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/nllb_moe/test_modeling_nllb_moe.py

@@ -24,6 +24,7 @@ from transformers.testing_utils import (
     require_sentencepiece,
     require_tokenizers,
     require_torch,
+    require_torch_fp16,
     slow,
     torch_device,
 )
@@ -327,13 +328,13 @@ class NllbMoeModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMi
             with torch.no_grad():
                 model(**inputs)[0]
 
+    @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 = NllbMoeForConditionalGeneration(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)