[MoE] Move PF Methods to Folder (#35927)

tests/kernels/moe/modular_kernel_tools/mk_objects.py

@@ -199,10 +199,10 @@ register_experts(
 
 # Disable on blackwell for now
 if has_deep_ep() and not current_platform.has_device_capability(100):
-    from vllm.model_executor.layers.fused_moe.deepep_ht_prepare_finalize import (
+    from vllm.model_executor.layers.fused_moe.prepare_finalize.deepep_ht import (
         DeepEPHTPrepareAndFinalize,
     )
-    from vllm.model_executor.layers.fused_moe.deepep_ll_prepare_finalize import (
+    from vllm.model_executor.layers.fused_moe.prepare_finalize.deepep_ll import (
         DeepEPLLPrepareAndFinalize,
     )
 
@@ -240,7 +240,7 @@ if has_flashinfer_cutlass_fused_moe() and current_platform.has_device_capability
     from vllm.model_executor.layers.fused_moe.flashinfer_cutlass_moe import (
         FlashInferExperts,
     )
-    from vllm.model_executor.layers.fused_moe.flashinfer_nvlink_two_sided_prepare_finalize import (  # noqa: E501
+    from vllm.model_executor.layers.fused_moe.prepare_finalize.flashinfer_nvlink_two_sided import (  # noqa: E501
         FlashInferNVLinkTwoSidedPrepareAndFinalize,
     )
 
@@ -271,7 +271,7 @@ if (
     and has_flashinfer_cutlass_fused_moe()
     and current_platform.has_device_capability(100)
 ):
-    from vllm.model_executor.layers.fused_moe.flashinfer_nvlink_one_sided_prepare_finalize import (  # noqa: E501
+    from vllm.model_executor.layers.fused_moe.prepare_finalize.flashinfer_nvlink_one_sided import (  # noqa: E501
         FlashInferNVLinkOneSidedPrepareAndFinalize,
     )
 

tests/kernels/moe/parallel_utils.py

@@ -19,10 +19,10 @@ from vllm.utils.import_utils import has_deep_ep
 from vllm.utils.network_utils import get_open_port
 
 if has_deep_ep():
-    from vllm.model_executor.layers.fused_moe.deepep_ht_prepare_finalize import (
+    from vllm.model_executor.layers.fused_moe.prepare_finalize.deepep_ht import (
         DeepEPHTPrepareAndFinalize,
     )
-    from vllm.model_executor.layers.fused_moe.deepep_ll_prepare_finalize import (
+    from vllm.model_executor.layers.fused_moe.prepare_finalize.deepep_ll import (
         DeepEPLLPrepareAndFinalize,
     )
 

tests/kernels/moe/test_deepep_deepgemm_moe.py

@@ -37,10 +37,10 @@ from .parallel_utils import ProcessGroupInfo, parallel_launch
 from .utils import make_dummy_moe_config, make_test_weights
 
 if has_deep_ep():
-    from vllm.model_executor.layers.fused_moe.deepep_ht_prepare_finalize import (
+    from vllm.model_executor.layers.fused_moe.prepare_finalize.deepep_ht import (
         DeepEPHTPrepareAndFinalize,
     )
-    from vllm.model_executor.layers.fused_moe.deepep_ll_prepare_finalize import (
+    from vllm.model_executor.layers.fused_moe.prepare_finalize.deepep_ll import (
         DeepEPLLPrepareAndFinalize,
     )
 

tests/kernels/moe/test_deepep_moe.py

@@ -32,10 +32,10 @@ from ...utils import multi_gpu_test
 from .parallel_utils import ProcessGroupInfo, parallel_launch
 
 if has_deep_ep():
-    from vllm.model_executor.layers.fused_moe.deepep_ht_prepare_finalize import (
+    from vllm.model_executor.layers.fused_moe.prepare_finalize.deepep_ht import (
         DeepEPHTPrepareAndFinalize,
     )
-    from vllm.model_executor.layers.fused_moe.deepep_ll_prepare_finalize import (
+    from vllm.model_executor.layers.fused_moe.prepare_finalize.deepep_ll import (
         DeepEPLLPrepareAndFinalize,
     )