| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | |
| 4 | from typing import Any |
| 5 | |
| 6 | import pytest |
| 7 | import requests |
| 8 | |
| 9 | from tests.utils import RemoteOpenAIServer |
| 10 | |
| 11 | MODEL_NAME = "Qwen/Qwen3-0.6B" |
| 12 | |
| 13 | |
| 14 | class TestWorkerExtension: |
| 15 | def get_model_name(self) -> str: |
| 16 | """Test non-pydantic return type.""" |
| 17 | return MODEL_NAME |
| 18 | |
| 19 | def echo_args_kwargs(self, *args, **kwargs) -> dict[str, Any]: |
| 20 | """Echo back both args and kwargs.""" |
| 21 | return dict( |
| 22 | args=list(args), |
| 23 | kwargs=kwargs, |
| 24 | total_items=len(args) + len(kwargs), |
| 25 | ) |
| 26 | |
| 27 | def return_none(self, *args, **kwargs) -> None: |
| 28 | """Test method that does not return anything.""" |
| 29 | return |
| 30 | |
| 31 | |
| 32 | @pytest.fixture(scope="module") |
| 33 | def server(): |
| 34 | args = [ |
| 35 | "--max-model-len", |
| 36 | "8192", |
| 37 | "--max-num-seqs", |
| 38 | "128", |
| 39 | "--worker-extension-cls", |
| 40 | "tests.entrypoints.serve.dev.rpc.test_collective_rpc.TestWorkerExtension", |
| 41 | ] |
| 42 | with RemoteOpenAIServer( |
| 43 | MODEL_NAME, |
| 44 | args, |
| 45 | env_dict={"VLLM_SERVER_DEV_MODE": "1", "CUDA_VISIBLE_DEVICES": "0"}, |
| 46 | ) as remote_server: |
| 47 | yield remote_server |
| 48 | |
| 49 | |
| 50 | def test_get_model_name(server): |
| 51 | """Test basic response.""" |
| 52 | response = requests.post( |
| 53 | server.url_for("collective_rpc"), json={"method": "get_model_name"} |
| 54 | ) |
| 55 | assert response.status_code == 200 |
| 56 | results = response.json() |
| 57 | assert "results" in results |
| 58 | assert results["results"] == [MODEL_NAME] |
| 59 | |
| 60 | |
| 61 | def test_return_none(server): |
| 62 | """Test return none.""" |
| 63 | response = requests.post( |
| 64 | server.url_for("collective_rpc"), json={"method": "return_none"} |
| 65 | ) |
| 66 | assert response.status_code == 200 |
| 67 | results = response.json() |
| 68 | assert results["results"] == [None] |
| 69 | |
| 70 | |
| 71 | def test_echo_args_kwargs(server): |
| 72 | """Test args, kwargs, and dict response.""" |
| 73 | args = ["arg1", "arg2"] |
| 74 | kwargs = {"key1": "value1", "key2": "value2"} |
| 75 | response = requests.post( |
| 76 | server.url_for("collective_rpc"), |
| 77 | json={"method": "echo_args_kwargs", "args": args, "kwargs": kwargs}, |
| 78 | ) |
| 79 | assert response.status_code == 200 |
| 80 | results = response.json() |
| 81 | result = results["results"][0] |
| 82 | assert result["args"] == args |
| 83 | assert result["kwargs"] == kwargs |
| 84 | assert result["total_items"] == len(args) + len(kwargs) |