[Core] Registry for processing model inputs (#5214)

Co-authored-by: ywang96 <ywang@roblox.com>

vllm/sequence.py

@@ -8,12 +8,12 @@ from typing import TYPE_CHECKING, Dict, List, Optional, Tuple, Union
 
 import torch
 
-from vllm.inputs import LLMInputs
 from vllm.lora.request import LoRARequest
 from vllm.pooling_params import PoolingParams
 from vllm.sampling_params import SamplingParams
 
 if TYPE_CHECKING:
+    from vllm.inputs import LLMInputs
     from vllm.multimodal import MultiModalData
     from vllm.spec_decode.metrics import SpecDecodeWorkerMetrics
 
@@ -221,7 +221,7 @@ class Sequence:
     def __init__(
         self,
         seq_id: int,
-        inputs: LLMInputs,
+        inputs: "LLMInputs",
         block_size: int,
         eos_token_id: Optional[int] = None,
         lora_request: Optional[LoRARequest] = None,