[Core] Rename `PromptInputs` and `inputs`(#8673)

benchmarks/benchmark_latency.py

@@ -11,7 +11,7 @@ from tqdm import tqdm
 
 from vllm import LLM, SamplingParams
 from vllm.engine.arg_utils import DEVICE_OPTIONS, EngineArgs
-from vllm.inputs import PromptInputs
+from vllm.inputs import PromptType
 from vllm.model_executor.layers.quantization import QUANTIZATION_METHODS
 from vllm.utils import FlexibleArgumentParser
 
@@ -61,7 +61,7 @@ def main(args: argparse.Namespace):
     dummy_prompt_token_ids = np.random.randint(10000,
                                                size=(args.batch_size,
                                                      args.input_len))
-    dummy_inputs: List[PromptInputs] = [{
+    dummy_prompts: List[PromptType] = [{
         "prompt_token_ids": batch
     } for batch in dummy_prompt_token_ids.tolist()]
 
@@ -74,13 +74,13 @@ def main(args: argparse.Namespace):
                     ],
                     on_trace_ready=torch.profiler.tensorboard_trace_handler(
                         str(profile_dir))) as p:
-                llm.generate(dummy_inputs,
+                llm.generate(dummy_prompts,
                              sampling_params=sampling_params,
                              use_tqdm=False)
             print(p.key_averages())
         else:
             start_time = time.perf_counter()
-            llm.generate(dummy_inputs,
+            llm.generate(dummy_prompts,
                          sampling_params=sampling_params,
                          use_tqdm=False)
             end_time = time.perf_counter()

docs/source/dev/multimodal/multimodal_index.rst

@@ -8,7 +8,7 @@ Multi-Modality
 vLLM provides experimental support for multi-modal models through the :mod:`vllm.multimodal` package.
 
 Multi-modal inputs can be passed alongside text and token prompts to :ref:`supported models <supported_vlms>`
-via the ``multi_modal_data`` field in :class:`vllm.inputs.PromptInputs`.
+via the ``multi_modal_data`` field in :class:`vllm.inputs.PromptType`.
 
 Currently, vLLM only has built-in support for image data. You can extend vLLM to process additional modalities
 by following :ref:`this guide <adding_multimodal_plugin>`.

docs/source/dev/offline_inference/llm_inputs.rst

 LLM Inputs
 ==========
 
-.. autodata:: vllm.inputs.PromptInputs
+.. autodata:: vllm.inputs.PromptType
 
 .. autoclass:: vllm.inputs.TextPrompt
     :show-inheritance:

docs/source/models/vlm.rst

@@ -27,7 +27,7 @@ The :class:`~vllm.LLM` class can be instantiated in much the same way as languag
     We have removed all vision language related CLI args in the ``0.5.1`` release. **This is a breaking change**, so please update your code to follow
     the above snippet. Specifically, ``image_feature_size`` can no longer be specified as we now calculate that internally for each model.
 
-To pass an image to the model, note the following in :class:`vllm.inputs.PromptInputs`:
+To pass an image to the model, note the following in :class:`vllm.inputs.PromptType`:
 
 * ``prompt``: The prompt should follow the format that is documented on HuggingFace.
 * ``multi_modal_data``: This is a dictionary that follows the schema defined in :class:`vllm.multimodal.MultiModalDataDict`. 

tests/mq_llm_engine/test_error_handling.py

@@ -61,7 +61,7 @@ async def test_evil_forward(tmp_socket):
 
         # Throws an error in first forward pass.
         with pytest.raises(RAISED_ERROR):
-            async for _ in client.generate(inputs="Hello my name is",
+            async for _ in client.generate(prompt="Hello my name is",
                                            sampling_params=SamplingParams(),
                                            request_id=uuid.uuid4()):
                 pass
@@ -69,7 +69,7 @@ async def test_evil_forward(tmp_socket):
 
         # Engine is errored, should get ENGINE_DEAD_ERROR.
         with pytest.raises(MQEngineDeadError):
-            async for _ in client.generate(inputs="Hello my name is",
+            async for _ in client.generate(prompt="Hello my name is",
                                            sampling_params=SamplingParams(),
                                            request_id=uuid.uuid4()):
                 pass
@@ -118,7 +118,7 @@ async def test_failed_health_check(tmp_socket):
 
         # Generate call should throw ENGINE_DEAD_ERROR
         with pytest.raises(MQEngineDeadError):
-            async for _ in client.generate(inputs="Hello my name is",
+            async for _ in client.generate(prompt="Hello my name is",
                                            sampling_params=SamplingParams(),
                                            request_id=uuid.uuid4()):
                 pass
@@ -165,7 +165,7 @@ async def test_failed_abort(tmp_socket):
         # with reference to the original KeyError("foo")
         with pytest.raises(MQEngineDeadError) as execinfo:
             async for _ in client.generate(
-                    inputs="Hello my name is",
+                    prompt="Hello my name is",
                     sampling_params=SamplingParams(max_tokens=2000),
                     request_id=uuid.uuid4()):
                 pass
@@ -190,7 +190,7 @@ async def test_bad_request(tmp_socket):
 
         # Invalid request should fail, but not crash the server.
         with pytest.raises(ValueError):
-            async for _ in client.generate(inputs="Hello my name is",
+            async for _ in client.generate(prompt="Hello my name is",
                                            sampling_params=SamplingParams(),
                                            request_id="abcd-1",
                                            lora_request=LoRARequest(
@@ -199,7 +199,7 @@ async def test_bad_request(tmp_socket):
                 pass
 
