom_state_dict_ckpt_file = get_model_ckpt_name_for_rank(
             os.path.join(custom_state_dict_ckpt_folder, 'global_step0'), '00')
-        loaded_custom_state_dict_param_model = torch.load(custom_state_dict_ckpt_file, weights_only=True)['module']
+        loaded_custom_state_dict_param_model = torch.load(custom_state_dict_ckpt_file, weights_only=False)['module']
         loaded_custom_state_dict_param_names = set(loaded_custom_state_dict_param_model.keys())
 
         custom_state_dict_param_names = set([k for k, v in model.state_dict().items()])
diff --git a/tests/unit/model_parallelism/test_configurable_parallel_mp.py b/tests/unit/model_parallelism/test_configurable_parallel_mp.py
index e495e946308d..a7b0d3431ee9 100644
--- a/tests/unit/model_parallelism/test_configurable_parallel_mp.py
+++ b/tests/unit/model_parallelism/test_configurable_parallel_mp.py
@@ -170,7 +170,7 @@ def test(self, baseline_mp2, inputs, class_tmpdir):
             test = model(inputs[0].to(device_name), inputs[1].to(device_name), inputs[2].to(device_name))
             if dist.get_rank() == 0:
                 load_path = os.path.join(class_tmpdir, "output.pt")
-                baseline = torch.load(load_path, weights_only=True)
+                baseline = torch.load(load_path, weights_only=False)
                 test = test.cpu()
                 assert torch.allclose(
                     baseline, test,
diff --git a/tests/unit/model_parallelism/test_configurable_parallel_pp.py b/tests/unit/model_parallelism/test_configurable_parallel_pp.py
index 7a62de77fc05..df469044e186 100644
--- a/tests/unit/model_parallelism/test_configurable_parallel_pp.py
+++ b/tests/unit/model_parallelism/test_configurable_parallel_pp.py
@@ -225,7 +225,7 @@ def _test(self, inputs, class_tmpdir, checkpoint_tag, mp_size, pp_size, mp_resiz
                 assert torch.is_tensor(test[0][0])
                 test = test[0][0].cpu()
                 load_path = os.path.join(class_tmpdir, f"output-{checkpoint_tag}.pt")
-                baseline = torch.load(load_path, weights_only=True)
+                baseline = torch.load(load_path, weights_only=False)
                 assert torch.allclose(
                     baseline, test,
                     atol=1e-03), f"Baseline output {baseline} is not equal to save-then-load output {test}"