remove monkey patch, openshift ai now works with max_completion_tokens

Signed-off-by: Shawn Hurley <shawn@hurley.page>

kai/llm_interfacing/model_provider.py

@@ -68,34 +68,6 @@ def __init__(
                 model_args = deep_update(defaults, config.args)
                 model_id = model_args["model"]
 
-                # NOTE(JonahSussman): This is a hack to prevent `max_tokens`
-                # from getting converted to `max_completion_tokens` for every
-                # model, except for the o1 and o3 family of models.
-
-                @property  # type: ignore[misc]
-                def _default_params(self: ChatOpenAI) -> dict[str, Any]:
-                    return super(ChatOpenAI, self)._default_params
-
-                def _get_request_payload(
-                    self: ChatOpenAI,
-                    input_: LanguageModelInput,
-                    *,
-                    stop: list[str] | None = None,
-                    **kwargs: Any,
-                ) -> dict:  # type: ignore[type-arg]
-                    return super(ChatOpenAI, self)._get_request_payload(
-                        input_, stop=stop, **kwargs
-                    )
-
-                if not (model_id.startswith("o1") or model_id.startswith("o3")):
-                    ChatOpenAI._default_params = _default_params  # type: ignore[method-assign]
-                    ChatOpenAI._get_request_payload = _get_request_payload  # type: ignore[method-assign]
-                else:
-                    if "streaming" in model_args:
-                        del model_args["streaming"]
-                    if "temperature" in model_args:
-                        del model_args["temperature"]
-
             case "ChatBedrock":
                 model_class = ChatBedrock
 
@@ -177,6 +149,8 @@ def _get_request_payload(
                 raise Exception(f"Unrecognized provider '{config.provider}'")
 
         self.provider_id: str = config.provider
+        # In the future we should consider https://github.com/langchain-ai/langchain/blob/b7a1705052f763b8a11e164041efef1465c07595/libs/langchain/langchain/chat_models/base.py#L85
+        # as this seems to be the more correct way of doing things.
         self.llm: BaseChatModel = model_class(**model_args)
         self.model_id: str = model_id
 
@@ -193,6 +167,20 @@ def validate_environment(
         current model provider.
         """
 
+        # The only time, that model must be invoked is when the pydantic model has not already validated
+        # for the model class. To see a quick list of models that use model_kwargs
+        # https://github.com/search?q=repo%3Alangchain-ai%2Flangchain%20_build_model_kwargs&type=code
+        # if there are extras, not defined in the model but saved/used in some other place then we need to
+        # validate them.
+        has_model_kwargs = (
+            hasattr(self.llm, "model_kwargs")
+            and isinstance(self.llm.model_kwargs, dict)
+            and self.llm.model_kwargs
+        )
+        if not self.llm.model_extra and not has_model_kwargs:
+            LOG.debug("validation done by pydantic")
+            return
+
         cpr = SimplePathResolver("validate_environment.json")
 
         def challenge(k: str) -> BaseMessage: