[Frontend] Add /v1/messages/render endpoint for the Anthropic Messages API (#45803)

Signed-off-by: HyunKyun Moon <mhg5303@gmail.com>
This commit is contained in:
Hyunkyun Moon
2026-08-26 11:33:59 +08:00
committed by GitHub
parent a5e0045433
commit f25c2692fe
5 changed files with 165 additions and 5 deletions
+3 -3
View File
@@ -191,7 +191,7 @@ class AnthropicServingMessages(OpenAIServingChat):
return f"data:{media_type};base64,{data}"
@classmethod
def _convert_anthropic_to_openai_request(
def to_chat_completion_request(
cls,
anthropic_request: AnthropicMessagesRequest | AnthropicCountTokensRequest,
*,
@@ -602,7 +602,7 @@ class AnthropicServingMessages(OpenAIServingChat):
"""
if logger.isEnabledFor(logging.DEBUG):
logger.debug("Received messages request %s", request.model_dump_json())
chat_req = self._convert_anthropic_to_openai_request(
chat_req = self.to_chat_completion_request(
request,
merge_inline_system=self._merge_inline_system,
)
@@ -1023,7 +1023,7 @@ class AnthropicServingMessages(OpenAIServingChat):
raw_request: Request | None = None,
) -> AnthropicCountTokensResponse | ErrorResponse:
"""Implements Anthropic's messages.count_tokens endpoint."""
chat_req = self._convert_anthropic_to_openai_request(
chat_req = self.to_chat_completion_request(
request,
merge_inline_system=self._merge_inline_system,
)
@@ -5,6 +5,7 @@ from http import HTTPStatus
from fastapi import APIRouter, Depends, Request
from fastapi.responses import JSONResponse
from vllm.entrypoints.anthropic.protocol import AnthropicMessagesRequest
from vllm.entrypoints.openai.chat_completion.protocol import ChatCompletionRequest
from vllm.entrypoints.openai.completion.protocol import CompletionRequest
from vllm.entrypoints.openai.engine.protocol import ErrorResponse
@@ -49,6 +50,30 @@ async def render_chat_completion(request: ChatCompletionRequest, raw_request: Re
return JSONResponse(content=result.model_dump())
@router.post(
"/v1/messages/render",
dependencies=[Depends(validate_json_request)],
response_model=GenerateRequest,
responses={
HTTPStatus.BAD_REQUEST.value: {"model": ErrorResponse},
HTTPStatus.NOT_FOUND.value: {"model": ErrorResponse},
HTTPStatus.NOT_IMPLEMENTED.value: {"model": ErrorResponse},
HTTPStatus.INTERNAL_SERVER_ERROR.value: {"model": ErrorResponse},
},
)
async def render_messages(request: AnthropicMessagesRequest, raw_request: Request):
handler = render(raw_request)
if handler is None:
raise NotImplementedError("The model does not support Messages Render API")
result = await handler.render_messages_request(request)
if isinstance(result, ErrorResponse):
return JSONResponse(content=result.model_dump(), status_code=result.error.code)
return JSONResponse(content=result.model_dump())
@router.post(
"/v1/completions/render",
dependencies=[Depends(validate_json_request)],
@@ -2,6 +2,8 @@
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
from typing import cast
from vllm.entrypoints.anthropic.protocol import AnthropicMessagesRequest
from vllm.entrypoints.anthropic.serving import AnthropicServingMessages
from vllm.entrypoints.openai.chat_completion.protocol import ChatCompletionRequest
from vllm.entrypoints.openai.completion.protocol import CompletionRequest
from vllm.entrypoints.openai.engine.protocol import ErrorResponse
@@ -51,6 +53,12 @@ class ServingRender(BaseServing):
self.online_renderer = online_renderer
self._merge_inline_system = (
AnthropicServingMessages._detect_merge_inline_system(
online_renderer.chat_template
)
)
self.default_sampling_params = (
online_renderer.model_config.get_diff_sampling_param()
)
@@ -149,6 +157,21 @@ class ServingRender(BaseServing):
token_offsets=engine_input.get("prompt_token_offsets"),
)
async def render_messages_request(
self,
request: AnthropicMessagesRequest,
) -> GenerateRequest | ErrorResponse:
"""Validate the model and preprocess an Anthropic Messages request.
Converts the request to the OpenAI chat format using the same
conversion as the /v1/messages server path, then delegates to
render_chat_request so the rendered tokens match the server exactly.
"""
chat_req = AnthropicServingMessages.to_chat_completion_request(
request, merge_inline_system=self._merge_inline_system
)
return await self.render_chat_request(chat_req)
async def render_completion_request(
self,
request: CompletionRequest,