[tests] properly skip tests instead of return (#11771)

model test updates

Author: sayakpaul

tests/models/test_modeling_common.py

@@ -30,6 +30,7 @@
 from typing import Dict, List, Optional, Tuple, Union
 
 import numpy as np
+import pytest
 import requests_mock
 import safetensors.torch
 import torch
@@ -938,8 +939,9 @@ def recursive_check(tuple_object, dict_object):
 
     @require_torch_accelerator_with_training
     def test_enable_disable_gradient_checkpointing(self):
+        # Skip test if model does not support gradient checkpointing
         if not self.model_class._supports_gradient_checkpointing:
-            return  # Skip test if model does not support gradient checkpointing
+            pytest.skip("Gradient checkpointing is not supported.")
 
         init_dict, _ = self.prepare_init_args_and_inputs_for_common()
 
@@ -957,8 +959,9 @@ def test_enable_disable_gradient_checkpointing(self):
 
     @require_torch_accelerator_with_training
     def test_effective_gradient_checkpointing(self, loss_tolerance=1e-5, param_grad_tol=5e-5, skip: set[str] = {}):
+        # Skip test if model does not support gradient checkpointing
         if not self.model_class._supports_gradient_checkpointing:
-            return  # Skip test if model does not support gradient checkpointing
+            pytest.skip("Gradient checkpointing is not supported.")
 
         # enable deterministic behavior for gradient checkpointing
         init_dict, inputs_dict = self.prepare_init_args_and_inputs_for_common()
@@ -1015,8 +1018,9 @@ def test_effective_gradient_checkpointing(self, loss_tolerance=1e-5, param_grad_
     def test_gradient_checkpointing_is_applied(
         self, expected_set=None, attention_head_dim=None, num_attention_heads=None, block_out_channels=None
     ):
+        # Skip test if model does not support gradient checkpointing
         if not self.model_class._supports_gradient_checkpointing:
-            return  # Skip test if model does not support gradient checkpointing
+            pytest.skip("Gradient checkpointing is not supported.")
 
         init_dict, inputs_dict = self.prepare_init_args_and_inputs_for_common()
 
@@ -1073,7 +1077,7 @@ def test_save_load_lora_adapter(self, rank, lora_alpha, use_dora=False):
         model = self.model_class(**init_dict).to(torch_device)
 
         if not issubclass(model.__class__, PeftAdapterMixin):
-            return
+            pytest.skip(f"PEFT is not supported for this model ({model.__class__.__name__}).")
 
         torch.manual_seed(0)
         output_no_lora = model(**inputs_dict, return_dict=False)[0]
@@ -1128,7 +1132,7 @@ def test_lora_wrong_adapter_name_raises_error(self):
         model = self.model_class(**init_dict).to(torch_device)
 
         if not issubclass(model.__class__, PeftAdapterMixin):
-            return
+            pytest.skip(f"PEFT is not supported for this model ({model.__class__.__name__}).")
 
         denoiser_lora_config = LoraConfig(
             r=4,
@@ -1159,7 +1163,7 @@ def test_lora_adapter_metadata_is_loaded_correctly(self, rank, lora_alpha, use_d
         model = self.model_class(**init_dict).to(torch_device)
 
         if not issubclass(model.__class__, PeftAdapterMixin):
-            return
+            pytest.skip(f"PEFT is not supported for this model ({model.__class__.__name__}).")
 
         denoiser_lora_config = LoraConfig(
             r=rank,
@@ -1196,7 +1200,7 @@ def test_lora_adapter_wrong_metadata_raises_error(self):
         model = self.model_class(**init_dict).to(torch_device)
 
         if not issubclass(model.__class__, PeftAdapterMixin):
-            return
+            pytest.skip(f"PEFT is not supported for this model ({model.__class__.__name__}).")
 
         denoiser_lora_config = LoraConfig(
             r=4,
@@ -1233,10 +1237,10 @@ def test_lora_adapter_wrong_metadata_raises_error(self):
 
     @require_torch_accelerator
     def test_cpu_offload(self):
+        if self.model_class._no_split_modules is None:
+            pytest.skip("Test not supported for this model as `_no_split_modules` is not set.")
         config, inputs_dict = self.prepare_init_args_and_inputs_for_common()
         model = self.model_class(**config).eval()
-        if model._no_split_modules is None:
-            return
 
         model = model.to(torch_device)
 
@@ -1263,10 +1267,10 @@ def test_cpu_offload(self):
 
