[V1] Initial support of multimodal models for V1 re-arch (#10699)

Signed-off-by: Roger Wang <ywang@roblox.com>

vllm/v1/engine/llm_engine.py

@@ -1,5 +1,7 @@
 from typing import Dict, List, Mapping, Optional, Type, Union
 
+from typing_extensions import TypeVar
+
 from vllm.config import VllmConfig
 from vllm.engine.arg_utils import EngineArgs
 from vllm.engine.metrics_types import StatLoggerBase
@@ -12,7 +14,8 @@ from vllm.outputs import RequestOutput
 from vllm.pooling_params import PoolingParams
 from vllm.prompt_adapter.request import PromptAdapterRequest
 from vllm.sampling_params import SamplingParams
-from vllm.transformers_utils.tokenizer_group import init_tokenizer_from_configs
+from vllm.transformers_utils.tokenizer_group import (
+    BaseTokenizerGroup, init_tokenizer_from_configs)
 from vllm.usage.usage_lib import UsageContext
 from vllm.v1.engine.core_client import EngineCoreClient
 from vllm.v1.engine.detokenizer import Detokenizer
@@ -21,6 +24,8 @@ from vllm.v1.executor.gpu_executor import GPUExecutor
 
 logger = init_logger(__name__)
 
+_G = TypeVar("_G", bound=BaseTokenizerGroup, default=BaseTokenizerGroup)
+
 
 class LLMEngine:
     """Legacy LLMEngine for backwards compatibility."""
@@ -169,5 +174,18 @@ class LLMEngine:
     def stop_profile(self):
         self.engine_core.profile(False)
 
-    def get_tokenizer_group(self, group_type):
-        pass
+    def get_tokenizer_group(
+        self,
+        group_type: Type[_G] = BaseTokenizerGroup,
+    ) -> _G:
+        tokenizer_group = self.tokenizer
+
+        if tokenizer_group is None:
+            raise ValueError("Unable to get tokenizer because "
+                             "skip_tokenizer_init is True")
+        if not isinstance(tokenizer_group, group_type):
+            raise TypeError("Invalid type of tokenizer group. "
+                            f"Expected type: {group_type}, but "
+                            f"found type: {type(tokenizer_group)}")
+
+        return tokenizer_group