[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
@@ -63,7 +63,7 @@ from vllm.entrypoints.openai.parser.harmony_utils import (
)
from vllm.entrypoints.openai.utils import maybe_filter_parallel_tool_calls
from vllm.entrypoints.utils import get_max_tokens, should_include_usage
from vllm.inputs.data import ProcessorInputs
from vllm.inputs import EngineInput
from vllm.logger import init_logger
from vllm.logprobs import Logprob
from vllm.outputs import CompletionOutput, RequestOutput
@@ -177,7 +177,7 @@ class OpenAIServingChat(OpenAIServing):
async def render_chat_request(
self,
request: ChatCompletionRequest,
) -> tuple[list[ConversationMessage], list[ProcessorInputs]] | ErrorResponse:
) -> tuple[list[ConversationMessage], list[EngineInput]] | ErrorResponse:
"""
Validate the model and preprocess a chat completion request.
@@ -185,7 +185,7 @@ class OpenAIServingChat(OpenAIServing):
engine-aware checks (LoRA model validation, engine health).
Returns:
A tuple of (conversation, engine_prompts) on success,
A tuple of (conversation, engine_inputs) on success,
or an ErrorResponse on failure.
"""
error_check_ret = await self._check_model(request)
@@ -231,7 +231,7 @@ class OpenAIServingChat(OpenAIServing):
if isinstance(result, ErrorResponse):
return result
conversation, engine_prompts = result
conversation, engine_inputs = result
request_id = (
f"chatcmpl-{self._base_request_id(raw_request, request.request_id)}"
@@ -251,13 +251,13 @@ class OpenAIServingChat(OpenAIServing):
# Schedule the request and get the result generator.
max_model_len = self.model_config.max_model_len
generators: list[AsyncGenerator[RequestOutput, None]] = []
for i, engine_prompt in enumerate(engine_prompts):
prompt_token_ids = self._extract_prompt_components(engine_prompt).token_ids
for i, engine_input in enumerate(engine_inputs):
prompt_token_ids = self._extract_prompt_components(engine_input).token_ids
# If we are creating sub requests for multiple prompts, ensure that they
# have unique request ids.
sub_request_id = (
request_id if len(engine_prompts) == 1 else f"{request_id}_{i}"
request_id if len(engine_inputs) == 1 else f"{request_id}_{i}"
)
max_tokens = get_max_tokens(
@@ -265,7 +265,7 @@ class OpenAIServingChat(OpenAIServing):
request.max_completion_tokens
if request.max_completion_tokens is not None
else request.max_tokens,
self._extract_prompt_len(engine_prompt),
self._extract_prompt_len(engine_input),
self.default_sampling_params,
self.override_max_tokens,
)
@@ -283,7 +283,7 @@ class OpenAIServingChat(OpenAIServing):
self._log_inputs(
sub_request_id,
engine_prompt,
engine_input,
params=sampling_params,
lora_request=lora_request,
)
@@ -296,7 +296,7 @@ class OpenAIServingChat(OpenAIServing):
if isinstance(sampling_params, BeamSearchParams):
generator = self.beam_search(
prompt=engine_prompt,
prompt=engine_input,
request_id=sub_request_id,
params=sampling_params,
lora_request=lora_request,
@@ -313,7 +313,7 @@ class OpenAIServingChat(OpenAIServing):
reasoning_ended = None
generator = self.engine_client.generate(
engine_prompt,
engine_input,
sampling_params,
sub_request_id,
lora_request=lora_request,