     @require_torch_accelerator
     def test_disk_offload_without_safetensors(self):
+        if self.model_class._no_split_modules is None:
+            pytest.skip("Test not supported for this model as `_no_split_modules` is not set.")
         config, inputs_dict = self.prepare_init_args_and_inputs_for_common()
         model = self.model_class(**config).eval()
-        if model._no_split_modules is None:
-            return
 
         model = model.to(torch_device)
 
@@ -1296,10 +1300,10 @@ def test_disk_offload_without_safetensors(self):
 
     @require_torch_accelerator
     def test_disk_offload_with_safetensors(self):
+        if self.model_class._no_split_modules is None:
+            pytest.skip("Test not supported for this model as `_no_split_modules` is not set.")
         config, inputs_dict = self.prepare_init_args_and_inputs_for_common()
         model = self.model_class(**config).eval()
-        if model._no_split_modules is None:
-            return
 
         model = model.to(torch_device)
 
@@ -1324,10 +1328,10 @@ def test_disk_offload_with_safetensors(self):
 
     @require_torch_multi_accelerator
     def test_model_parallelism(self):
+        if self.model_class._no_split_modules is None:
+            pytest.skip("Test not supported for this model as `_no_split_modules` is not set.")
         config, inputs_dict = self.prepare_init_args_and_inputs_for_common()
         model = self.model_class(**config).eval()
-        if model._no_split_modules is None:
-            return
 
         model = model.to(torch_device)
 
@@ -1426,10 +1430,10 @@ def test_sharded_checkpoints_with_variant(self):
 
     @require_torch_accelerator
     def test_sharded_checkpoints_device_map(self):
+        if self.model_class._no_split_modules is None:
+            pytest.skip("Test not supported for this model as `_no_split_modules` is not set.")
         config, inputs_dict = self.prepare_init_args_and_inputs_for_common()
         model = self.model_class(**config).eval()
-        if model._no_split_modules is None:
-            return
         model = model.to(torch_device)
 
         torch.manual_seed(0)
@@ -1497,7 +1501,7 @@ def test_variant_sharded_ckpt_right_format(self):
     def test_layerwise_casting_training(self):
         def test_fn(storage_dtype, compute_dtype):
             if torch.device(torch_device).type == "cpu" and compute_dtype == torch.bfloat16:
-                return
+                pytest.skip("Skipping test because CPU doesn't go well with bfloat16.")
             init_dict, inputs_dict = self.prepare_init_args_and_inputs_for_common()
 
             model = self.model_class(**init_dict)
@@ -1617,6 +1621,9 @@ def get_memory_usage(storage_dtype, compute_dtype):
     @parameterized.expand([False, True])
     @require_torch_accelerator
     def test_group_offloading(self, record_stream):
+        if not self.model_class._supports_group_offloading:
+            pytest.skip("Model does not support group offloading.")
+
         init_dict, inputs_dict = self.prepare_init_args_and_inputs_for_common()
         torch.manual_seed(0)
 
@@ -1633,8 +1640,6 @@ def run_forward(model):
             return model(**inputs_dict)[0]
 
         model = self.model_class(**init_dict)
-        if not getattr(model, "_supports_group_offloading", True):
-            return
 
         model.to(torch_device)
         output_without_group_offloading = run_forward(model)
@@ -1670,13 +1675,13 @@ def run_forward(model):
     @require_torch_accelerator
     @torch.no_grad()
     def test_group_offloading_with_layerwise_casting(self, record_stream, offload_type):
+        if not self.model_class._supports_group_offloading:
+            pytest.skip("Model does not support group offloading.")
+
         torch.manual_seed(0)
         init_dict, inputs_dict = self.prepare_init_args_and_inputs_for_common()
         model = self.model_class(**init_dict)
 
-        if not getattr(model, "_supports_group_offloading", True):
-            return
-
         model.to(torch_device)
         model.eval()
         _ = model(**inputs_dict)[0]
@@ -1698,13 +1703,13 @@ def test_group_offloading_with_layerwise_casting(self, record_stream, offload_ty
     @require_torch_accelerator
     @torch.no_grad()
     def test_group_offloading_with_disk(self, record_stream, offload_type):
+        if not self.model_class._supports_group_offloading:
+            pytest.skip("Model does not support group offloading.")
+
         torch.manual_seed(0)
         init_dict, inputs_dict = self.prepare_init_args_and_inputs_for_common()
         model = self.model_class(**init_dict)
 
-        if not getattr(model, "_supports_group_offloading", True):
-            return
-
         torch.manual_seed(0)
         init_dict, inputs_dict = self.prepare_init_args_and_inputs_for_common()
         model = self.model_class(**init_dict)