         # This request should be okay.
-        async for _ in client.generate(inputs="Hello my name is",
+        async for _ in client.generate(prompt="Hello my name is",
                                        sampling_params=SamplingParams(),
                                        request_id="abcd-2"):
             pass

tests/mq_llm_engine/utils.py

@@ -20,7 +20,7 @@ async def generate(
     count = 0
     async for out in client.generate(
             request_id=request_id,
-            inputs="Hello my name is Robert and",
+            prompt="Hello my name is Robert and",
             sampling_params=SamplingParams(max_tokens=num_tokens,
                                            temperature=0)):
 

vllm/__init__.py

@@ -5,7 +5,7 @@ from vllm.engine.async_llm_engine import AsyncLLMEngine
 from vllm.engine.llm_engine import LLMEngine
 from vllm.entrypoints.llm import LLM
 from vllm.executor.ray_utils import initialize_ray_cluster
-from vllm.inputs import PromptInputs, TextPrompt, TokensPrompt
+from vllm.inputs import PromptType, TextPrompt, TokensPrompt
 from vllm.model_executor.models import ModelRegistry
 from vllm.outputs import (CompletionOutput, EmbeddingOutput,
                           EmbeddingRequestOutput, RequestOutput)
@@ -19,7 +19,7 @@ __all__ = [
     "__version__",
     "LLM",
     "ModelRegistry",
-    "PromptInputs",
+    "PromptType",
     "TextPrompt",
     "TokensPrompt",
     "SamplingParams",

vllm/engine/async_llm_engine.py

@@ -17,7 +17,7 @@ from vllm.engine.metrics_types import StatLoggerBase
 from vllm.executor.executor_base import ExecutorAsyncBase
 from vllm.executor.gpu_executor import GPUExecutorAsync
 from vllm.executor.ray_utils import initialize_ray_cluster
-from vllm.inputs import PromptInputs
+from vllm.inputs import PromptType
 from vllm.logger import init_logger
 from vllm.lora.request import LoRARequest
 from vllm.model_executor.layers.sampler import SamplerOutput
@@ -405,7 +405,7 @@ class _AsyncLLMEngine(LLMEngine):
     async def add_request_async(
         self,
         request_id: str,
-        inputs: PromptInputs,
+        prompt: PromptType,
         params: Union[SamplingParams, PoolingParams],
         arrival_time: Optional[float] = None,
         lora_request: Optional[LoRARequest] = None,
@@ -420,7 +420,7 @@ class _AsyncLLMEngine(LLMEngine):
             arrival_time = time.time()
 
         preprocessed_inputs = await self.input_preprocessor.preprocess_async(
-            inputs,
+            prompt,
             request_id=request_id,
             lora_request=lora_request,
             prompt_adapter_request=prompt_adapter_request,
@@ -777,7 +777,7 @@ class AsyncLLMEngine:
     async def add_request(
         self,
         request_id: str,
-        inputs: PromptInputs,
+        prompt: PromptType,
         params: Union[SamplingParams, PoolingParams],
         arrival_time: Optional[float] = None,
         lora_request: Optional[LoRARequest] = None,
@@ -797,7 +797,7 @@ class AsyncLLMEngine:
         stream = self._request_tracker.add_request(
             request_id,
             verbose=self.log_requests,
-            inputs=inputs,
+            prompt=prompt,
             params=params,
             arrival_time=arrival_time or time.time(),
             lora_request=lora_request,
@@ -808,7 +808,7 @@ class AsyncLLMEngine:
 
     async def generate(
         self,
-        inputs: PromptInputs,
+        prompt: PromptType,
         sampling_params: SamplingParams,
         request_id: str,
         lora_request: Optional[LoRARequest] = None,
@@ -822,8 +822,7 @@ class AsyncLLMEngine:
         from the LLMEngine to the caller.
 
         Args:
-            inputs: The inputs to the LLM. See
-                :class:`~vllm.inputs.PromptInputs`
+            prompt: The prompt to the LLM. See :class:`~vllm.inputs.PromptType`
                 for more details about the format of each input.
             sampling_params: The sampling parameters of the request.
             request_id: The unique id of the request.
@@ -881,7 +880,7 @@ class AsyncLLMEngine:
         """
         async for output in await self.add_request(
                 request_id,
-                inputs,
+                prompt,
                 sampling_params,
                 lora_request=lora_request,
                 trace_headers=trace_headers,
@@ -891,7 +890,7 @@ class AsyncLLMEngine:
 
     async def encode(
         self,
-        inputs: PromptInputs,
+        prompt: PromptType,
         pooling_params: PoolingParams,
         request_id: str,
         lora_request: Optional[LoRARequest] = None,
@@ -904,8 +903,7 @@ class AsyncLLMEngine:
         from the LLMEngine to the caller.
 
