[XPU] Fix XPU UTs for basic coverage (#1164)

Signed-off-by: Yan Ma <yan.ma@intel.com>

tests/e2e/offline_inference/test_cache_dit.py

@@ -24,6 +24,7 @@ if str(REPO_ROOT) not in sys.path:
 
 from vllm_omni import Omni
 from vllm_omni.outputs import OmniRequestOutput
+from vllm_omni.platforms import current_omni_platform
 
 os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "1"
 
@@ -62,7 +63,7 @@ def test_cache_dit(model_name: str):
                 width=width,
                 num_inference_steps=num_inference_steps,
                 guidance_scale=0.0,
-                generator=torch.Generator("cuda").manual_seed(42),
+                generator=torch.Generator(current_omni_platform.device_type).manual_seed(42),
                 num_outputs_per_prompt=1,  # Single output for speed
             ),
         )

tests/e2e/offline_inference/test_t2i_model.py

@@ -47,7 +47,7 @@ def test_diffusion_model(model_name: str):
                 width=width,
                 num_inference_steps=2,
                 guidance_scale=0.0,
-                generator=torch.Generator("cuda").manual_seed(42),
+                generator=torch.Generator(current_omni_platform.device_type).manual_seed(42),
                 num_outputs_per_prompt=2,
             ),
         )

tests/e2e/offline_inference/test_t2v_model.py

@@ -14,6 +14,7 @@ if str(REPO_ROOT) not in sys.path:
 
 from vllm_omni import Omni
 from vllm_omni.outputs import OmniRequestOutput
+from vllm_omni.platforms import current_omni_platform
 
 # os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "1"
 os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
@@ -42,7 +43,7 @@ def test_video_diffusion_model(model_name: str):
             num_frames=num_frames,
             num_inference_steps=2,
             guidance_scale=1.0,
-            generator=torch.Generator("cuda").manual_seed(42),
+            generator=torch.Generator(current_omni_platform.device_type).manual_seed(42),
         ),
     )
     first_output = outputs[0]