Adapt some test case on npu (#42335)

* Adapt some test case on npu

* Adapt some test case on npu

---------

Co-authored-by: mamba-chen <chenhao388@huawei.com>
Co-authored-by: Yih-Dar <2521628+ydshieh@users.noreply.github.com>

Author: 3 people

src/transformers/testing_utils.py

@@ -222,10 +222,12 @@
     IS_ROCM_SYSTEM = torch.version.hip is not None
     IS_CUDA_SYSTEM = torch.version.cuda is not None
     IS_XPU_SYSTEM = getattr(torch.version, "xpu", None) is not None
+    IS_NPU_SYSTEM = getattr(torch, "npu", None) is not None
 else:
     IS_ROCM_SYSTEM = False
     IS_CUDA_SYSTEM = False
     IS_XPU_SYSTEM = False
+    IS_NPU_SYSTEM = False
 
 logger = transformers_logging.get_logger(__name__)
 
@@ -3174,6 +3176,8 @@ def get_device_properties() -> DeviceProperties:
         gen_mask = 0x000000FF00000000
         gen = (arch & gen_mask) >> 32
         return ("xpu", gen, None)
+    elif IS_NPU_SYSTEM:
+        return ("npu", None, None)
     else:
         return (torch_device, None, None)
 

tests/models/qwen3/test_modeling_qwen3.py

@@ -168,6 +168,7 @@ def test_speculative_generation(self):
                 ("xpu", 3): "My favourite condiment is 100% beef and comes in a 12 oz. jar. It is sold in",
                 ("cuda", 7): "My favourite condiment is 100% natural. It's a little spicy and a little sweet, but it's the",
                 ("cuda", 8): "My favourite condiment is 100% beef, 100% beef, 100% beef.",
+                ("npu", None): "My favourite condiment is 100% chicken and beef. I love it because it's so good and I love it",
             }
         )  # fmt: skip
         EXPECTED_TEXT_COMPLETION = EXPECTED_TEXT_COMPLETIONS.get_expectation()
@@ -214,6 +215,7 @@ def test_export_static_cache(self):
                 ("xpu", None): ["My favourite condiment is 100% plain, unflavoured, and unadulterated. It is"],
                 ("rocm", (9, 5)): ["My favourite condiment is 100% plain, unflavoured, and unadulterated."],
                 ("cuda", None): cuda_expectation,
+                ("npu", None): ["My favourite condiment is 100% plain, unsalted, unsweetened, and unflavored. It is"],
             }
         )  # fmt: skip
         EXPECTED_TEXT_COMPLETION = expected_text_completions.get_expectation()

tests/test_modeling_common.py

@@ -448,7 +448,7 @@ def _can_output_attn(model):
             if torch_device in ["cpu", "cuda"]:
                 atol = atols[torch_device, enable_kernels, dtype]
                 rtol = rtols[torch_device, enable_kernels, dtype]
-            elif torch_device == "hpu":
+            elif torch_device in ["hpu", "npu"]:
                 atol = atols["cuda", enable_kernels, dtype]
                 rtol = rtols["cuda", enable_kernels, dtype]
             elif torch_device == "xpu":