         Args:
-            inputs: The inputs to the LLM. See
-                :class:`~vllm.inputs.PromptInputs`
+            prompt: The prompt to the LLM. See :class:`~vllm.inputs.PromptType`
                 for more details about the format of each input.
             pooling_params: The pooling parameters of the request.
             request_id: The unique id of the request.
@@ -959,7 +957,7 @@ class AsyncLLMEngine:
         """
         async for output in await self.add_request(
                 request_id,
-                inputs,
+                prompt,
                 pooling_params,
                 lora_request=lora_request,
                 trace_headers=trace_headers,

vllm/engine/llm_engine.py

@@ -29,7 +29,7 @@ from vllm.executor.executor_base import ExecutorBase
 from vllm.executor.gpu_executor import GPUExecutor
 from vllm.executor.ray_utils import initialize_ray_cluster
 from vllm.inputs import (INPUT_REGISTRY, EncoderDecoderLLMInputs,
-                         InputRegistry, LLMInputs, PromptInputs)
+                         InputRegistry, LLMInputs, PromptType)
 from vllm.inputs.preprocess import InputPreprocessor
 from vllm.logger import init_logger
 from vllm.lora.request import LoRARequest
@@ -680,7 +680,7 @@ class LLMEngine:
     def add_request(
         self,
         request_id: str,
-        inputs: PromptInputs,
+        prompt: PromptType,
         params: Union[SamplingParams, PoolingParams],
         arrival_time: Optional[float] = None,
         lora_request: Optional[LoRARequest] = None,
@@ -695,8 +695,7 @@ class LLMEngine:
 
         Args:
             request_id: The unique ID of the request.
-            inputs: The inputs to the LLM. See
-                :class:`~vllm.inputs.PromptInputs`
+            prompt: The prompt to the LLM. See :class:`~vllm.inputs.PromptType`
                 for more details about the format of each input.
             params: Parameters for sampling or pooling.
                 :class:`~vllm.SamplingParams` for text generation.
@@ -736,7 +735,7 @@ class LLMEngine:
             arrival_time = time.time()
 
         preprocessed_inputs = self.input_preprocessor.preprocess(
-            inputs,
+            prompt,
             request_id=request_id,
             lora_request=lora_request,
             prompt_adapter_request=prompt_adapter_request,

vllm/engine/multiprocessing/__init__.py

@@ -3,7 +3,7 @@ from enum import Enum
 from typing import List, Mapping, Optional, Union
 
 from vllm import PoolingParams
-from vllm.inputs import PromptInputs
+from vllm.inputs import PromptType
 from vllm.lora.request import LoRARequest
 from vllm.outputs import RequestOutput
 from vllm.prompt_adapter.request import PromptAdapterRequest
@@ -23,7 +23,7 @@ class MQEngineDeadError(RuntimeError):
 
 @dataclass
 class RPCProcessRequest:
-    inputs: PromptInputs
+    prompt: PromptType
     params: Union[SamplingParams, PoolingParams]
     request_id: str
     lora_request: Optional[LoRARequest] = None

vllm/engine/multiprocessing/client.py

@@ -25,7 +25,7 @@ from vllm.engine.multiprocessing import (ENGINE_DEAD_ERROR, IPC_DATA_EXT,
                                          RPCStartupResponse)
 # yapf: enable
 from vllm.envs import VLLM_RPC_TIMEOUT
-from vllm.inputs import PromptInputs
+from vllm.inputs import PromptType
 from vllm.logger import init_logger
 from vllm.lora.request import LoRARequest
 from vllm.outputs import EmbeddingRequestOutput, RequestOutput
@@ -375,7 +375,7 @@ class MQLLMEngineClient:
 
     def generate(
         self,
-        inputs: PromptInputs,
+        prompt: PromptType,
         sampling_params: SamplingParams,
         request_id: str,
         lora_request: Optional[LoRARequest] = None,
@@ -389,8 +389,7 @@ class MQLLMEngineClient:
         from the LLMEngine to the caller.
 
         Args:
-            inputs: The inputs to the LLM. See
-                :class:`~vllm.inputs.PromptInputs`
+            prompt: The prompt to the LLM. See :class:`~vllm.inputs.PromptType`
                 for more details about the format of each input.
             sampling_params: The sampling parameters of the request.
             request_id: The unique id of the request.
@@ -399,13 +398,13 @@ class MQLLMEngineClient:
             prompt_adapter_request: Prompt Adapter request to use
                                             for generation, if any.
         """
-        return self._process_request(inputs, sampling_params, request_id,
+        return self._process_request(prompt, sampling_params, request_id,
                                      lora_request, trace_headers,
                                      prompt_adapter_request)
 
     def encode(
         self,
-        inputs: PromptInputs,
+        prompt: PromptType,
         pooling_params: PoolingParams,
         request_id: str,
         lora_request: Optional[LoRARequest] = None,
@@ -418,8 +417,7 @@ class MQLLMEngineClient:
         from the LLMEngine to the caller.
 
         Args:
-            inputs: The inputs to the LLM. See
-                :class:`~vllm.inputs.PromptInputs`
+            prompt: The prompt to the LLM. See :class:`~vllm.inputs.PromptType`
                 for more details about the format of each input.
             pooling_params: The pooling parameters of the request.
             request_id: The unique id of the request.
@@ -430,12 +428,12 @@ class MQLLMEngineClient:
             The output `EmbeddingRequestOutput` objects from the LLMEngine
             for the request.
         """
-        return self._process_request(inputs, pooling_params, request_id,
+        return self._process_request(prompt, pooling_params, request_id,
                                      lora_request, trace_headers)
 
     async def _process_request(
         self,
-        inputs: PromptInputs,
+        prompt: PromptType,
         params: Union[SamplingParams, PoolingParams],
         request_id: str,
         lora_request: Optional[LoRARequest] = None,
@@ -468,7 +466,7 @@ class MQLLMEngineClient:
 
             request_bytes = pickle.dumps(
                 RPCProcessRequest(
-                    inputs=inputs,
+                    prompt=prompt,
                     params=params,
                     request_id=request_id,
                     lora_request=lora_request,

