|
| 1 | +"""Unit tests for discriminated inference.providers[] config entries.""" |
| 2 | + |
| 3 | +import pytest |
| 4 | +from pydantic import TypeAdapter, ValidationError |
| 5 | + |
| 6 | +from models.config import ( |
| 7 | + AllowedModel, |
| 8 | + BedrockInferenceProvider, |
| 9 | + InferenceConfiguration, |
| 10 | + OpenAIInferenceProvider, |
| 11 | + SentenceTransformersInferenceProvider, |
| 12 | + UnifiedInferenceProvider, |
| 13 | + VllmInferenceProvider, |
| 14 | +) |
| 15 | + |
| 16 | +PROVIDER_ADAPTER = TypeAdapter(UnifiedInferenceProvider) |
| 17 | + |
| 18 | + |
| 19 | +@pytest.mark.parametrize( |
| 20 | + "provider_type", |
| 21 | + [ |
| 22 | + "openai", |
| 23 | + "azure", |
| 24 | + "vertexai", |
| 25 | + "bedrock", |
| 26 | + "watsonx", |
| 27 | + "vllm", |
| 28 | + "sentence_transformers", |
| 29 | + ], |
| 30 | +) |
| 31 | +def test_all_seven_provider_types_parse(provider_type: str) -> None: |
| 32 | + """Each supported provider type parses into the discriminated union.""" |
| 33 | + payload: dict[str, object] = {"type": provider_type} |
| 34 | + if provider_type == "sentence_transformers": |
| 35 | + payload["allowed_models"] = ["all-MiniLM-L6-v2"] |
| 36 | + provider = PROVIDER_ADAPTER.validate_python(payload) |
| 37 | + assert provider.type == provider_type |
| 38 | + assert provider.id == provider_type |
| 39 | + |
| 40 | + |
| 41 | +def test_allowed_models_string_form_normalizes() -> None: |
| 42 | + """Bare string allow-list entries set name and provider_model_id equal.""" |
| 43 | + model = AllowedModel.model_validate("gpt-4o-mini") |
| 44 | + assert model.name == "gpt-4o-mini" |
| 45 | + assert model.provider_model_id == "gpt-4o-mini" |
| 46 | + assert model.model_type is None |
| 47 | + assert model.metadata == {} |
| 48 | + |
| 49 | + |
| 50 | +def test_allowed_models_structured_form() -> None: |
| 51 | + """Structured allow-list entries keep local name, wire id, type, metadata.""" |
| 52 | + model = AllowedModel.model_validate( |
| 53 | + { |
| 54 | + "name": "fast-chat", |
| 55 | + "provider_model_id": "gpt-4o-mini", |
| 56 | + "model_type": "llm", |
| 57 | + "metadata": {"context_window": 128000}, |
| 58 | + } |
| 59 | + ) |
| 60 | + assert model.name == "fast-chat" |
| 61 | + assert model.provider_model_id == "gpt-4o-mini" |
| 62 | + assert model.model_type == "llm" |
| 63 | + assert model.metadata["context_window"] == 128000 |
| 64 | + |
| 65 | + |
| 66 | +def test_allowed_models_defaults_name_to_provider_model_id() -> None: |
| 67 | + """Omitted name defaults to provider_model_id.""" |
| 68 | + model = AllowedModel.model_validate( |
| 69 | + {"provider_model_id": "openai/gpt-4o-mini", "model_type": "llm"} |
| 70 | + ) |
| 71 | + assert model.name == "openai/gpt-4o-mini" |
| 72 | + assert model.provider_model_id == "openai/gpt-4o-mini" |
| 73 | + |
| 74 | + |
| 75 | +def test_omitted_id_defaults_to_type() -> None: |
| 76 | + """Provider id defaults to type when omitted.""" |
| 77 | + provider = OpenAIInferenceProvider.model_validate({"type": "openai"}) |
| 78 | + assert provider.id == "openai" |
| 79 | + |
| 80 | + |
| 81 | +def test_explicit_id_is_kept() -> None: |
| 82 | + """Explicit provider id is preserved.""" |
| 83 | + provider = VllmInferenceProvider.model_validate({"type": "vllm", "id": "vllm-prod"}) |
| 84 | + assert provider.id == "vllm-prod" |
| 85 | + |
| 86 | + |
