NotImplementedError for bind_tools method in AzureMLChatOnlineEndpoint class

[informatica92]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

import os

from langchain_openai import AzureChatOpenAI
from langchain_community.chat_models.azureml_endpoint import AzureMLChatOnlineEndpoint,  AzureMLEndpointApiType, CustomOpenAIChatContentFormatter
from langchain_community.utilities import SQLDatabase
from langchain_community.agent_toolkits.sql.toolkit import SQLDatabaseToolkit
from langchain_community.tools.sql_database.tool import (
    InfoSQLDatabaseTool,
    ListSQLDatabaseTool,
    QuerySQLCheckerTool,
    QuerySQLDataBaseTool,
)
from langchain import hub
from langgraph.prebuilt import create_react_agent
from langchain_core.language_models import BaseChatModel

if __name__ == "__main__":
    db = SQLDatabase.from_uri("sqlite:///Chinook.db")

    url = os.environ.get("AZUREML_ENDPOINT_URL")
    key = os.environ.get("AZUREML_ENDPOINT_KEY")
    timeout = 60 * 5  # default = 60 * 5 = 5 minutes

    llm = AzureMLChatOnlineEndpoint(
        endpoint_url=url,
        endpoint_api_type=AzureMLEndpointApiType.serverless,
        endpoint_api_key=key,
        content_formatter=CustomOpenAIChatContentFormatter(),
        timeout=timeout  # default = 60 * 5 = 5 minutes
    )

    toolkit = SQLDatabaseToolkit(db=db, llm=llm)

    prompt_template = hub.pull("langchain-ai/sql-agent-system-prompt")

    system_message = prompt_template.format(dialect="SQLite", top_k=5)

    agent_executor = create_react_agent(llm, toolkit.get_tools(), state_modifier=system_message)

    example_query = "Which country's customers spent the most?"

    events = agent_executor.stream(
        {"messages": [("user", example_query)]},
        stream_mode="values",
    )
    for event in events:
        event["messages"][-1].pretty_print()

Description

I am trying to use the tutorial reported here to create a simple SQL agent powered by an LLM.

My issue is about the AzureMLChatOnlineEndpoint class I'd like to use instead of the one included into the tutorial (AzureChatOpenAI) since I'd like to switch from a GPT-X model to an open-weights one hosted in my Azure infrastructure.
The LLM is a Meta LLama 3.1 70B for the record.

Well, in this situation reported above, if I run the code, I get the following error: "NotImplementedError" for the bind_tools method.

Full stack trace reported below:

Traceback (most recent call last):
  File "C:\Users\...\git\Text2SQL\text2sql\main_agent_llama.py", line 59, in <module>
    agent_executor = create_react_agent(llm, toolkit.get_tools(), state_modifier=system_message)
                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\...\git\Text2SQL\Text2SQL - ENV\Lib\site-packages\langgraph\_api\deprecation.py", line 80, in wrapper
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\...\git\Text2SQL\Text2SQL - ENV\Lib\site-packages\langgraph\prebuilt\chat_agent_executor.py", line 512, in create_react_agent
    model = cast(BaseChatModel, model).bind_tools(tool_classes)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\...\git\Text2SQL\Text2SQL - ENV\Lib\site-packages\langchain_core\language_models\chat_models.py", line 1115, in bind_tools
    raise NotImplementedError
NotImplementedError

What am I missing? Can I integrate the bind_tools implementation somehow? Is there another class instead of AzureMLChatOnlineEndpoint to access Azure AI studio deployed models to do this?

System Info

langchain==0.3.6
langchain-community==0.3.4
langchain-core==0.3.14
langchain-openai==0.2.4
langchain-text-splitters==0.3.1
langgraph==0.2.41
langgraph-checkpoint==2.0.2
langgraph-sdk==0.1.35
langsmith==0.1.138

Platform WINDOWS

Python version 3.11

[feijoes]

Try using create_sql_agent_

from langchain.agents.agent import AgentExecutor
from langchain_community.agent_toolkits import create_sql_agent
agent_executor: AgentExecutor = create_sql_agent(
    llm=llm,
    toolkit=toolkit, 
)
agent_executor.run("Which country's customers spent the most?")

[informatica92]

Unfortunately this doesn't work either.

import os

from langchain_community.chat_models.azureml_endpoint import AzureMLChatOnlineEndpoint,  AzureMLEndpointApiType, CustomOpenAIChatContentFormatter
from langchain_community.utilities import SQLDatabase
from langchain_community.agent_toolkits.sql.toolkit import SQLDatabaseToolkit

from langchain import hub

from langchain.agents.agent import AgentExecutor
from langchain_community.agent_toolkits import create_sql_agent

from langchain_groq import ChatGroq


if __name__ == "__main__":
    db = SQLDatabase.from_uri("sqlite:///Chinook.db")

    url = os.environ.get("AZUREML_ENDPOINT_URL")
    key = os.environ.get("AZUREML_ENDPOINT_KEY")
    timeout = 60 * 5  # default = 60 * 5 = 5 minutes

    llm = AzureMLChatOnlineEndpoint(
        endpoint_url=url,
        endpoint_api_type=AzureMLEndpointApiType.serverless,
        endpoint_api_key=key,
        content_formatter=CustomOpenAIChatContentFormatter(),
        timeout=timeout  # default = 60 * 5 = 5 minutes
    )

    # llm = ChatGroq(model="llama3-8b-8192", api_key="...")

    toolkit = SQLDatabaseToolkit(db=db, llm=llm)

    agent_executor: AgentExecutor = create_sql_agent(
        llm=llm,
        toolkit=toolkit,
        # verbose=True,
        handle_parsing_error=True,
        return_intermediate_steps=True
    )

    example_query = "Which country's customers spent the most?"

    response = agent_executor.invoke({"input": example_query})

    print(response)

Executing the above code using the Azure LLM I get:

ValueError: An output parsing error occurred. In order to pass this error back to the agent and have it try again, pass `handle_parsing_errors=True` to the AgentExecutor. 

This is the error: Parsing LLM output produced both a final answer and a parse-able action:: Thought: I should query the schema of the most relevant tables, such as Customer and Invoice.\nAction: sql_db_schema\nAction Input: Customer, Invoice\nObservation: 

tested multiple times, always same result.

If I switch to Groq LLM the same code works well returning

{'input': "Which country's customers spent the most?", 'output': 'The country whose customers spent the most is the USA, with a total spent of 523.06.'}

[feijoes]

@informatica92 it's strange that even with handle_parsing_error=True, you still get the same error. Try checking this solution: link.

[informatica92]

Found it! The issue is that the kwargs fir the AgentExecutor can not be passed as create_sql_agent kwargs but they need to be basses as dict into the agent_executor_kwargs parameter.
So basically, code needs to be changed to:

    agent_executor: AgentExecutor = create_sql_agent(
        llm=llm,
        toolkit=toolkit,
        # verbose=True,
        max_iterations=10,
        agent_executor_kwargs={  # <--- please note this 
            "handle_parsing_errors": True,
            "return_intermediate_steps": True
        }
    )

The complete code, then, looks like this:

import os

from langchain_community.chat_models.azureml_endpoint import AzureMLChatOnlineEndpoint,  AzureMLEndpointApiType, CustomOpenAIChatContentFormatter
from langchain_community.utilities import SQLDatabase
from langchain_community.agent_toolkits.sql.toolkit import SQLDatabaseToolkit

from langchain.agents.agent import AgentExecutor
from langchain_community.agent_toolkits import create_sql_agent

if __name__ == "__main__":
    db = SQLDatabase.from_uri("sqlite:///Chinook.db")

    url = os.environ.get("AZUREML_ENDPOINT_URL")
    key = os.environ.get("AZUREML_ENDPOINT_KEY")
    timeout = 60 * 5  # default = 60 * 5 = 5 minutes

    llm = AzureMLChatOnlineEndpoint(
        endpoint_url=url,
        endpoint_api_type=AzureMLEndpointApiType.serverless,
        endpoint_api_key=key,
        content_formatter=CustomOpenAIChatContentFormatter(),
        timeout=timeout  # default = 60 * 5 = 5 minutes
    )

    toolkit = SQLDatabaseToolkit(db=db, llm=llm)

    agent_executor: AgentExecutor = create_sql_agent(
        llm=llm,
        toolkit=toolkit,
        # verbose=True,
        max_iterations=10,
        agent_executor_kwargs={
            "handle_parsing_errors": True,
            "return_intermediate_steps": True
        }
    )

    example_query = "Which country's customers spent the most?"

    response = agent_executor.invoke({"input": example_query})

    print(response)

Output:

{
    "input": "Which country's customers spent the most?",
    "output": "The country whose customers spent the most is the USA, with a total of $523.06.",
    "intermediate_steps": [<list of tuple (AgentAction, step_output)>]
}