[Misc] Reorganize inputs (#35182)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2026-03-25 10:22:54 -07:00
committed by GitHub
parent 678b3c99e8
commit ba2f0acc2d
142 changed files with 1212 additions and 1342 deletions
+10 -10
View File
@@ -18,7 +18,7 @@ from vllm.entrypoints.pooling.typing import (
PoolingCompletionLikeRequest,
PoolingServeContext,
)
from vllm.inputs.data import ProcessorInputs, SingletonPrompt
from vllm.inputs import EngineInput, SingletonPrompt
from vllm.renderers import BaseRenderer, merge_kwargs
from vllm.renderers.inputs.preprocess import parse_model_prompt, prompt_to_seq
from vllm.tool_parsers import ToolParser
@@ -60,7 +60,7 @@ class PoolingIOProcessor:
chat_template_kwargs=request.chat_template_kwargs,
trust_request_chat_template=self.trust_request_chat_template,
)
_, engine_prompts = self._preprocess_chat_online(
_, engine_inputs = self._preprocess_chat_online(
request,
request.messages,
default_template=self.chat_template,
@@ -68,7 +68,7 @@ class PoolingIOProcessor:
default_template_kwargs=None,
)
elif isinstance(request, PoolingCompletionLikeRequest):
engine_prompts = self._preprocess_completion_online(
engine_inputs = self._preprocess_completion_online(
request,
prompt_input=request.input,
prompt_embeds=None,
@@ -76,7 +76,7 @@ class PoolingIOProcessor:
else:
raise ValueError(f"Invalid {self.name} request type")
ctx.engine_prompts = engine_prompts
ctx.engine_inputs = engine_inputs
async def pre_process_online_async(self, ctx: PoolingServeContext):
self.pre_process_online(ctx)
@@ -100,7 +100,7 @@ class PoolingIOProcessor:
self,
prompts: PromptType | Sequence[PromptType],
tokenization_kwargs: dict[str, Any] | None = None,
) -> Sequence[ProcessorInputs]:
) -> Sequence[EngineInput]:
return self._preprocess_completion_offline(
prompts=prompts, tokenization_kwargs=tokenization_kwargs
)
@@ -128,7 +128,7 @@ class PoolingIOProcessor:
request: RendererRequest,
prompt_input: str | list[str] | list[int] | list[list[int]] | None,
prompt_embeds: bytes | list[bytes] | None,
) -> list[ProcessorInputs]:
) -> list[EngineInput]:
renderer = self.renderer
model_config = self.model_config
@@ -167,7 +167,7 @@ class PoolingIOProcessor:
default_template_kwargs: dict[str, Any] | None,
tool_dicts: list[dict[str, Any]] | None = None,
tool_parser: type[ToolParser] | None = None,
) -> tuple[list[ConversationMessage], list[ProcessorInputs]]:
) -> tuple[list[ConversationMessage], list[EngineInput]]:
renderer = self.renderer
default_template_kwargs = merge_kwargs(
@@ -188,7 +188,7 @@ class PoolingIOProcessor:
default_media_io_kwargs=(mm_config.media_io_kwargs if mm_config else None),
)
(conversation,), (engine_prompt,) = renderer.render_chat(
(conversation,), (engine_input,) = renderer.render_chat(
[messages],
chat_params,
tok_params,
@@ -199,13 +199,13 @@ class PoolingIOProcessor:
},
)
return conversation, [engine_prompt]
return conversation, [engine_input]
def _preprocess_completion_offline(
self,
prompts: PromptType | Sequence[PromptType],
tokenization_kwargs: dict[str, Any] | None = None,
) -> Sequence[ProcessorInputs]:
) -> Sequence[EngineInput]:
renderer = self.renderer
model_config = self.model_config
+8 -8
View File
@@ -20,7 +20,7 @@ from vllm.entrypoints.openai.engine.protocol import ErrorResponse
from vllm.entrypoints.openai.models.serving import OpenAIServingModels
from vllm.entrypoints.pooling.typing import AnyPoolingRequest, PoolingServeContext
from vllm.exceptions import VLLMNotFoundError
from vllm.inputs.data import ProcessorInputs
from vllm.inputs import EngineInput
from vllm.lora.request import LoRARequest
from vllm.renderers.base import BaseRenderer
from vllm.renderers.inputs.preprocess import extract_prompt_components
@@ -106,7 +106,7 @@ class PoolingServing:
self,
ctx: PoolingServeContext,
):
if ctx.engine_prompts is None:
if ctx.engine_inputs is None:
raise ValueError("Engine prompts not available")
generators: list[AsyncGenerator[PoolingRequestOutput, None]] = []
@@ -120,7 +120,7 @@ class PoolingServing:
pooling_params = self.io_processor.create_pooling_params(ctx.request)
pooling_params.verify(self.model_config)
for i, engine_prompt in enumerate(ctx.engine_prompts):
for i, engine_input in enumerate(ctx.engine_inputs):
prompt_request_id = (
f"{ctx.request_id}-{i}"
if ctx.prompt_request_ids is None
@@ -129,13 +129,13 @@ class PoolingServing:
self._log_inputs(
prompt_request_id,
engine_prompt,
engine_input,
params=pooling_params,
lora_request=ctx.lora_request,
)
generator = self.engine_client.encode(
engine_prompt,
engine_input,
pooling_params,
prompt_request_id,
lora_request=ctx.lora_request,
@@ -151,13 +151,13 @@ class PoolingServing:
self,
ctx: PoolingServeContext,
):
if ctx.engine_prompts is None:
if ctx.engine_inputs is None:
raise ValueError("Engine prompts not available")
if ctx.result_generator is None:
raise ValueError("Result generator not available")
num_inputs = len(ctx.engine_prompts)
num_inputs = len(ctx.engine_inputs)
final_res_batch: list[PoolingRequestOutput | None]
final_res_batch = [None] * num_inputs
@@ -317,7 +317,7 @@ class PoolingServing:
def _log_inputs(
self,
request_id: str,
inputs: ProcessorInputs,
inputs: EngineInput,
params: PoolingParams,
lora_request: LoRARequest | None,
) -> None: