[Bug] Fix Import paths for `encoder_cudagraph` modules (#38997)

Signed-off-by: greg pereira <grpereir@redhat.com>
Signed-off-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>
Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>

tests/v1/cudagraph/test_encoder_cudagraph.py

@@ -14,17 +14,17 @@ from typing import Any
 
 import pytest
 import torch
-from vllm.v1.worker.gpu.mm.encoder_cudagraph import (
+
+from vllm.platforms import current_platform
+from vllm.v1.worker.encoder_cudagraph import (
     EncoderCudaGraphManager,
 )
-from vllm.v1.worker.gpu.mm.encoder_cudagraph_defs import (
+from vllm.v1.worker.encoder_cudagraph_defs import (
     EncoderCudaGraphCaptureInputs,
     EncoderCudaGraphConfig,
     EncoderCudaGraphReplayBuffers,
 )
 
-from vllm.platforms import current_platform
-
 # ---------------------------------------------------------------------------
 # Helpers
 # ---------------------------------------------------------------------------

vllm/model_executor/models/interfaces.py

@@ -46,7 +46,7 @@ if TYPE_CHECKING:
     from vllm.multimodal.inputs import MultiModalFeatureSpec
     from vllm.multimodal.registry import _ProcessorFactories
     from vllm.sequence import IntermediateTensors
-    from vllm.v1.worker.gpu.mm.encoder_cudagraph_defs import (
+    from vllm.v1.worker.encoder_cudagraph_defs import (
         EncoderCudaGraphCaptureInputs,
         EncoderCudaGraphConfig,
         EncoderCudaGraphReplayBuffers,

vllm/model_executor/models/qwen3_vl.py

@@ -1733,7 +1733,7 @@ class Qwen3VLForConditionalGeneration(
     # -- SupportsEncoderCudaGraph protocol methods --
 
     def get_encoder_cudagraph_config(self):
-        from vllm.v1.worker.gpu.mm.encoder_cudagraph_defs import (
+        from vllm.v1.worker.encoder_cudagraph_defs import (
             EncoderCudaGraphConfig,
         )
 
@@ -1818,7 +1818,7 @@ class Qwen3VLForConditionalGeneration(
         device: torch.device,
         dtype: torch.dtype,
     ):
-        from vllm.v1.worker.gpu.mm.encoder_cudagraph_defs import (
+        from vllm.v1.worker.encoder_cudagraph_defs import (
             EncoderCudaGraphCaptureInputs,
         )
 
@@ -1872,7 +1872,7 @@ class Qwen3VLForConditionalGeneration(
         mm_kwargs: dict[str, Any],
         max_batch_size: int,
     ):
-        from vllm.v1.worker.gpu.mm.encoder_cudagraph_defs import (
+        from vllm.v1.worker.encoder_cudagraph_defs import (
             EncoderCudaGraphReplayBuffers,
         )
 

vllm/v1/worker/encoder_cudagraph.py

@@ -16,7 +16,7 @@ from vllm.distributed import (
 from vllm.logger import init_logger
 from vllm.model_executor.models.interfaces import SupportsEncoderCudaGraph
 from vllm.model_executor.models.vision import get_load_balance_assignment
-from vllm.v1.worker.gpu.mm.encoder_cudagraph_defs import (
+from vllm.v1.worker.encoder_cudagraph_defs import (
     EncoderCudaGraphConfig,
 )
 

vllm/v1/worker/gpu_model_runner.py

@@ -211,7 +211,7 @@ from .utils import (
 if TYPE_CHECKING:
     from vllm.v1.core.sched.output import GrammarOutput, SchedulerOutput
     from vllm.v1.spec_decode.ngram_proposer import NgramProposer
-    from vllm.v1.worker.gpu.mm.encoder_cudagraph import EncoderCudaGraphManager
+    from vllm.v1.worker.encoder_cudagraph import EncoderCudaGraphManager
 
 logger = init_logger(__name__)
 
@@ -5988,7 +5988,7 @@ class GPUModelRunner(
                 SupportsEncoderCudaGraph,
                 supports_encoder_cudagraph,
             )
-            from vllm.v1.worker.gpu.mm.encoder_cudagraph import (
+            from vllm.v1.worker.encoder_cudagraph import (
                 EncoderCudaGraphManager,
             )