Feature/langchain integration for agent #9
Conversation
Co-authored-by: andresdiverGPT <[email protected]>
Co-authored-by: andresdiverGPT <[email protected]>
Co-authored-by: andresdiverGPT <[email protected]>
Co-authored-by: andresdiverGPT <[email protected]>
|
|
||
| functions = [format_tool_to_openai_function(tool) for tool in tools] | ||
| llm = ChatOpenAI(temperature=0, model="gpt-4-1106-preview") | ||
| model = ChatOpenAI(temperature=0, model="gpt-4-1106-preview").bind(functions=functions) |
There was a problem hiding this comment.
It seems like you are initializing two separate instances of ChatOpenAI which are exactly the same.
Consider creating a single instance and then using it across your code to save resources and avoid redundancy.
| model = ChatOpenAI(temperature=0, model="gpt-4-1106-preview").bind(functions=functions) | |
| llm = ChatOpenAI(temperature=0, model="gpt-4-1106-preview") | |
| model = llm.bind(functions=functions) |
| functions = [format_tool_to_openai_function(tool) for tool in tools] | ||
| llm = ChatOpenAI(temperature=0, model="gpt-4-1106-preview") | ||
| model = ChatOpenAI(temperature=0, model="gpt-4-1106-preview").bind(functions=functions) | ||
| prompt = ChatPromptTemplate.from_messages([ |
There was a problem hiding this comment.
The variable system_message is used but not defined in this scope, which would result in a NameError.
Make sure that system_message is defined before using it.
| functions = [format_tool_to_openai_function(tool) for tool in tools] | |
| llm = ChatOpenAI(temperature=0, model="gpt-4-1106-preview") | |
| model = ChatOpenAI(temperature=0, model="gpt-4-1106-preview").bind(functions=functions) | |
| prompt = ChatPromptTemplate.from_messages([ | |
| # Assuming system_message is defined elsewhere and imported correctly | |
| system_message = "Your system message here" |
| """ | ||
|
|
||
| functions = [format_tool_to_openai_function(tool) for tool in tools] | ||
| llm = ChatOpenAI(temperature=0, model="gpt-4-1106-preview") |
There was a problem hiding this comment.
Be cautious with hardcoding model strings like "gpt-4-1106-preview". It's recommended to use a configuration file or environment variables
for such parameters, to increase the flexibility and maintainability of your code.
| llm = ChatOpenAI(temperature=0, model="gpt-4-1106-preview") | |
| # Example using environment variable | |
| llm = ChatOpenAI(temperature=0, model=os.environ.get("OPENAI_MODEL", "gpt-3")) |
| self.chat_agent = AgentExecutor(agent=chain, memory=memory, tools=tools, verbose=True) | ||
|
|
There was a problem hiding this comment.
In the instantiation of AgentExecutor, the verbose flag is set to True. Ensure that this does not leak sensitive
information to the logs in a production environment.
| self.chat_agent = AgentExecutor(agent=chain, memory=memory, tools=tools, verbose=True) | |
| self.chat_agent = AgentExecutor(agent=chain, memory=memory, tools=tools, verbose=False) |
| def create_query_engine(self, agents): | ||
| """ | ||
| Create a query engine using the specified agents. | ||
|
|
||
| Args: | ||
| agents (dict): A dictionary of agents to be used in the query engine. | ||
| """ | ||
| nodes = [ | ||
| IndexNode(text=f"Law Information about {file}", index_id=file) | ||
| for file in self.docs.keys() | ||
| ] | ||
|
|
||
| vector_index = VectorStoreIndex(nodes) | ||
| vector_retriever = vector_index.as_retriever(similarity_top_k=1) | ||
| recursive_retriever = RecursiveRetriever( | ||
| "vector", retriever_dict={"vector": vector_retriever}, query_engine_dict=agents, verbose=True | ||
| ) | ||
|
|
||
| response_synthesizer = get_response_synthesizer(response_mode="compact") | ||
| self.query_engine = RetrieverQueryEngine.from_args( | ||
| recursive_retriever, response_synthesizer=response_synthesizer, service_context=self.service_context | ||
| ) |
There was a problem hiding this comment.
This method create_query_engine relies on a parameter agents that is not passed to it,
leading to a potential NameError. Ensure that all necessary parameters are being passed and used correctly.
| def create_query_engine(self, agents): | |
| """ | |
| Create a query engine using the specified agents. | |
| Args: | |
| agents (dict): A dictionary of agents to be used in the query engine. | |
| """ | |
| nodes = [ | |
| IndexNode(text=f"Law Information about {file}", index_id=file) | |
| for file in self.docs.keys() | |
| ] | |
| vector_index = VectorStoreIndex(nodes) | |
| vector_retriever = vector_index.as_retriever(similarity_top_k=1) | |
| recursive_retriever = RecursiveRetriever( | |
| "vector", retriever_dict={"vector": vector_retriever}, query_engine_dict=agents, verbose=True | |
| ) | |
| response_synthesizer = get_response_synthesizer(response_mode="compact") | |
| self.query_engine = RetrieverQueryEngine.from_args( | |
| recursive_retriever, response_synthesizer=response_synthesizer, service_context=self.service_context | |
| ) | |
| def create_query_engine(self, agents = None): | |
| if agents is None: | |
| agents = self.agents | |
| ... |
| index_path (str): The path to the index directory. | ||
| """ | ||
| self.llm = OpenAI(temperature=0, model="gpt-4-1106-preview") | ||
| self.service_context = ServiceContext.from_defaults(llm=self.llm) |
There was a problem hiding this comment.
Creating a ServiceContext could be a heavyweight operation. Please verify that the context is supposed to be initiated with each Retriever instance,
or if it can be shared amongst them as a single instance, improving efficiency.
| self.service_context = ServiceContext.from_defaults(llm=self.llm) | |
| # Assuming that ServiceContext can be created once and reused | |
| self.service_context = shared_service_context |
| ) | ||
| ] | ||
|
|
||
| function_llm = OpenAI(model="gpt-4-1106-preview") |
There was a problem hiding this comment.
The instantiation of OpenAI here seems to be redundant with the one initialized in __init__ as self.llm.
Consider reusing self.llm instead of creating a new instance.
| function_llm = OpenAI(model="gpt-4-1106-preview") | |
| self.agents[file] = OpenAIAgent.from_tools(query_engine_tools, llm=self.llm, verbose=True) |
| @@ -0,0 +1 @@ | |||
| """PLACEHOLDER FOR THE SYSTEM MESSAGE""" | |||
There was a problem hiding this comment.
The placeholder here must be replaced with actual system message content or configuration.
Ensure this placeholder has been appropriately updated before merging.
| """PLACEHOLDER FOR THE SYSTEM MESSAGE""" | |
| SYSTEM_MESSAGE_CONTENT = "Actual system message content goes here." |
No description provided.