vllm/engine/multiprocessing/engine.py

@@ -245,7 +245,7 @@ class MQLLMEngine:
         try:
             self.engine.add_request(
                 request_id=request_id,
-                inputs=request.inputs,
+                prompt=request.prompt,
                 params=request.params,
                 lora_request=request.lora_request,
                 trace_headers=request.trace_headers,

vllm/engine/protocol.py

@@ -3,7 +3,7 @@ from typing import (AsyncGenerator, List, Mapping, Optional, Protocol,
 
 from vllm.config import DecodingConfig, ModelConfig
 from vllm.core.scheduler import SchedulerOutputs
-from vllm.inputs.data import PromptInputs
+from vllm.inputs.data import PromptType
 from vllm.lora.request import LoRARequest
 from vllm.model_executor.layers.sampler import SamplerOutput
 from vllm.outputs import EmbeddingRequestOutput, RequestOutput
@@ -35,19 +35,19 @@ class EngineClient(Protocol):
 
     def generate(
         self,
-        inputs: PromptInputs,
+        prompt: PromptType,
         sampling_params: SamplingParams,
         request_id: str,
         lora_request: Optional[LoRARequest] = None,
         trace_headers: Optional[Mapping[str, str]] = None,
         prompt_adapter_request: Optional[PromptAdapterRequest] = None
     ) -> AsyncGenerator[RequestOutput, None]:
-        """Generates outputs for a request"""
+        """Generate outputs for a request."""
         ...
 
     def encode(
         self,
-        inputs: PromptInputs,
+        prompt: PromptType,
         pooling_params: PoolingParams,
         request_id: str,
         lora_request: Optional[LoRARequest] = None,

vllm/entrypoints/llm.py

@@ -10,7 +10,7 @@ from vllm.entrypoints.chat_utils import (ChatCompletionMessageParam,
                                          apply_hf_chat_template,
                                          apply_mistral_chat_template,
                                          parse_chat_messages)
-from vllm.inputs import PromptInputs, TextPrompt, TokensPrompt
+from vllm.inputs import PromptType, TextPrompt, TokensPrompt
 from vllm.inputs.parse import parse_and_batch_prompt
 from vllm.logger import init_logger
 from vllm.lora.request import LoRARequest
@@ -258,8 +258,8 @@ class LLM:
     @overload
     def generate(
         self,
-        inputs: Union[PromptInputs, Sequence[PromptInputs]],
-        /,  # We may enable `inputs` keyword after removing the old API
+        prompts: Union[PromptType, Sequence[PromptType]],
+        /,
         *,
         sampling_params: Optional[Union[SamplingParams,
                                         Sequence[SamplingParams]]] = None,
@@ -276,7 +276,7 @@ class LLM:
     )
     def generate(
         self,
-        prompts: Union[Union[PromptInputs, Sequence[PromptInputs]],
+        prompts: Union[Union[PromptType, Sequence[PromptType]],
                        Optional[Union[str, List[str]]]] = None,
         sampling_params: Optional[Union[SamplingParams,
                                         Sequence[SamplingParams]]] = None,
@@ -294,7 +294,9 @@ class LLM:
         into a single list and pass it to this method.
 
         Args:
-            inputs: A list of inputs to generate completions for.
+            prompts: The prompts to the LLM. You may pass a sequence of prompts
+                for batch inference. See :class:`~vllm.inputs.PromptType`
+                for more details about the format of each prompts.
             sampling_params: The sampling parameters for text generation. If
                 None, we use the default sampling parameters.
                 When it is a single value, it is applied to every prompt.
@@ -320,12 +322,13 @@ class LLM:
                 "models (XForCausalLM, XForConditionalGeneration).")
 
         if prompt_token_ids is not None:
-            inputs = self._convert_v1_inputs(
+            parsed_prompts = self._convert_v1_inputs(
                 prompts=cast(Optional[Union[str, List[str]]], prompts),
                 prompt_token_ids=prompt_token_ids,
             )
         else:
-            inputs = cast(Union[PromptInputs, Sequence[PromptInputs]], prompts)
+            parsed_prompts = cast(Union[PromptType, Sequence[PromptType]],
+                                  prompts)
 
         if isinstance(guided_options_request, dict):
             if len(guided_options_request) > 1:
@@ -340,7 +343,7 @@ class LLM:
             sampling_params = SamplingParams()
 
         self._validate_and_add_requests(
-            inputs=inputs,
+            prompts=parsed_prompts,
             params=sampling_params,
             lora_request=lora_request,
             prompt_adapter_request=prompt_adapter_request,
@@ -396,9 +399,9 @@ class LLM:
         conversation, mm_data = parse_chat_messages(messages, model_config,
                                                     tokenizer)
 
-        prompt: Union[str, List[int]]
+        prompt_data: Union[str, List[int]]
         if isinstance(tokenizer, MistralTokenizer):
-            prompt = apply_mistral_chat_template(
+            prompt_data = apply_mistral_chat_template(
                 tokenizer,
                 messages=messages,
                 chat_template=chat_template,
@@ -406,7 +409,7 @@ class LLM:
                 tools=tools,
             )
         else:
-            prompt = apply_hf_chat_template(
+            prompt_data = apply_hf_chat_template(
                 tokenizer,
                 conversation=conversation,
                 chat_template=chat_template,
@@ -414,17 +417,17 @@ class LLM:
                 tools=tools,
             )
 
