fix: Validate the completion params that are passed as input to the API #3

tests/unit/types/test_chat_completion_validation.py

@@ -0,0 +1,74 @@
+import pytest
+
+from tlm.utils.chat_completion_validation import _validate_chat_completion_params
+
+
+def test_validate_chat_completion_params_allows_valid_openai_keys() -> None:
+    params = {"messages": [], "model": "gpt-4.1", "temperature": 0.5}
+
+    _validate_chat_completion_params(params)
+
+
+def test_validate_chat_completion_params_allows_provider_as_none() -> None:
+    params = {"messages": [], "model": "gpt-4.1", "temperature": 0.5}
+
+    _validate_chat_completion_params(params)
+
+
+def test_validate_chat_completion_params_requires_messages() -> None:
+    params = {"model": "gpt-4.1-mini"}
+
+    with pytest.raises(ValueError) as exc_info:
+        _validate_chat_completion_params(params)
+
+    assert "openai_args must include the following parameter(s): messages" in str(exc_info.value)
+
+
+def test_validate_chat_completion_params_requires_messages_list() -> None:
+    params = {"messages": "not-a-list"}
+
+    with pytest.raises(ValueError) as exc_info:
+        _validate_chat_completion_params(params)
+
+    assert "`messages` must be provided as a list" in str(exc_info.value)
+
+
+def test_validate_chat_completion_params_requires_message_dict() -> None:
+    params = {"messages": ["not-a-dict"]}
+
+    with pytest.raises(ValueError) as exc_info:
+        _validate_chat_completion_params(params)
+
+    assert "messages[0] must be a dictionary" in str(exc_info.value)
+
+
+def test_validate_chat_completion_params_requires_role_and_content_strings() -> None:
+    params = {"messages": [{"role": 123, "content": None}]}
+
+    with pytest.raises(ValueError) as exc_info:
+        _validate_chat_completion_params(params)
+
+    assert "messages[0]['role']" in str(exc_info.value)
+
+
+def test_validate_chat_completion_params_allows_function_call_without_content() -> None:
+    params = {
+        "messages": [
+            {
+                "role": "assistant",
+                "content": None,
+                "function_call": {"name": "foo", "arguments": '{"bar": 1}'},
+            }
+        ]
+    }
+
+    _validate_chat_completion_params(params)
+
+
+def test_validate_chat_completion_params_requires_content_when_no_function_call() -> None:
+    params = {"messages": [{"role": "assistant", "content": None}]}
+
+    with pytest.raises(ValueError) as exc_info:
+        _validate_chat_completion_params(params)
+
+    assert "messages[0]['content'] must be a string." in str(exc_info.value)

tlm/api.py

@@ -4,6 +4,7 @@
 from tlm.config.presets import WorkflowType
 from tlm.inference import InferenceResult, tlm_inference
 from tlm.types import SemanticEval, CompletionParams
+from tlm.utils.chat_completion_validation import _validate_chat_completion_params
 
 
 async def inference(
@@ -14,6 +15,7 @@ async def inference(
     evals: list[SemanticEval] | None = None,
     config_input: ConfigInput = ConfigInput(),
 ) -> InferenceResult:
+    _validate_chat_completion_params(openai_args)
     workflow_type = WorkflowType.from_inference_params(
         openai_args=openai_args,
         score=response is not None,

tlm/utils/chat_completion_validation.py

@@ -0,0 +1,54 @@
+"""Validation helpers for chat completion parameter dictionaries."""
+
+from typing import Any, Callable, FrozenSet, Mapping
+
+from tlm.types.base import CompletionParams
+
+
+ParamValidator = Callable[[Any], None]
+
+REQUIRED_CHAT_COMPLETION_PARAMS: FrozenSet[str] = frozenset({"messages"})
+
+
+def _validate_messages_param(messages: Any) -> None:
+    """Validate the shape of a `messages` param for chat completions."""
+
+    if not isinstance(messages, list):
+        raise ValueError("`messages` must be provided as a list of message dictionaries.")
+
+    for index, message in enumerate(messages):
+        if not isinstance(message, dict):
+            raise ValueError(f"messages[{index}] must be a dictionary.")
+
+        role = message.get("role")
+        content = message.get("content")
+
+        if role is None or not isinstance(role, str):
+            raise ValueError(f"messages[{index}]['role'] must be a non-empty string.")
+
+        if content is None or not isinstance(content, str):
+            function_call = message.get("function_call")
+            if role != "assistant":
+                raise ValueError(f"Non-assistant message at index {index} must have content.")
+            if function_call is None:
+                raise ValueError(f"Assistant message at index {index} must have content or a function call.")
+
+
+REQUIRED_PARAM_VALIDATORS: Mapping[str, ParamValidator] = {
+    "messages": _validate_messages_param,
+}
+
+
+def _validate_chat_completion_params(params: CompletionParams) -> None:  # type: ignore
+    """Ensure only supported chat completion params are passed into inference."""
+
+    missing_required = [param for param in REQUIRED_CHAT_COMPLETION_PARAMS if param not in params]
+    if missing_required:
+        required_str = ", ".join(sorted(REQUIRED_CHAT_COMPLETION_PARAMS))
+        raise ValueError(f"openai_args must include the following parameter(s): {required_str}")
+
+    for param in REQUIRED_CHAT_COMPLETION_PARAMS:
+        validator = REQUIRED_PARAM_VALIDATORS.get(param)
+        if validator is None:
+            continue
+        validator(params[param])