[T5 failing CI] Fix generate test (#11770)

* fix_torch_device_generate_test

* remove @

tests/test_generation_utils.py

@@ -1084,9 +1084,13 @@ class GenerationTesterMixin:
                 continue
 
             head_masking = {
-                "head_mask": torch.zeros(config.encoder_layers, config.encoder_attention_heads),
-                "decoder_head_mask": torch.zeros(config.decoder_layers, config.decoder_attention_heads),
-                "cross_attn_head_mask": torch.zeros(config.decoder_layers, config.decoder_attention_heads),
+                "head_mask": torch.zeros(config.encoder_layers, config.encoder_attention_heads, device=torch_device),
+                "decoder_head_mask": torch.zeros(
+                    config.decoder_layers, config.decoder_attention_heads, device=torch_device
+                ),
+                "cross_attn_head_mask": torch.zeros(
+                    config.decoder_layers, config.decoder_attention_heads, device=torch_device
+                ),
             }
 
             signature = inspect.signature(model.forward)

tests/test_modeling_t5.py

@@ -605,19 +605,22 @@ class T5ModelTest(ModelTesterMixin, GenerationTesterMixin, unittest.TestCase):
         config_and_inputs = self.model_tester.prepare_config_and_inputs()
         config = config_and_inputs[0]
         max_length = config_and_inputs[1].shape[-1] + 3
-        model = T5ForConditionalGeneration(config)
+        model = T5ForConditionalGeneration(config).eval()
+        model.to(torch_device)
 
         head_masking = {
-            "head_mask": torch.zeros(config.num_layers, config.num_heads),
-            "decoder_head_mask": torch.zeros(config.num_decoder_layers, config.num_heads),
-            "cross_attn_head_mask": torch.zeros(config.num_decoder_layers, config.num_heads),
+            "head_mask": torch.zeros(config.num_layers, config.num_heads, device=torch_device),
+            "decoder_head_mask": torch.zeros(config.num_decoder_layers, config.num_heads, device=torch_device),
+            "cross_attn_head_mask": torch.zeros(config.num_decoder_layers, config.num_heads, device=torch_device),
         }
 
         for attn_name, (name, mask) in zip(attention_names, head_masking.items()):
             head_masks = {name: mask}
             # Explicitly pass decoder_head_mask as it is required from T5 model when head_mask specified
             if name == "head_mask":
-                head_masks["decoder_head_mask"] = torch.ones(config.num_decoder_layers, config.num_heads)
+                head_masks["decoder_head_mask"] = torch.ones(
+                    config.num_decoder_layers, config.num_heads, device=torch_device
+                )
 
             out = model.generate(
                 config_and_inputs[1],