-        inputs: PromptInputs
-        if is_list_of(prompt, int):
-            inputs = TokensPrompt(prompt_token_ids=prompt)
+        prompt: PromptType
+        if is_list_of(prompt_data, int):
+            prompt = TokensPrompt(prompt_token_ids=prompt_data)
         else:
-            inputs = TextPrompt(prompt=prompt)
+            prompt = TextPrompt(prompt=prompt_data)
 
         if mm_data is not None:
-            inputs["multi_modal_data"] = mm_data
+            prompt["multi_modal_data"] = mm_data
 
         return self.generate(
-            inputs,
+            prompt,
             sampling_params=sampling_params,
             use_tqdm=use_tqdm,
             lora_request=lora_request,
@@ -494,8 +497,8 @@ class LLM:
     @overload
     def encode(
         self,
-        inputs: Union[PromptInputs, Sequence[PromptInputs]],
-        /,  # We may enable `inputs` keyword after removing the old API
+        prompts: Union[PromptType, Sequence[PromptType]],
+        /,
         *,
         pooling_params: Optional[Union[PoolingParams,
                                        Sequence[PoolingParams]]] = None,
@@ -512,7 +515,7 @@ class LLM:
     )
     def encode(
         self,
-        prompts: Union[Union[PromptInputs, Sequence[PromptInputs]],
+        prompts: Union[Union[PromptType, Sequence[PromptType]],
                        Optional[Union[str, List[str]]]] = None,
         pooling_params: Optional[Union[PoolingParams,
                                        Sequence[PoolingParams]]] = None,
@@ -528,9 +531,9 @@ class LLM:
         into a single list and pass it to this method.
 
         Args:
-            inputs: The inputs to the LLM. You may pass a sequence of inputs for
-                batch inference. See :class:`~vllm.inputs.PromptInputs`
-                for more details about the format of each input.
+            prompts: The prompts to the LLM. You may pass a sequence of prompts
+                for batch inference. See :class:`~vllm.inputs.PromptType`
+                for more details about the format of each prompts.
             pooling_params: The pooling parameters for pooling. If None, we
                 use the default pooling parameters.
             use_tqdm: Whether to use tqdm to display the progress bar.
@@ -553,19 +556,20 @@ class LLM:
             )
 
         if prompt_token_ids is not None:
-            inputs = self._convert_v1_inputs(
+            parsed_prompts = self._convert_v1_inputs(
                 prompts=cast(Optional[Union[str, List[str]]], prompts),
                 prompt_token_ids=prompt_token_ids,
             )
         else:
-            inputs = cast(Union[PromptInputs, Sequence[PromptInputs]], prompts)
+            parsed_prompts = cast(Union[PromptType, Sequence[PromptType]],
+                                  prompts)
 
         if pooling_params is None:
             # Use default pooling params.
             pooling_params = PoolingParams()
 
         self._validate_and_add_requests(
-            inputs=inputs,
+            prompts=parsed_prompts,
             params=pooling_params,
             lora_request=lora_request,
             prompt_adapter_request=prompt_adapter_request,
@@ -609,9 +613,9 @@ class LLM:
             raise ValueError("Either prompts or prompt_token_ids must be "
                              "provided.")
 
-        inputs: List[PromptInputs] = []
+        parsed_prompts: List[PromptType] = []
         for i in range(num_requests):
-            item: PromptInputs
+            item: PromptType
 
             if prompts is not None:
                 item = TextPrompt(prompt=prompts[i])
@@ -620,24 +624,24 @@ class LLM:
             else:
                 raise AssertionError
 
-            inputs.append(item)
+            parsed_prompts.append(item)
 
-        return inputs
+        return parsed_prompts
 
     def _validate_and_add_requests(
         self,
-        inputs: Union[PromptInputs, Sequence[PromptInputs]],
+        prompts: Union[PromptType, Sequence[PromptType]],
         params: Union[SamplingParams, Sequence[SamplingParams], PoolingParams,
                       Sequence[PoolingParams]],
         lora_request: Optional[Union[Sequence[LoRARequest], LoRARequest]],
         prompt_adapter_request: Optional[PromptAdapterRequest],
         guided_options: Optional[GuidedDecodingRequest] = None,
     ) -> None:
-        if isinstance(inputs, (str, dict)):
+        if isinstance(prompts, (str, dict)):
             # Convert a single prompt to a list.
-            inputs = [inputs]
+            prompts = [prompts]
 
-        num_requests = len(inputs)
+        num_requests = len(prompts)
         if isinstance(params, list) and len(params) != num_requests:
             raise ValueError("The lengths of prompts and params "
                              "must be the same.")
@@ -654,9 +658,9 @@ class LLM:
                 sp.output_kind = RequestOutputKind.FINAL_ONLY
 
         # Add requests to the engine.
-        for i, request_inputs in enumerate(inputs):
+        for i, prompt in enumerate(prompts):
             self._add_request(
-                request_inputs,
+                prompt,
                 params[i] if isinstance(params, Sequence) else params,
                 lora_request=lora_request[i] if isinstance(
                     lora_request, Sequence) else lora_request,
@@ -665,7 +669,7 @@ class LLM:
 
     def _add_request(
         self,
-        inputs: PromptInputs,
+        prompt: PromptType,
         params: Union[SamplingParams, PoolingParams],
         lora_request: Optional[LoRARequest] = None,
         prompt_adapter_request: Optional[PromptAdapterRequest] = None,
@@ -673,7 +677,7 @@ class LLM:
         request_id = str(next(self.request_counter))
         self.llm_engine.add_request(
             request_id,
-            inputs,
+            prompt,
             params,
             lora_request=lora_request,
             prompt_adapter_request=prompt_adapter_request,

