community[patch]: allows using `text-generation-inference` /generate route with `HuggingFaceEndpoint` (#20100)

- **Description:** allows to use the /generate route of
`text-generation-inference` with the `HuggingFaceEndpoint`

libs/community/langchain_community/llms/huggingface_endpoint.py

@@ -258,7 +258,10 @@ class HuggingFaceEndpoint(LLM):
                 stream=False,
                 task=self.task,
             )
-            response_text = json.loads(response.decode())[0]["generated_text"]
+            try:
+                response_text = json.loads(response.decode())[0]["generated_text"]
+            except KeyError:
+                response_text = json.loads(response.decode())["generated_text"]
 
             # Maybe the generation has stopped at one of the stop sequences:
             # then we remove this stop sequence from the end of the generated text
@@ -289,7 +292,10 @@ class HuggingFaceEndpoint(LLM):
                 stream=False,
                 task=self.task,
             )
-            response_text = json.loads(response.decode())[0]["generated_text"]
+            try:
+                response_text = json.loads(response.decode())[0]["generated_text"]
+            except KeyError:
+                response_text = json.loads(response.decode())["generated_text"]
 
             # Maybe the generation has stopped at one of the stop sequences:
             # then remove this stop sequence from the end of the generated text