A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 5, 12:15 UTC
[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>
Greg Pereira committed
Apr 5, 2026 at 12:11 UTC
4dd49b06f81af2238ac5a86cfb0b7220083eb125
5 files changed
+11
-11
tests/v1/cudagraph/test_encoder_cudagraph.py
+4
-4
@@ -14,17 +14,17 @@ from typing import Any
14
15
import pytest
16
import torch
17
-from vllm.v1.worker.gpu.mm.encoder_cudagraph import (
17
+
18
+from vllm.platforms import current_platform
19
+from vllm.v1.worker.encoder_cudagraph import (
20
EncoderCudaGraphManager,
21
)
20
-from vllm.v1.worker.gpu.mm.encoder_cudagraph_defs import (
22
+from vllm.v1.worker.encoder_cudagraph_defs import (
23
EncoderCudaGraphCaptureInputs,
24
EncoderCudaGraphConfig,
25
EncoderCudaGraphReplayBuffers,
26
)
27
26
-from vllm.platforms import current_platform
27
-
28
# ---------------------------------------------------------------------------
29
# Helpers
30
# ---------------------------------------------------------------------------
vllm/model_executor/models/interfaces.py
+1
-1
@@ -46,7 +46,7 @@ if TYPE_CHECKING:
46
from vllm.multimodal.inputs import MultiModalFeatureSpec
47
from vllm.multimodal.registry import _ProcessorFactories
48
from vllm.sequence import IntermediateTensors
49
- from vllm.v1.worker.gpu.mm.encoder_cudagraph_defs import (
49
+ from vllm.v1.worker.encoder_cudagraph_defs import (
50
EncoderCudaGraphCaptureInputs,
51
EncoderCudaGraphConfig,
52
EncoderCudaGraphReplayBuffers,
vllm/model_executor/models/qwen3_vl.py
+3
-3
@@ -1733,7 +1733,7 @@ class Qwen3VLForConditionalGeneration(
1733
# -- SupportsEncoderCudaGraph protocol methods --
1734
1735
def get_encoder_cudagraph_config(self):
1736
- from vllm.v1.worker.gpu.mm.encoder_cudagraph_defs import (
1736
+ from vllm.v1.worker.encoder_cudagraph_defs import (
1737
EncoderCudaGraphConfig,
1738
)
1739
@@ -1818,7 +1818,7 @@ class Qwen3VLForConditionalGeneration(
1818
device: torch.device,
1819
dtype: torch.dtype,
1820
):
1821
- from vllm.v1.worker.gpu.mm.encoder_cudagraph_defs import (
1821
+ from vllm.v1.worker.encoder_cudagraph_defs import (
1822
EncoderCudaGraphCaptureInputs,
1823
)
1824
@@ -1872,7 +1872,7 @@ class Qwen3VLForConditionalGeneration(
1872
mm_kwargs: dict[str, Any],
1873
max_batch_size: int,
1874
):
1875
- from vllm.v1.worker.gpu.mm.encoder_cudagraph_defs import (
1875
+ from vllm.v1.worker.encoder_cudagraph_defs import (
1876
EncoderCudaGraphReplayBuffers,
1877
)
1878
vllm/v1/worker/encoder_cudagraph.py
+1
-1
@@ -16,7 +16,7 @@ from vllm.distributed import (
16
from vllm.logger import init_logger
17
from vllm.model_executor.models.interfaces import SupportsEncoderCudaGraph
18
from vllm.model_executor.models.vision import get_load_balance_assignment
19
-from vllm.v1.worker.gpu.mm.encoder_cudagraph_defs import (
19
+from vllm.v1.worker.encoder_cudagraph_defs import (
20
EncoderCudaGraphConfig,
21
)
22
vllm/v1/worker/gpu_model_runner.py
+2
-2
@@ -211,7 +211,7 @@ from .utils import (
211
if TYPE_CHECKING:
212
from vllm.v1.core.sched.output import GrammarOutput, SchedulerOutput
213
from vllm.v1.spec_decode.ngram_proposer import NgramProposer
214
- from vllm.v1.worker.gpu.mm.encoder_cudagraph import EncoderCudaGraphManager
214
+ from vllm.v1.worker.encoder_cudagraph import EncoderCudaGraphManager
215
216
logger = init_logger(__name__)
217
@@ -5988,7 +5988,7 @@ class GPUModelRunner(
5988
SupportsEncoderCudaGraph,
5989
supports_encoder_cudagraph,
5990
)
5991
- from vllm.v1.worker.gpu.mm.encoder_cudagraph import (
5991
+ from vllm.v1.worker.encoder_cudagraph import (
5992
EncoderCudaGraphManager,
5993
)
5994