vllm/inputs/__init__.py

 from .data import (EncoderDecoderLLMInputs, ExplicitEncoderDecoderPrompt,
-                   LLMInputs, PromptInputs, SingletonPromptInputs, TextPrompt,
+                   LLMInputs, PromptType, SingletonPrompt, TextPrompt,
                    TokensPrompt, build_explicit_enc_dec_prompt,
                    to_enc_dec_tuple_list, zip_enc_dec_prompts)
 from .registry import InputContext, InputRegistry
@@ -16,8 +16,8 @@ See also:
 __all__ = [
     "TextPrompt",
     "TokensPrompt",
-    "PromptInputs",
-    "SingletonPromptInputs",
+    "PromptType",
+    "SingletonPrompt",
     "ExplicitEncoderDecoderPrompt",
     "LLMInputs",
     "EncoderDecoderLLMInputs",

vllm/inputs/data.py

@@ -33,7 +33,7 @@ class TokensPrompt(TypedDict):
     """
 
 
-SingletonPromptInputs = Union[str, TextPrompt, TokensPrompt]
+SingletonPrompt = Union[str, TextPrompt, TokensPrompt]
 """
 Set of possible schemas for a single LLM input:
 
@@ -46,7 +46,7 @@ which may be utilized for encoder/decoder models when
 the user desires to express both the encoder & decoder
 prompts explicitly, i.e. :class:`ExplicitEncoderDecoderPrompt`
 
-A prompt of type :class:`SingletonPromptInputs` may be employed
+A prompt of type :class:`SingletonPromptType` may be employed
 as (1) input to a decoder-only model, (2) input to
 the encoder of an encoder/decoder model, in the scenario
 where the decoder-prompt is not specified explicitly, or
@@ -55,12 +55,12 @@ more than one prompt, i.e. :class:`ExplicitEncoderDecoderPrompt`
 """
 
 _T1_co = TypeVar("_T1_co",
-                 bound=SingletonPromptInputs,
-                 default=SingletonPromptInputs,
+                 bound=SingletonPrompt,
+                 default=SingletonPrompt,
                  covariant=True)
 _T2_co = TypeVar("_T2_co",
-                 bound=SingletonPromptInputs,
-                 default=SingletonPromptInputs,
+                 bound=SingletonPrompt,
+                 default=SingletonPrompt,
                  covariant=True)
 
 
@@ -72,7 +72,7 @@ class ExplicitEncoderDecoderPrompt(TypedDict, Generic[_T1_co, _T2_co]):
 
     The encoder and decoder prompts, respectively,
     may formatted according to any of the
-    :class:`SingletonPromptInputs` schemas, and are not
+    :class:`SingletonPromptType` schemas, and are not
     required to have the same schema.
 
     Only the encoder prompt may have multi-modal data.
@@ -81,7 +81,7 @@ class ExplicitEncoderDecoderPrompt(TypedDict, Generic[_T1_co, _T2_co]):
     be used as an input to a decoder-only model,
     and that the `encoder_prompt` and `decoder_prompt`
     fields of this data structure themselves must be
-    :class:`SingletonPromptInputs` instances.
+    :class:`SingletonPromptType` instances.
     """
 
     encoder_prompt: _T1_co
@@ -89,7 +89,7 @@ class ExplicitEncoderDecoderPrompt(TypedDict, Generic[_T1_co, _T2_co]):
     decoder_prompt: Optional[_T2_co]
 
 
-PromptInputs = Union[SingletonPromptInputs, ExplicitEncoderDecoderPrompt]
+PromptType = Union[SingletonPrompt, ExplicitEncoderDecoderPrompt]
 """
 Set of possible schemas for an LLM input, including
 both decoder-only and encoder/decoder input types:
@@ -140,12 +140,8 @@ class EncoderDecoderLLMInputs(LLMInputs):
     """
 
 
-_T1 = TypeVar("_T1",
-              bound=SingletonPromptInputs,
-              default=SingletonPromptInputs)
-_T2 = TypeVar("_T2",
-              bound=SingletonPromptInputs,
-              default=SingletonPromptInputs)
+_T1 = TypeVar("_T1", bound=SingletonPrompt, default=SingletonPrompt)
+_T2 = TypeVar("_T2", bound=SingletonPrompt, default=SingletonPrompt)
 
 
 def build_explicit_enc_dec_prompt(

vllm/inputs/parse.py

@@ -5,7 +5,7 @@ from typing_extensions import TypeIs
 from vllm.utils import is_list_of
 
 from .data import (EncoderDecoderLLMInputs, ExplicitEncoderDecoderPrompt,
-                   LLMInputs, PromptInputs, SingletonPromptInputs, TextPrompt,
+                   LLMInputs, PromptType, SingletonPrompt, TextPrompt,
                    TokensPrompt)
 
