[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
@@ -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,