Adds anthropic /v1/messages endpoint to openai api_server

tests/entrypoints/anthropic/test_messages.py

@@ -105,37 +105,36 @@ async def test_anthropic_tool_call(client: anthropic.AsyncAnthropic):
 
     print(f"Anthropic response: {resp.model_dump_json()}")
 
-    @pytest.mark.asyncio
-    async def test_anthropic_tool_call_streaming(client: anthropic.AsyncAnthropic):
-        resp = await client.messages.create(
-            model="claude-3-7-sonnet-latest",
-            max_tokens=1024,
-            messages=[
-                {
-                    "role": "user",
-                    "content": "What's the weather like in New York today?",
-                }
-            ],
-            tools=[
-                {
-                    "name": "get_current_weather",
-                    "description": "Useful for querying the weather "
-                    "in a specified city.",
-                    "input_schema": {
-                        "type": "object",
-                        "properties": {
-                            "location": {
-                                "type": "string",
-                                "description": "City or region, for example: "
-                                "New York, London, Tokyo, etc.",
-                            }
-                        },
-                        "required": ["location"],
+@pytest.mark.asyncio
+async def test_anthropic_tool_call_streaming(client: anthropic.AsyncAnthropic):
+    resp = await client.messages.create(
+        model="claude-3-7-sonnet-latest",
+        max_tokens=1024,
+        messages=[
+            {
+                "role": "user",
+                "content": "What's the weather like in New York today?",
+            }
+        ],
+        tools=[
+            {
+                "name": "get_current_weather",
+                "description": "Useful for querying the weather in a specified city.",
+                "input_schema": {
+                    "type": "object",
+                    "properties": {
+                        "location": {
+                            "type": "string",
+                            "description": "City or region, for example: "
+                            "New York, London, Tokyo, etc.",
+                        }
                     },
-                }
-            ],
-            stream=True,
-        )
+                    "required": ["location"],
+                },
+            }
+        ],
+        stream=True,
+    )
 
-        async for chunk in resp:
-            print(chunk.model_dump_json())
+    async for chunk in resp:
+        print(chunk.model_dump_json())

tests/entrypoints/openai/test_messages.py

@@ -0,0 +1,140 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+import anthropic
+import pytest
+import pytest_asyncio
+
+from ...utils import RemoteOpenAIServer
+
+MODEL_NAME = "Qwen/Qwen3-0.6B"
+
+
+@pytest.fixture(scope="module")
+def server():  # noqa: F811
+    args = [
+        "--max-model-len",
+        "2048",
+        "--enforce-eager",
+        "--enable-auto-tool-choice",
+        "--tool-call-parser",
+        "hermes",
+        "--served-model-name",
+        "claude-3-7-sonnet-latest",
+    ]
+
+    with RemoteOpenAIServer(MODEL_NAME, args) as remote_server:
+        yield remote_server
+
+
+@pytest_asyncio.fixture
+async def client(server):
+    async with server.get_async_client() as async_client:
+        yield async_client
+
+
+@pytest.mark.asyncio
+async def test_simple_messages(client: anthropic.AsyncAnthropic):
+    resp = await client.messages.create(
+        model="claude-3-7-sonnet-latest",
+        max_tokens=1024,
+        messages=[{"role": "user", "content": "how are you!"}],
+    )
+    assert resp.stop_reason == "end_turn"
+    assert resp.role == "assistant"
+
+    print(f"Anthropic response: {resp.model_dump_json()}")
+
+
+@pytest.mark.asyncio
+async def test_system_message(client: anthropic.AsyncAnthropic):
+    resp = await client.messages.create(
+        model="claude-3-7-sonnet-latest",
+        max_tokens=1024,
+        system="you are a helpful assistant",
+        messages=[{"role": "user", "content": "how are you!"}],
+    )
+    assert resp.stop_reason == "end_turn"
+    assert resp.role == "assistant"
+
+    print(f"Anthropic response: {resp.model_dump_json()}")
+
+
+@pytest.mark.asyncio
+async def test_anthropic_streaming(client: anthropic.AsyncAnthropic):
+    resp = await client.messages.create(
+        model="claude-3-7-sonnet-latest",
+        max_tokens=1024,
+        messages=[{"role": "user", "content": "how are you!"}],
+        stream=True,
+    )
+
+    async for chunk in resp:
+        print(chunk.model_dump_json())
+
+
+@pytest.mark.asyncio
+async def test_anthropic_tool_call(client: anthropic.AsyncAnthropic):
+    resp = await client.messages.create(
+        model="claude-3-7-sonnet-latest",
+        max_tokens=1024,
+        messages=[
+            {"role": "user", "content": "What's the weather like in New York today?"}
+        ],
+        tools=[
+            {
+                "name": "get_current_weather",
+                "description": "Useful for querying the weather in a specified city.",
+                "input_schema": {
+                    "type": "object",
+                    "properties": {
+                        "location": {
+                            "type": "string",
+                            "description": "City or region, for example: "
+                            "New York, London, Tokyo, etc.",
+                        }
+                    },
+                    "required": ["location"],
+                },
+            }
+        ],
+        stream=False,
+    )
+    assert resp.stop_reason == "tool_use"
+    assert resp.role == "assistant"
+
+    print(f"Anthropic response: {resp.model_dump_json()}")
+
+@pytest.mark.asyncio
+async def test_anthropic_tool_call_streaming(client: anthropic.AsyncAnthropic):
+    resp = await client.messages.create(
+        model="claude-3-7-sonnet-latest",
+        max_tokens=1024,
+        messages=[
+            {
+                "role": "user",
+                "content": "What's the weather like in New York today?",
+            }
+        ],
+        tools=[
+            {
+                "name": "get_current_weather",
+                "description": "Useful for querying the weather in a specified city.",
+                "input_schema": {
+                    "type": "object",
+                    "properties": {
+                        "location": {
+                            "type": "string",
+                            "description": "City or region, for example: "
+                            "New York, London, Tokyo, etc.",
+                        }
+                    },
+                    "required": ["location"],
+                },
+            }
+        ],
+        stream=True,
+    )
+
+    async for chunk in resp:
+        print(chunk.model_dump_json())

