expose memory key name

[hwchase17]

No description provided.

[rowillia]

Is this constant? If so you can use the const parameter from Pydantic:

chat_history_key: str = Field("history", const=True)

[VBoB13]

While I appreciate the efforts, I still can't use this feature due to the very same issue.
Error:

| ERROR | 2023-03-03 18:03:15 | /api/prompt/prompt.py | PromptMaker Could NOT execute LangChain's agent code!
Details:
Missing some input keys: {'chat_history'}  File "/opt/gpt3/api/prompt/prompt.py", line 134, in _run
    results = agent_chain.run(
  File "/usr/local/lib/python3.10/site-packages/langchain/chains/base.py", line 242, in run
    return self(kwargs)[self.output_keys[0]]
  File "/usr/local/lib/python3.10/site-packages/langchain/chains/base.py", line 132, in __call__
    inputs = self.prep_inputs(inputs)
  File "/usr/local/lib/python3.10/site-packages/langchain/chains/base.py", line 221, in prep_inputs
    self._validate_inputs(inputs)
  File "/usr/local/lib/python3.10/site-packages/langchain/chains/base.py", line 101, in _validate_inputs
    raise ValueError(f"Missing some input keys: {missing_keys}")

But my code does indeed specify to look for 'chat_history'; in multiple places even:

prompt = ConversationalAgent.create_prompt(
    tools,
    prefix=TEMPLATE_OBJECT[self.language]["prefix"],
    suffix=TEMPLATE_OBJECT[self.language]["suffix"],
    input_variables=['input', 'chat_history',
                     'agent_scratchpad', 'context']
)
memory = ConversationBufferMemory(memory_key="chat_history")
llm_chain = LLMChain(llm=self.llm, prompt=prompt)
tools = load_tools(["serpapi", "llm-math"], llm=self.llm)
# ConversationalAgent seems to have issues with chat_history
# TODO: Check ConversationalAgent + {chat_history} !!!
# agent = ConversationalAgent(
#     llm_chain=llm_chain, tools=tools, verbose=True)
agent = ConversationalAgent(
    llm_chain=llm_chain, tools=tools, verbose=True)
agent_chain = AgentExecutor.from_agent_and_tools(
    agent, tools, verbose=True, mempry=memory)
results = agent_chain.run(
    input=self.prompt, context=self.context)

Is there anything I am missing? Am I the one being silly? 😢

[VBoB13]

Oh my goodness.... It was all my fault in the end.... Spelling mistakes 😭

agent_chain = AgentExecutor.from_agent_and_tools(
    agent, tools, verbose=True, mempry=memory)

-->

agent_chain = AgentExecutor.from_agent_and_tools(
    agent, tools, verbose=True, memory=memory)

Thanks for your great work! This tool is freakin' amazeballs :)

[postix]

about the chat_history ... how do you handle the sessions with the client ??? using a fingerprint or something to make a history not shared with others users ...

[VBoB13]

about the chat_history ... how do you handle the sessions with the client ??? using a fingerprint or something to make a history not shared with others users ...

For that, I just use session_id as an identifier and save each question and answer into a DB.
Then, for every time a user asks something from my Agent, I load memory through Elasticsearch and shove it into the Agent.

This is my code for loading the memory (part of a larger client class). You're welcome :)

def _load_memory(self, index: str, session: str):
        """
        Method that loads memory (if it exists).
        """
        history = ConversationBufferWindowMemory(
            k=3, return_messages=True, memory_key='chat_history')
        hist_index = "_".join(["hist", index, session])
        if self.indices.exists(index=hist_index).body:
            query = {
                "query": {
                    "match_all": {}
                },
                "sort": [
                    {
                        "timestamp": {
                            "order": "desc",
                            "unmapped_type": "date"
                        }
                    }
                ]
            }
            results = self.search(
                index=hist_index, query=query['query'], size=3, sort=query['sort'])
            hist_docs = results['hits']['hits']
            # self.logger.msg = "Documents found:"
            # self.logger.info(extra_msg=str(hist_docs))

            for doc in hist_docs:
                history.chat_memory.add_user_message(doc['_source']['user'])
                history.chat_memory.add_ai_message(doc['_source']['ai'])
        else:
            settings = {
                "settings": {
                    "index": {
                        "number_of_shards": 1,
                        "number_of_replicas": 0
                    }
                }
            }
            mappings = {
                "mappings": {
                    "properties": {
                        "user": {
                            "type": "text"
                        },
                        "ai": {
                            "type": "text"
                        },
                        "timestamp": {
                            "type": "date"
                        }
                    }
                }
            }
            mappings.update(settings)
            try:
                self.indices.create(
                    index=hist_index, mappings=mappings['mappings'], settings=mappings['settings'])
            except Exception as err:
                self.logger.msg = "Something went wrong when trying to create index " +\
                    Fore.LIGHTRED_EX + hist_index + Fore.RESET + "!"
                self.logger.error(extra_msg=str(err))
                raise self.logger from err

        return history