[Frontend] Refactor prompt processing (#4028)

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

vllm/inputs/__init__.py

@@ -1,6 +1,5 @@
 from .data import (LLMInputs, ParsedText, ParsedTokens, PromptInputs,
-                   PromptStrictInputs, TextPrompt, TextTokensPrompt,
-                   TokensPrompt, parse_and_batch_prompt)
+                   TextPrompt, TokensPrompt, parse_and_batch_prompt)
 from .registry import InputContext, InputRegistry
 
 INPUT_REGISTRY = InputRegistry()
@@ -14,6 +13,6 @@ See also:
 
 __all__ = [
     "ParsedText", "ParsedTokens", "parse_and_batch_prompt", "TextPrompt",
-    "TokensPrompt", "TextTokensPrompt", "PromptStrictInputs", "PromptInputs",
-    "LLMInputs", "INPUT_REGISTRY", "InputContext", "InputRegistry"
+    "TokensPrompt", "PromptInputs", "LLMInputs", "INPUT_REGISTRY",
+    "InputContext", "InputRegistry"
 ]

vllm/inputs/data.py

@@ -92,25 +92,7 @@ class TokensPrompt(TypedDict):
     """
 
 
-class TextTokensPrompt(TypedDict):
-    """It is assumed that :attr:`prompt` is consistent with
-    :attr:`prompt_token_ids`. This is currently used in
-    :class:`AsyncLLMEngine` for logging both the text and token IDs."""
-
-    prompt: str
-    """The prompt text."""
-
-    prompt_token_ids: List[int]
-    """The token IDs of the prompt."""
-
-    multi_modal_data: NotRequired["MultiModalDataDict"]
-    """
-    Optional multi-modal data to pass to the model,
-    if the model supports it.
-    """
-
-
-PromptStrictInputs = Union[str, TextPrompt, TokensPrompt]
+PromptInputs = Union[str, TextPrompt, TokensPrompt]
 """
 The inputs to the LLM, which can take one of the following forms:
 
@@ -118,10 +100,6 @@ The inputs to the LLM, which can take one of the following forms:
 - A tokenized prompt (:class:`TokensPrompt`)
 """
 
-PromptInputs = Union[str, TextPrompt, TokensPrompt, TextTokensPrompt]
-"""Same as :const:`PromptStrictInputs` but additionally accepts
-:class:`TextTokensPrompt`."""
-
 
 class LLMInputs(TypedDict):
     """