enhance require_deterministic_for_xpu (#37437)

* enhance require_deterministic_for_xpu

Signed-off-by: YAO Matrix <matrix.yao@intel.com>

* fix style

Signed-off-by: YAO Matrix <matrix.yao@intel.com>

* fix style

Signed-off-by: YAO Matrix <matrix.yao@intel.com>

---------

Signed-off-by: YAO Matrix <matrix.yao@intel.com>

src/transformers/testing_utils.py

@@ -139,7 +139,6 @@ from .utils import (
     is_torch_available,
     is_torch_bf16_available_on_device,
     is_torch_bf16_gpu_available,
-    is_torch_deterministic,
     is_torch_fp16_available_on_device,
     is_torch_greater_or_equal,
     is_torch_hpu_available,
@@ -1073,12 +1072,19 @@ def require_torch_bf16_gpu(test_case):
 
 
 def require_deterministic_for_xpu(test_case):
-    if is_torch_xpu_available():
-        return unittest.skipUnless(is_torch_deterministic(), "test requires torch to use deterministic algorithms")(
-            test_case
-        )
-    else:
-        return test_case
+    @wraps(test_case)
+    def wrapper(*args, **kwargs):
+        if is_torch_xpu_available():
+            original_state = torch.are_deterministic_algorithms_enabled()
+            try:
+                torch.use_deterministic_algorithms(True)
+                return test_case(*args, **kwargs)
+            finally:
+                torch.use_deterministic_algorithms(original_state)
+        else:
+            return test_case(*args, **kwargs)
+
+    return wrapper
 
 
 def require_torch_tf32(test_case):

tests/models/encoder_decoder/test_modeling_encoder_decoder.py

@@ -936,6 +936,7 @@ class BertGenerationEncoderDecoderModelTest(EncoderDecoderMixin, unittest.TestCa
         }
 
     @slow
+    @require_deterministic_for_xpu
     def test_roberta2roberta_summarization(self):
         model = EncoderDecoderModel.from_pretrained("google/roberta2roberta_L-24_bbc")
         model.to(torch_device)
@@ -1080,6 +1081,7 @@ class GPT2EncoderDecoderModelTest(EncoderDecoderMixin, unittest.TestCase):
         pass
 
     @slow
+    @require_deterministic_for_xpu
     def test_bert2gpt2_summarization(self):
         model = EncoderDecoderModel.from_pretrained("patrickvonplaten/bert2gpt2-cnn_dailymail-fp16")
 

tests/models/speech_encoder_decoder/test_modeling_speech_encoder_decoder.py

@@ -634,6 +634,7 @@ class Speech2TextBertModelTest(EncoderDecoderMixin, unittest.TestCase):
     def test_encoder_decoder_model_from_pretrained_configs(self):
         pass
 
+    @require_deterministic_for_xpu
     @unittest.skip(reason="Cannot save full model as Speech2TextModel != Speech2TextEncoder")
     def test_save_and_load_from_pretrained(self):
         pass