fix: use openai llm via langchain (#670)

Co-authored-by: Deshraj Yadav <deshrajdry@gmail.com>

embedchain/embedchain.py

@@ -21,7 +21,8 @@ from embedchain.embedder.base import BaseEmbedder
 from embedchain.helper.json_serializable import JSONSerializable
 from embedchain.llm.base import BaseLlm
 from embedchain.loaders.base_loader import BaseLoader
-from embedchain.models.data_type import DataType, DirectDataType, IndirectDataType, SpecialDataType
+from embedchain.models.data_type import (DataType, DirectDataType,
+                                         IndirectDataType, SpecialDataType)
 from embedchain.utils import detect_datatype
 from embedchain.vectordb.base import BaseVectorDB
 

embedchain/llm/openai.py

@@ -1,6 +1,7 @@
 from typing import Optional
 
-import openai
+from langchain.chat_models import ChatOpenAI
+from langchain.schema import HumanMessage, SystemMessage
 
 from embedchain.config import BaseLlmConfig
 from embedchain.helper.json_serializable import register_deserializable
@@ -12,31 +13,32 @@ class OpenAILlm(BaseLlm):
     def __init__(self, config: Optional[BaseLlmConfig] = None):
         super().__init__(config=config)
 
-    # NOTE: This class does not use langchain. One reason is that `top_p` is not supported.
-
     def get_llm_model_answer(self, prompt):
-        messages = []
-        if self.config.system_prompt:
-            messages.append({"role": "system", "content": self.config.system_prompt})
-        messages.append({"role": "user", "content": prompt})
-        response = openai.ChatCompletion.create(
-            model=self.config.model or "gpt-3.5-turbo-0613",
-            messages=messages,
-            temperature=self.config.temperature,
-            max_tokens=self.config.max_tokens,
-            top_p=self.config.top_p,
-            stream=self.config.stream,
-        )
+        response = OpenAILlm._get_answer(prompt, self.config)
 
         if self.config.stream:
-            return self._stream_llm_model_response(response)
+            return response
         else:
-            return response["choices"][0]["message"]["content"]
+            return response.content
 
-    def _stream_llm_model_response(self, response):
-        """
-        This is a generator for streaming response from the OpenAI completions API
-        """
-        for line in response:
-            chunk = line["choices"][0].get("delta", {}).get("content", "")
-            yield chunk
+    def _get_answer(prompt: str, config: BaseLlmConfig) -> str:
+        messages = []
+        if config.system_prompt:
+            messages.append(SystemMessage(content=config.system_prompt))
+        messages.append(HumanMessage(content=prompt))
+        kwargs = {
+            "model": config.model or "gpt-3.5-turbo-0613",
+            "temperature": config.temperature,
+            "max_tokens": config.max_tokens,
+            "model_kwargs": {},
+        }
+        if config.top_p:
+            kwargs["model_kwargs"]["top_p"] = config.top_p
+        if config.stream:
+            from langchain.callbacks.streaming_stdout import \
+                StreamingStdOutCallbackHandler
+
+            chat = ChatOpenAI(**kwargs, streaming=config.stream, callbacks=[StreamingStdOutCallbackHandler()])
+        else:
+            chat = ChatOpenAI(**kwargs)
+        return chat(messages)