| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | |
| 4 | from http import HTTPStatus |
| 5 | from unittest.mock import MagicMock |
| 6 | |
| 7 | import pytest |
| 8 | |
| 9 | import vllm.envs as envs |
| 10 | from vllm.entrypoints.generate.base.serving import GenerateBaseServing |
| 11 | from vllm.envs import disable_envs_cache |
| 12 | from vllm.exceptions import GenerationError |
| 13 | |
| 14 | |
| 15 | @pytest.mark.asyncio |
| 16 | async def test_raise_if_error_raises_generation_error(): |
| 17 | """Test _raise_if_error raises GenerationError.""" |
| 18 | # create a minimal GenerateBaseServing instance |
| 19 | mock_engine = MagicMock() |
| 20 | mock_engine.model_config = MagicMock() |
| 21 | mock_engine.model_config.max_model_len = 100 |
| 22 | mock_models = MagicMock() |
| 23 | |
| 24 | serving = GenerateBaseServing( |
| 25 | engine_client=mock_engine, |
| 26 | models=mock_models, |
| 27 | request_logger=None, |
| 28 | ) |
| 29 | |
| 30 | # test that error finish_reason raises GenerationError |
| 31 | with pytest.raises(GenerationError) as exc_info: |
| 32 | serving._raise_if_error("error", "test-request-id") |
| 33 | |
| 34 | assert str(exc_info.value) == "Internal server error" |
| 35 | assert exc_info.value.status_code == HTTPStatus.INTERNAL_SERVER_ERROR |
| 36 | |
| 37 | # test that other finish_reasons don't raise |
| 38 | serving._raise_if_error("stop", "test-request-id") # should not raise |
| 39 | serving._raise_if_error("length", "test-request-id") # should not raise |
| 40 | serving._raise_if_error(None, "test-request-id") # should not raise |
| 41 | |
| 42 | |
| 43 | @pytest.mark.asyncio |
| 44 | async def test_convert_generation_error_to_streaming_response(): |
| 45 | """Test _convert_generation_error_to_streaming_response output.""" |
| 46 | mock_engine = MagicMock() |
| 47 | mock_engine.model_config = MagicMock() |
| 48 | mock_engine.model_config.max_model_len = 100 |
| 49 | mock_models = MagicMock() |
| 50 | |
| 51 | serving = GenerateBaseServing( |
| 52 | engine_client=mock_engine, |
| 53 | models=mock_models, |
| 54 | request_logger=None, |
| 55 | ) |
| 56 | |
| 57 | # create a GenerationError |
| 58 | gen_error = GenerationError("Internal server error") |
| 59 | |
| 60 | # convert to streaming error response |
| 61 | error_json = serving._convert_generation_error_to_streaming_response(gen_error) |
| 62 | |
| 63 | assert isinstance(error_json, str) |
| 64 | assert "Internal server error" in error_json |
| 65 | assert "InternalServerError" in error_json |
| 66 | |
| 67 | |
| 68 | def test_is_model_supported_skip_name_validation_env( |
| 69 | monkeypatch: pytest.MonkeyPatch, |
| 70 | ) -> None: |
| 71 | """When VLLM_SKIP_MODEL_NAME_VALIDATION is set, accept any model id.""" |
| 72 | disable_envs_cache() |
| 73 | monkeypatch.delenv("VLLM_SKIP_MODEL_NAME_VALIDATION", raising=False) |
| 74 | |
| 75 | mock_engine = MagicMock() |
| 76 | mock_engine.model_config = MagicMock() |
| 77 | mock_engine.model_config.max_model_len = 100 |
| 78 | mock_models = MagicMock() |
| 79 | mock_models.is_base_model.return_value = False |
| 80 | |
| 81 | serving = GenerateBaseServing( |
| 82 | engine_client=mock_engine, |
| 83 | models=mock_models, |
| 84 | request_logger=None, |
| 85 | ) |
| 86 | |
| 87 | assert serving._is_model_supported("not-a-registered-model") is False |
| 88 | |
| 89 | monkeypatch.setenv("VLLM_SKIP_MODEL_NAME_VALIDATION", "1") |
| 90 | disable_envs_cache() |
| 91 | assert envs.VLLM_SKIP_MODEL_NAME_VALIDATION is True |
| 92 | assert serving._is_model_supported("not-a-registered-model") is True |
| 93 | |
| 94 | monkeypatch.setenv("VLLM_SKIP_MODEL_NAME_VALIDATION", "true") |
| 95 | disable_envs_cache() |
| 96 | assert envs.VLLM_SKIP_MODEL_NAME_VALIDATION is True |
| 97 | assert serving._is_model_supported("another-alias") is True |