[Frontend] Complete OpenAI render delegation (#37287)

Signed-off-by: Sage Ahrac <sagiahrak@gmail.com>
This commit is contained in:
Sage
2026-03-17 13:53:55 +00:00
committed by GitHub
parent 56cb1baa66
commit 59192dfd39
8 changed files with 140 additions and 244 deletions
+4 -1
View File
@@ -29,6 +29,7 @@ from vllm.entrypoints.serve.disagg.protocol import (
GenerateResponse,
GenerateResponseChoice,
)
from vllm.entrypoints.serve.render.serving import OpenAIServingRender
from vllm.logger import init_logger
from vllm.logprobs import Logprob
from vllm.outputs import RequestOutput
@@ -45,6 +46,7 @@ class ServingTokens(OpenAIServing):
self,
engine_client: EngineClient,
models: OpenAIServingModels,
openai_serving_render: OpenAIServingRender,
*,
request_logger: RequestLogger | None,
force_no_detokenize: bool = False,
@@ -58,6 +60,7 @@ class ServingTokens(OpenAIServing):
request_logger=request_logger,
return_tokens_as_token_ids=return_tokens_as_token_ids,
)
self.openai_serving_render = openai_serving_render
self.enable_prompt_tokens_details = enable_prompt_tokens_details
self.enable_log_outputs = enable_log_outputs
self.force_no_detokenize = force_no_detokenize
@@ -96,7 +99,7 @@ class ServingTokens(OpenAIServing):
if raw_request:
raw_request.state.request_metadata = request_metadata
engine_prompts = await self._preprocess_completion(
engine_prompts = await self.openai_serving_render.preprocess_completion(
request,
prompt_input=request.token_ids,
prompt_embeds=None,
+7 -10
View File
@@ -24,6 +24,7 @@ from vllm.entrypoints.openai.parser.harmony_utils import (
parse_chat_inputs_to_harmony_messages,
render_for_completion,
)
from vllm.entrypoints.openai.responses.protocol import ResponsesRequest
from vllm.entrypoints.serve.disagg.protocol import (
GenerateRequest,
MultiModalFeatures,
@@ -459,9 +460,9 @@ class OpenAIServingRender:
prompts.extend(prompt_to_seq(prompt_embeds))
if prompt_input is not None:
prompts.extend(prompt_to_seq(prompt_input))
return await self._preprocess_cmpl(request, prompts)
return await self.preprocess_cmpl(request, prompts)
async def _preprocess_cmpl(
async def preprocess_cmpl(
self,
request: Any,
prompts: Sequence[PromptType | bytes],
@@ -500,11 +501,7 @@ class OpenAIServingRender:
tool_dicts: list[dict[str, Any]] | None = None,
tool_parser: Callable[[TokenizerLike], ToolParser] | None = None,
) -> tuple[list[ConversationMessage], list[ProcessorInputs]]:
"""Copied from OpenAIServing._preprocess_chat.
Differences: isinstance check is ChatCompletionRequest-only
(ResponsesRequest not supported here); TODO comment dropped accordingly.
"""
"""Copied from OpenAIServing._preprocess_chat."""
renderer = self.renderer
mm_config = self.model_config.multimodal_config
@@ -542,11 +539,11 @@ class OpenAIServingRender:
if tool_parser is not None:
tool_choice = getattr(request, "tool_choice", "none")
if tool_choice != "none":
if not isinstance(request, ChatCompletionRequest):
if not isinstance(request, ChatCompletionRequest | ResponsesRequest):
msg = (
"Tool usage is only supported "
" for ChatCompletionRequest, but got "
f"{type(request).__name__}"
"for Chat Completions API or Responses API requests, "
f"but got {type(request).__name__}"
)
raise NotImplementedError(msg)
tokenizer = renderer.get_tokenizer()