[XPU] move xpu_attn_backend.py to FastDeploy/fastdeploy/model_executor/layers/backends/xpu (#5878)

fastdeploy/model_executor/layers/attention/__init__.py

@@ -23,14 +23,12 @@ from .iluvatar_attn_backend import IluvatarAttnBackend
 from .mla_attention_backend import MLAAttentionBackend
 from .moba_attention_backend import PlasAttentionBackend
 from .native_paddle_backend import PaddleNativeAttnBackend
-from .xpu_attn_backend import XPUAttentionBackend
 
 __all__ = [
     "AttentionBackend",
     "PaddleNativeAttnBackend",
     "get_attention_backend",
     "AppendAttentionBackend",
-    "XPUAttentionBackend",
     "MLAAttentionBackend",
     "FlashAttentionBackend",
     "IluvatarAttnBackend",

fastdeploy/model_executor/layers/backends/xpu/__init__.py

@@ -16,6 +16,7 @@
 xpu backend methods
 """
 
+from .attention import XPUAttentionBackend
 from .moe.fused_moe import XPUMoEMethod, XPUWeightOnlyMoEMethod
 from .quantization.weight_only import XPUWeightOnlyLinearMethod
 
@@ -23,4 +24,5 @@ __all__ = [
     "XPUWeightOnlyLinearMethod",
     "XPUMoEMethod",
     "XPUWeightOnlyMoEMethod",
+    "XPUAttentionBackend",
 ]

fastdeploy/platforms/xpu.py

@@ -51,8 +51,8 @@ class XPUPlatform(Platform):
         get_attention_backend_cls
         """
         # TODO: 等支持配置 attention engine 之后再改回去
-        return "fastdeploy.model_executor.layers.attention.XPUAttentionBackend"
+        return "fastdeploy.model_executor.layers.backends.xpu.XPUAttentionBackend"
         if selected_backend == _Backend.NATIVE_ATTN:
-            return "fastdeploy.model_executor.layers.attention.XPUAttentionBackend"
+            return "fastdeploy.model_executor.layers.backends.xpu.XPUAttentionBackend"
         else:
             logger.warning("Other backends are not supported for now for XPU.")

tests/platforms/test_platforms.py

@@ -220,7 +220,7 @@ class TestXPUPlatform(unittest.TestCase):
 
     def test_get_attention_backend_cls(self):
         """Verify NATIVE_ATTN returns correct XPU backend class"""
-        expected_cls = "fastdeploy.model_executor.layers.attention.XPUAttentionBackend"
+        expected_cls = "fastdeploy.model_executor.layers.backends.xpu.XPUAttentionBackend"
         self.assertEqual(XPUPlatform.get_attention_backend_cls(_Backend.NATIVE_ATTN), expected_cls)