| 87 | +def test_duplicate_provider_ids_fail() -> None: |
| 88 | + """Two providers resolving to the same id fail validation.""" |
| 89 | + with pytest.raises(ValidationError, match="duplicate inference provider id"): |
| 90 | + InferenceConfiguration( |
| 91 | + providers=[ |
| 92 | + {"type": "vllm"}, |
| 93 | + {"type": "vllm"}, |
| 94 | + ] |
| 95 | + ) |
| 96 | + |
| 97 | + |
| 98 | +def test_duplicate_local_model_names_within_provider_fail() -> None: |
| 99 | + """Duplicate local names in one provider's allow-list fail validation.""" |
| 100 | + with pytest.raises(ValidationError, match="allowed model names must be unique"): |
| 101 | + OpenAIInferenceProvider.model_validate( |
| 102 | + { |
| 103 | + "type": "openai", |
| 104 | + "allowed_models": [ |
| 105 | + "same", |
| 106 | + {"name": "same", "provider_model_id": "other"}, |
| 107 | + ], |
| 108 | + } |
| 109 | + ) |
| 110 | + |
| 111 | + |
| 112 | +def test_same_local_model_name_across_providers_allowed() -> None: |
| 113 | + """The same local model name may appear on different providers.""" |
| 114 | + config = InferenceConfiguration( |
| 115 | + providers=[ |
| 116 | + {"type": "openai", "allowed_models": ["shared"]}, |
| 117 | + {"type": "vllm", "id": "vllm-prod", "allowed_models": ["shared"]}, |
| 118 | + ] |
| 119 | + ) |
| 120 | + assert config.providers[0].allowed_models is not None |
| 121 | + assert config.providers[1].allowed_models is not None |
| 122 | + assert config.providers[0].allowed_models[0].name == "shared" |
| 123 | + assert config.providers[1].allowed_models[0].name == "shared" |
| 124 | + |
| 125 | + |
| 126 | +def test_skip_model_discovery_requires_allowed_models() -> None: |
| 127 | + """skip_model_discovery: true requires a non-empty allow-list.""" |
| 128 | + with pytest.raises(ValidationError, match="skip_model_discovery"): |
| 129 | + OpenAIInferenceProvider.model_validate( |
| 130 | + {"type": "openai", "skip_model_discovery": True} |
| 131 | + ) |
| 132 | + |
| 133 | + |
| 134 | +def test_skip_model_discovery_with_allowed_models() -> None: |
| 135 | + """skip_model_discovery succeeds when allowed_models is non-empty.""" |
| 136 | + provider = OpenAIInferenceProvider.model_validate( |
| 137 | + { |
| 138 | + "type": "openai", |
| 139 | + "skip_model_discovery": True, |
| 140 | + "allowed_models": ["gpt-4o-mini"], |
| 141 | + } |
| 142 | + ) |
| 143 | + assert provider.skip_model_discovery is True |
| 144 | + |
| 145 | + |
| 146 | +def test_sentence_transformers_requires_allowed_models() -> None: |
| 147 | + """Sentence-transformers rejects an empty/missing allow-list.""" |
| 148 | + with pytest.raises( |
| 149 | + ValidationError, |
| 150 | + match="allowed_models must be non-empty for sentence_transformers", |
| 151 | + ): |
| 152 | + SentenceTransformersInferenceProvider.model_validate( |
| 153 | + {"type": "sentence_transformers"} |
| 154 | + ) |
| 155 | + |
| 156 | + |
| 157 | +def test_nested_config_parses() -> None: |
| 158 | + """Typed nested config blocks parse for a remote provider.""" |
| 159 | + provider = BedrockInferenceProvider.model_validate( |
| 160 | + { |
| 161 | + "type": "bedrock", |
| 162 | + "config": {"region": "${env.AWS_REGION}", "profile_name": "default"}, |
| 163 | + } |
| 164 | + ) |
| 165 | + assert provider.config.region == "${env.AWS_REGION}" |
| 166 | + assert provider.config.profile_name == "default" |
0 commit comments