vllm/entrypoints/openai/api_server.py

@@ -41,6 +41,13 @@
 from vllm.config import VllmConfig
 from vllm.engine.arg_utils import AsyncEngineArgs
 from vllm.engine.protocol import Device, EngineClient
+from vllm.entrypoints.anthropic.protocol import (
+    AnthropicError,
+    AnthropicErrorResponse,
+    AnthropicMessagesRequest,
+    AnthropicMessagesResponse,
+)
+from vllm.entrypoints.anthropic.serving_messages import AnthropicServingMessages
 from vllm.entrypoints.launcher import serve_http
 from vllm.entrypoints.logger import RequestLogger
 from vllm.entrypoints.openai.cli_args import make_arg_parser, validate_parsed_serve_args
@@ -307,6 +314,10 @@
     return request.app.state.openai_serving_responses
 
 
+def messages(request: Request) -> AnthropicServingMessages:
+    return request.app.state.anthropic_serving_messages
+
+
 def chat(request: Request) -> OpenAIServingChat | None:
     return request.app.state.openai_serving_chat
 
@@ -590,6 +601,63 @@
     return JSONResponse(content=response.model_dump())
 
 
+@router.post(
+    "/v1/messages",
+    dependencies=[Depends(validate_json_request)],
+    responses={
+        HTTPStatus.OK.value: {"content": {"text/event-stream": {}}},
+        HTTPStatus.BAD_REQUEST.value: {"model": AnthropicErrorResponse},
+        HTTPStatus.NOT_FOUND.value: {"model": AnthropicErrorResponse},
+        HTTPStatus.INTERNAL_SERVER_ERROR.value: {"model": AnthropicErrorResponse},
+    },
+)
+@with_cancellation
+@load_aware_call
+async def create_messages(request: AnthropicMessagesRequest, raw_request: Request):
+    def translate_error_response(response: ErrorResponse) -> JSONResponse:
+        anthropic_error = AnthropicErrorResponse(
+            error=AnthropicError(
+                type=response.error.type,
+                message=response.error.message,
+            )
+        )
+        return JSONResponse(status_code=response.error.code,
+                            content=anthropic_error.model_dump())
+
+    handler = messages(raw_request)
+    if handler is None:
+        error = base(raw_request).create_error_response(
+            message="The model does not support Messages API"
+        )
+        return translate_error_response(error)
+
+    try:
+        generator = await handler.create_messages(request, raw_request)
+    except Exception as e:
+        logger.exception("Error in create_messages: %s", e, exc_info=True)
+        return JSONResponse(
+            status_code=HTTPStatus.INTERNAL_SERVER_ERROR.value,
+            content=AnthropicErrorResponse(
+                error=AnthropicError(
+                    type="internal_error",
+                    message=str(e),
+                )
+            ).model_dump(),
+        )
+
+    if isinstance(generator, ErrorResponse):
+        return translate_error_response(generator)
+
+    elif isinstance(generator, AnthropicMessagesResponse):
+        logger.debug(
+            "Anthropic Messages Response: %s", generator.model_dump(exclude_none=True)
+        )
+        return JSONResponse(content=generator.model_dump(exclude_none=True))
+
+    return StreamingResponse(content=generator, media_type="text/event-stream")
+
+
+
 @router.post(
     "/v1/chat/completions",
     dependencies=[Depends(validate_json_request)],
@@ -1816,6 +1884,24 @@
         if "transcription" in supported_tasks
         else None
     )
+    state.anthropic_serving_messages = (
+        AnthropicServingMessages(
+            engine_client,
+            state.openai_serving_models,
+            args.response_role,
+            request_logger=request_logger,
+            chat_template=resolved_chat_template,
+            chat_template_content_format=args.chat_template_content_format,
+            return_tokens_as_token_ids=args.return_tokens_as_token_ids,
+            enable_auto_tools=args.enable_auto_tool_choice,
+            tool_parser=args.tool_call_parser,
+            reasoning_parser=args.structured_outputs_config.reasoning_parser,
+            enable_prompt_tokens_details=args.enable_prompt_tokens_details,
+            enable_force_include_usage=args.enable_force_include_usage,
+        )
+        if "generate" in supported_tasks
+        else None
+    )
 
     state.enable_server_load_tracking = args.enable_server_load_tracking
     state.server_load_metrics = 0