make connectors use user provided base_url when the provided model is…

… also hosted

previously, if a model was known to be hosted on a custom endpoint and a user ran the same model locally, the connectors would favor the hosted version. for instance, NVIDIAEmbeddings(model="NV-Embed-QA", base_url="http://localhost/v1") would contact the hosted NV-Embed-QA. likewise, ChatNVIDIA(model="mistralai/mixtral-8x7b-instruct-v0.1", base_url="http://localhost/v1") would contact the hosted mistralai/mixtral-8x7b-instruct-v0.1.

fixes #31

libs/ai-endpoints/langchain_nvidia_ai_endpoints/_common.py

@@ -524,6 +524,11 @@ def _postprocess_args(cls, values: Any) -> Any:
             name = values.get("model")
             if model := determine_model(name):
                 values["model"] = model.id
+                # not all models are on https://integrate.api.nvidia.com/v1,
+                # those that are not are served from their own endpoints
+                if model.endpoint:
+                    # we override the infer_path to use the custom endpoint
+                    values["client"].infer_path = model.endpoint
             else:
                 if not (client := values.get("client")):
                     warnings.warn(f"Unable to determine validity of {name}")

libs/ai-endpoints/langchain_nvidia_ai_endpoints/chat_models.py

@@ -168,17 +168,11 @@ def __init__(self, **kwargs: Any):
             environment variable.
         """
         super().__init__(**kwargs)
-        infer_path = "{base_url}/chat/completions"
-        # not all chat models are on https://integrate.api.nvidia.com/v1,
-        # those that are not are served from their own endpoints
-        if model := determine_model(self.model):
-            if model.endpoint:  # some models have custom endpoints
-                infer_path = model.endpoint
         self._client = _NVIDIAClient(
             base_url=self.base_url,
             model=self.model,
             api_key=kwargs.get("nvidia_api_key", kwargs.get("api_key", None)),
-            infer_path=infer_path,
+            infer_path="{base_url}/chat/completions",
         )
         # todo: only store the model in one place
         # the model may be updated to a newer name during initialization

libs/ai-endpoints/langchain_nvidia_ai_endpoints/embeddings.py

@@ -69,17 +69,11 @@ def __init__(self, **kwargs: Any):
             environment variable.
         """
         super().__init__(**kwargs)
-        infer_path = "{base_url}/embeddings"
-        # not all embedding models are on https://integrate.api.nvidia.com/v1,
-        # those that are not are served from their own endpoints
-        if model := determine_model(self.model):
-            if model.endpoint:  # some models have custom endpoints
-                infer_path = model.endpoint
         self._client = _NVIDIAClient(
             base_url=self.base_url,
             model=self.model,
             api_key=kwargs.get("nvidia_api_key", kwargs.get("api_key", None)),
-            infer_path=infer_path,
+            infer_path="{base_url}/embeddings",
         )
         # todo: only store the model in one place
         # the model may be updated to a newer name during initialization

libs/ai-endpoints/langchain_nvidia_ai_endpoints/reranking.py

@@ -62,17 +62,11 @@ def __init__(self, **kwargs: Any):
             environment variable.
         """
         super().__init__(**kwargs)
-        infer_path = "{base_url}/ranking"
-        # not all models are on https://integrate.api.nvidia.com/v1,
-        # those that are not are served from their own endpoints
-        if model := determine_model(self.model):
-            if model.endpoint:  # some models have custom endpoints
-                infer_path = model.endpoint
         self._client = _NVIDIAClient(
             base_url=self.base_url,
             model=self.model,
             api_key=kwargs.get("nvidia_api_key", kwargs.get("api_key", None)),
-            infer_path=infer_path,
+            infer_path="{base_url}/ranking",
         )
         # todo: only store the model in one place
         # the model may be updated to a newer name during initialization