device agnostic models testing (#27146)

* device agnostic models testing

* add decorator `require_torch_fp16`

* make style

* apply review suggestion

* Oops, the fp16 decorator was misused

tests/models/blenderbot/test_modeling_blenderbot.py

@@ -18,7 +18,15 @@ import tempfile
 import unittest
 
 from transformers import BlenderbotConfig, is_torch_available
-from transformers.testing_utils import require_sentencepiece, require_tokenizers, require_torch, slow, torch_device
+from transformers.testing_utils import (
+    backend_empty_cache,
+    require_sentencepiece,
+    require_tokenizers,
+    require_torch,
+    require_torch_fp16,
+    slow,
+    torch_device,
+)
 from transformers.utils import cached_property
 
 from ...generation.test_utils import GenerationTesterMixin
@@ -269,13 +277,13 @@ class BlenderbotModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTeste
         config_and_inputs = self.model_tester.prepare_config_and_inputs_for_common()
         self.model_tester.check_encoder_decoder_model_standalone(*config_and_inputs)
 
+    @require_torch_fp16
     def test_generate_fp16(self):
         config, input_dict = self.model_tester.prepare_config_and_inputs()
         input_ids = input_dict["input_ids"]
         attention_mask = input_ids.ne(1).to(torch_device)
         model = BlenderbotForConditionalGeneration(config).eval().to(torch_device)
-        if torch_device == "cuda":
-            model.half()
+        model.half()
         model.generate(input_ids, attention_mask=attention_mask)
         model.generate(num_beams=4, do_sample=True, early_stopping=False, num_return_sequences=3)
 
@@ -315,7 +323,7 @@ class Blenderbot3BIntegrationTests(unittest.TestCase):
         FASTER_GEN_KWARGS = {"num_beams": 1, "early_stopping": True, "min_length": 15, "max_length": 25}
         TOK_DECODE_KW = {"skip_special_tokens": True, "clean_up_tokenization_spaces": True}
 
-        torch.cuda.empty_cache()
+        backend_empty_cache(torch_device)
         model = BlenderbotForConditionalGeneration.from_pretrained(self.ckpt).half().to(torch_device)
 
         src_text = ["Sam"]