fix: adapt the import path change for ToT vLLM

components/src/dynamo/frontend/vllm_processor.py

@@ -14,7 +14,6 @@
 from typing import Any
 
 from vllm.config import CacheConfig, LoadConfig, ModelConfig, VllmConfig
-from vllm.inputs.data import TokensPrompt
 from vllm.reasoning import ReasoningParser, ReasoningParserManager
 from vllm.sampling_params import RequestOutputKind, SamplingParams
 from vllm.tasks import GENERATION_TASKS
@@ -39,6 +38,13 @@
 from .prepost import StreamingPostProcessor, preprocess_chat_request
 from .utils import random_uuid
 
+# Try importing from new vLLM (https://github.com/vllm-project/vllm/pull/35182), fallback to old structure
+try:
+    from vllm.inputs.llm import TokensPrompt
+except ImportError:
+    from vllm.inputs.data import TokensPrompt
+
+
 logger = logging.getLogger(__name__)
 
 

components/src/dynamo/vllm/multimodal_utils/chat_processor.py

@@ -20,7 +20,6 @@
 from vllm.config import ModelConfig
 from vllm.engine.arg_utils import AsyncEngineArgs
 from vllm.entrypoints.chat_utils import ConversationMessage
-from vllm.inputs.data import TokensPrompt
 from vllm.sampling_params import SamplingParams
 from vllm.tokenizers import TokenizerLike as AnyTokenizer
 
@@ -46,6 +45,12 @@
         OpenAIServingModels,
     )
 
+# Try importing from new vLLM (https://github.com/vllm-project/vllm/pull/35182), fallback to old structure
+try:
+    from vllm.inputs.llm import TokensPrompt
+except ImportError:
+    from vllm.inputs.data import TokensPrompt
+
 
 class StubEngineClient:
     """

components/src/dynamo/vllm/multimodal_utils/protocol.py

@@ -22,15 +22,21 @@
 from pydantic import BaseModel, ConfigDict, Field, field_serializer, field_validator
 from pydantic_core import core_schema
 from typing_extensions import NotRequired
-from vllm.inputs.data import TokensPrompt
 from vllm.logprobs import PromptLogprobs
-from vllm.multimodal.inputs import MultiModalUUIDDict  # noqa: F401
 from vllm.outputs import CompletionOutput
 from vllm.sampling_params import SamplingParams
 from vllm.v1.metrics.stats import RequestStateStats
 
 from dynamo.common.multimodal.embedding_transfer import TransferRequest
 
+# Try importing from new vLLM (https://github.com/vllm-project/vllm/pull/35182), fallback to old structure
+try:
+    from vllm.inputs.llm import MultiModalUUIDDict  # noqa: F401
+    from vllm.inputs.llm import TokensPrompt
+except ImportError:
+    from vllm.inputs.data import TokensPrompt
+    from vllm.multimodal.inputs import MultiModalUUIDDict  # noqa: F401
+
 
 class Request(BaseModel):
     prompt: str

components/src/dynamo/vllm/tests/test_vllm_renderer_api.py

@@ -23,7 +23,6 @@
     "vllm.entrypoints.openai.chat_completion.protocol"
 )
 _engine_protocol = importlib.import_module("vllm.entrypoints.openai.engine.protocol")
-_inputs_data = importlib.import_module("vllm.inputs.data")
 _reasoning = importlib.import_module("vllm.reasoning")
 _sampling_params = importlib.import_module("vllm.sampling_params")
 _tool_parsers = importlib.import_module("vllm.tool_parsers")
@@ -34,7 +33,6 @@
 ChatCompletionRequest = _chat_protocol.ChatCompletionRequest
 DeltaMessage = _engine_protocol.DeltaMessage
 DeltaToolCall = _engine_protocol.DeltaToolCall
-TokensPrompt = _inputs_data.TokensPrompt
 ReasoningParser = _reasoning.ReasoningParser
 ReasoningParserManager = _reasoning.ReasoningParserManager
 RequestOutputKind = _sampling_params.RequestOutputKind
@@ -48,6 +46,14 @@
 OutputProcessor = _output_processor_mod.OutputProcessor
 OutputProcessorOutput = _output_processor_mod.OutputProcessorOutput
 
+# Try importing from new vLLM (https://github.com/vllm-project/vllm/pull/35182), fallback to legacy location
+try:
+    _inputs_mod = importlib.import_module("vllm.inputs.llm")
+except (ImportError, ModuleNotFoundError):
+    _inputs_mod = importlib.import_module("vllm.inputs.data")
+
+TokensPrompt = _inputs_mod.TokensPrompt
+
 pytestmark = [
     pytest.mark.vllm,
     pytest.mark.unit,

examples/multimodal/components/worker.py

@@ -17,7 +17,6 @@
 import torch
 import uvloop
 from vllm.distributed.kv_events import ZmqEventPublisher
-from vllm.inputs.data import TokensPrompt
 from vllm.usage.usage_lib import UsageContext
 from vllm.utils.argparse_utils import FlexibleArgumentParser
 from vllm.v1.engine.async_llm import AsyncLLM
@@ -40,6 +39,12 @@
 from utils.model import construct_mm_data
 from utils.protocol import MyRequestOutput, vLLMMultimodalRequest
 
+# Try importing from new vLLM (https://github.com/vllm-project/vllm/pull/35182), fallback to old structure
+try:
+    from vllm.inputs.llm import TokensPrompt
+except ImportError:
+    from vllm.inputs.data import TokensPrompt
+
 configure_dynamo_logging()
 logger = logging.getLogger(__name__)
 

examples/multimodal/utils/chat_processor.py

@@ -27,11 +27,16 @@
 from vllm.entrypoints.openai.engine.protocol import RequestResponseMetadata
 from vllm.entrypoints.openai.models.protocol import BaseModelPath
 from vllm.entrypoints.openai.models.serving import OpenAIServingModels
-from vllm.inputs.data import TokensPrompt
 from vllm.renderers.registry import renderer_from_config
 from vllm.sampling_params import SamplingParams
 from vllm.tokenizers import TokenizerLike as AnyTokenizer
 
+# Try importing from new vLLM (https://github.com/vllm-project/vllm/pull/35182), fallback to old structure
+try:
+    from vllm.inputs.llm import TokensPrompt
+except ImportError:
+    from vllm.inputs.data import TokensPrompt
+
 
 class StubEngineClient:
     """

examples/multimodal/utils/protocol.py

@@ -21,15 +21,21 @@
 from pydantic import BaseModel, ConfigDict, Field, field_serializer, field_validator
 from pydantic_core import core_schema
 from typing_extensions import NotRequired
-from vllm.inputs.data import TokensPrompt
 from vllm.logprobs import PromptLogprobs
-from vllm.multimodal.inputs import MultiModalUUIDDict  # noqa: F401
 from vllm.outputs import CompletionOutput
 from vllm.sampling_params import SamplingParams
 from vllm.v1.metrics.stats import RequestStateStats
 
 import dynamo.nixl_connect as connect
 
+# Try importing from new vLLM (https://github.com/vllm-project/vllm/pull/35182), fallback to old structure
+try:
+    from vllm.inputs.llm import MultiModalUUIDDict  # noqa: F401
+    from vllm.inputs.llm import TokensPrompt
+except ImportError:
+    from vllm.inputs.data import TokensPrompt
+    from vllm.multimodal.inputs import MultiModalUUIDDict  # noqa: F401
+
 
 class Request(BaseModel):
     prompt: str