 
@@ -81,23 +81,23 @@ class ParsedTokensPrompt(TypedDict):
 
 
 def parse_singleton_prompt(
-    inputs: SingletonPromptInputs,
+    prompt: SingletonPrompt,
 ) -> Union[ParsedStrPrompt, ParsedTextPrompt, ParsedTokensPrompt]:
-    if isinstance(inputs, str):
-        return ParsedStrPrompt(type="str", content=inputs)
-    elif isinstance(inputs, dict):
-        if "prompt_token_ids" in inputs:
+    if isinstance(prompt, str):
+        return ParsedStrPrompt(type="str", content=prompt)
+    elif isinstance(prompt, dict):
+        if "prompt_token_ids" in prompt:
             return ParsedTokensPrompt(type="tokens",
-                                      content=inputs)  # type: ignore
-        elif "prompt" in inputs:
-            return ParsedTextPrompt(type="text", content=inputs)
+                                      content=prompt)  # type: ignore
+        elif "prompt" in prompt:
+            return ParsedTextPrompt(type="text", content=prompt)
 
     raise TypeError("inputs must be a string, TextPrompt, or TokensPrompt")
 
 
 def is_explicit_encoder_decoder_prompt(
-        inputs: PromptInputs) -> TypeIs[ExplicitEncoderDecoderPrompt]:
-    return isinstance(inputs, dict) and "encoder_prompt" in inputs
+        prompt: PromptType) -> TypeIs[ExplicitEncoderDecoderPrompt]:
+    return isinstance(prompt, dict) and "encoder_prompt" in prompt
 
 
 def is_valid_encoder_decoder_llm_inputs(

vllm/inputs/preprocess.py

@@ -9,8 +9,8 @@ from vllm.lora.request import LoRARequest
 from vllm.prompt_adapter.request import PromptAdapterRequest
 from vllm.transformers_utils.tokenizer_group import BaseTokenizerGroup
 
-from .data import (EncoderDecoderLLMInputs, LLMInputs, PromptInputs,
-                   SingletonPromptInputs)
+from .data import (EncoderDecoderLLMInputs, LLMInputs, PromptType,
+                   SingletonPrompt)
 from .parse import is_explicit_encoder_decoder_prompt, parse_singleton_prompt
 
 if TYPE_CHECKING:
@@ -206,7 +206,7 @@ class InputPreprocessor:
 
     def _extract_prompt_components(
         self,
-        inputs: SingletonPromptInputs,
+        prompt: SingletonPrompt,
         request_id: str,
         lora_request: Optional[LoRARequest] = None,
     ) -> PromptComponents:
@@ -216,7 +216,7 @@ class InputPreprocessor:
         Arguments:
 
         * request_id
-        * inputs: single encoder or decoder input prompt
+        * prompt: single encoder or decoder input prompt
         * lora_request: this is only valid for decoder prompts
 
         Returns:
@@ -226,24 +226,24 @@ class InputPreprocessor:
         * multi_modal_data
         '''
 
-        parsed = parse_singleton_prompt(inputs)
+        parsed = parse_singleton_prompt(prompt)
 
         if parsed["type"] == "str":
-            prompt = parsed["content"]
+            prompt_text = parsed["content"]
             prompt_token_ids = self._tokenize_prompt(
-                prompt,
+                prompt_text,
                 request_id=request_id,
                 lora_request=lora_request,
             )
             multi_modal_data = None
         elif parsed["type"] == "tokens":
-            prompt = None
+            prompt_text = None
             prompt_token_ids = parsed["content"]["prompt_token_ids"]
             multi_modal_data = parsed["content"].get("multi_modal_data")
         elif parsed["type"] == "text":
-            prompt = parsed["content"]["prompt"]
+            prompt_text = parsed["content"]["prompt"]
             prompt_token_ids = self._tokenize_prompt(
-                prompt,
+                prompt_text,
                 request_id=request_id,
                 lora_request=lora_request,
             )
@@ -251,33 +251,33 @@ class InputPreprocessor:
         else:
             assert_never(parsed)
 
-        return prompt, prompt_token_ids, multi_modal_data
+        return prompt_text, prompt_token_ids, multi_modal_data
 
     async def _extract_prompt_components_async(
         self,
-        inputs: SingletonPromptInputs,
+        prompt: SingletonPrompt,
         request_id: str,
         lora_request: Optional[LoRARequest] = None,
     ) -> PromptComponents:
         """Async version of :meth:`_extract_prompt_components`."""
-        parsed = parse_singleton_prompt(inputs)
+        parsed = parse_singleton_prompt(prompt)
 
         if parsed["type"] == "str":
-            prompt = parsed["content"]
+            prompt_text = parsed["content"]
             prompt_token_ids = await self._tokenize_prompt_async(
-                prompt,
+                prompt_text,
                 request_id=request_id,
                 lora_request=lora_request,
             )
             multi_modal_data = None
         elif parsed["type"] == "tokens":
-            prompt = None
+            prompt_text = None
             prompt_token_ids = parsed["content"]["prompt_token_ids"]
             multi_modal_data = parsed["content"].get("multi_modal_data")
         elif parsed["type"] == "text":
-            prompt = parsed["content"]["prompt"]
+            prompt_text = parsed["content"]["prompt"]
             prompt_token_ids = await self._tokenize_prompt_async(
-                prompt,
+                prompt_text,
                 request_id=request_id,
                 lora_request=lora_request,
             )
@@ -285,7 +285,7 @@ class InputPreprocessor:
         else:
             assert_never(parsed)
 
-        return prompt, prompt_token_ids, multi_modal_data
+        return prompt_text, prompt_token_ids, multi_modal_data
 
     def _build_enc_dec_llm_inputs(
         self,
@@ -311,7 +311,7 @@ class InputPreprocessor:
 
     def _process_encoder_decoder_prompt(
         self,
-        inputs: PromptInputs,
+        prompt: PromptType,
         request_id: str,
     ) -> EncoderDecoderLLMInputs:
         '''
@@ -339,7 +339,7 @@ class InputPreprocessor:
         
         Arguments:
 
-        * inputs: an input prompt
+        * prompt: an input prompt
         * request_id
 
         Returns:
@@ -350,13 +350,13 @@ class InputPreprocessor:
         encoder_comps: PromptComponents
         decoder_comps: DecoderPromptComponents
 
