feat: optimize xinference stream (#989)

Author: takatost

api/core/third_party/langchain/llms/xinference_llm.py

@@ -108,12 +108,12 @@ def _stream_generate(
         Yields:
             A string token.
         """
-        if isinstance(model, RESTfulGenerateModelHandle):
-            streaming_response = model.generate(
+        if isinstance(model, (RESTfulChatModelHandle, RESTfulChatglmCppChatModelHandle)):
+            streaming_response = model.chat(
                 prompt=prompt, generate_config=generate_config
             )
         else:
-            streaming_response = model.chat(
+            streaming_response = model.generate(
                 prompt=prompt, generate_config=generate_config
             )