-        if is_explicit_encoder_decoder_prompt(inputs):
+        if is_explicit_encoder_decoder_prompt(prompt):
             encoder_comps = self._extract_prompt_components(
-                inputs["encoder_prompt"],
+                prompt["encoder_prompt"],
                 request_id=request_id,
             )
 
-            if (decoder_input := inputs["decoder_prompt"]) is None:
+            if (decoder_input := prompt["decoder_prompt"]) is None:
                 decoder_comps = None, None, None
             else:
                 decoder_comps = self._extract_prompt_components(
@@ -365,7 +365,7 @@ class InputPreprocessor:
                 )
         else:
             encoder_comps = self._extract_prompt_components(
-                inputs,
+                prompt,
                 request_id=request_id,
             )
 
@@ -375,20 +375,20 @@ class InputPreprocessor:
 
     async def _process_encoder_decoder_prompt_async(
         self,
-        inputs: PromptInputs,
+        prompt: PromptType,
         request_id: str,
     ) -> EncoderDecoderLLMInputs:
         """Async version of :meth:`_process_encoder_decoder_prompt`."""
         encoder_comps: PromptComponents
         decoder_comps: DecoderPromptComponents
 
-        if is_explicit_encoder_decoder_prompt(inputs):
+        if is_explicit_encoder_decoder_prompt(prompt):
             encoder_task = self._extract_prompt_components_async(
-                inputs["encoder_prompt"],
+                prompt["encoder_prompt"],
                 request_id=request_id,
             )
 
-            if (decoder_input := inputs["decoder_prompt"]) is None:
+            if (decoder_input := prompt["decoder_prompt"]) is None:
                 encoder_comps = await encoder_task
                 decoder_comps = None, None, None
             else:
@@ -401,7 +401,7 @@ class InputPreprocessor:
                     encoder_task, decoder_task)
         else:
             encoder_comps = await self._extract_prompt_components_async(
-                inputs,
+                prompt,
                 request_id=request_id,
             )
 
@@ -425,7 +425,7 @@ class InputPreprocessor:
 
     def _process_decoder_only_prompt(
         self,
-        inputs: SingletonPromptInputs,
+        prompt: SingletonPrompt,
         request_id: str,
         lora_request: Optional[LoRARequest] = None,
         prompt_adapter_request: Optional[PromptAdapterRequest] = None,
@@ -436,7 +436,7 @@ class InputPreprocessor:
 
         Arguments:
 
-        * inputs: input prompt
+        * prompt: input prompt
         * request_id
         * lora_request
         * prompt_adapter_request
@@ -447,7 +447,7 @@ class InputPreprocessor:
         '''
 
         prompt_comps = self._extract_prompt_components(
-            inputs,
+            prompt,
             request_id=request_id,
             lora_request=lora_request,
         )
@@ -459,14 +459,14 @@ class InputPreprocessor:
 
     async def _process_decoder_only_prompt_async(
         self,
-        inputs: SingletonPromptInputs,
+        prompt: SingletonPrompt,
         request_id: str,
         lora_request: Optional[LoRARequest] = None,
         prompt_adapter_request: Optional[PromptAdapterRequest] = None,
     ) -> LLMInputs:
         """Async version of :meth:`_process_decoder_only_prompt`."""
         prompt_comps = await self._extract_prompt_components_async(
-            inputs,
+            prompt,
             request_id=request_id,
             lora_request=lora_request,
         )
@@ -478,7 +478,7 @@ class InputPreprocessor:
 
     def preprocess(
         self,
-        inputs: PromptInputs,
+        prompt: PromptType,
         request_id: str,
         lora_request: Optional[LoRARequest] = None,
         prompt_adapter_request: Optional[PromptAdapterRequest] = None,
@@ -488,17 +488,17 @@ class InputPreprocessor:
             # Encoder-decoder model requires special mapping of
             # input prompts to encoder & decoder
             return self._process_encoder_decoder_prompt(
-                inputs,
+                prompt,
                 request_id=request_id,
             )
 
-        if is_explicit_encoder_decoder_prompt(inputs):
+        if is_explicit_encoder_decoder_prompt(prompt):
             raise ValueError("Cannot pass encoder-decoder prompt "
                              "to decoder-only models")
 
         # Decoder-only operation
         return self._process_decoder_only_prompt(
-            inputs,
+            prompt,
             request_id=request_id,
             lora_request=lora_request,
             prompt_adapter_request=prompt_adapter_request,
@@ -506,7 +506,7 @@ class InputPreprocessor:
 
     async def preprocess_async(
         self,
-        inputs: PromptInputs,
+        prompt: PromptType,
         request_id: str,
         lora_request: Optional[LoRARequest] = None,
         prompt_adapter_request: Optional[PromptAdapterRequest] = None,
@@ -516,17 +516,17 @@ class InputPreprocessor:
             # Encoder-decoder model requires special mapping of
             # input prompts to encoder & decoder
             return await self._process_encoder_decoder_prompt_async(
-                inputs,
+                prompt,
                 request_id=request_id,
             )
 
-        if is_explicit_encoder_decoder_prompt(inputs):
+        if is_explicit_encoder_decoder_prompt(prompt):
             raise ValueError("Cannot pass encoder-decoder prompt "
                              "to decoder-only models")
 
         # Decoder-only operation
         return await self._process_decoder_only_prompt_async(
-            inputs,
+            prompt,
             request_id=request_id,
             lora_request=lora_request,
             prompt_adapter_request=prompt_adapter_request,