Make LangChain bind_tools work with Granite function calling

[adampingel]

While using Replicate Granite model via Langchain Langgraph following error is encountered for agent creation

from langchain_community.llms import Replicate
from ibm_granite_community.notebook_utils import set_env_var, get_env_var
   
llm = Replicate (
    model = "ibm-granite/granite-20b-code-instruct-8k",
    model_kwargs={"temperature": 0.3,}
)

AttributeError: 'Replicate' object has no attribute 'bind_tools'

# Research agent and node
research_agent = create_agent(
    llm,
    [tavily_tool],
    system_message="You should provide accurate data for the chart_generator to use.",
)

[fayvor]

Taking a look at this.

[fayvor]

It looks like we can just bypass bind_tools if we want to, when using the Granite models (on Replicate or WatsonX). This PR of the Function_Calling recipe just calls invoke() on the WatsonX llm client, using a properly-constructed prompt.

The same thing works for Replicate; I've tested that.

[sree24lakshmi]

Hi @fayvor ,

If we were to use simple sql agents from langchain to leverage its specialized text-to-sql capabilities (tool-calling) below error is encountered.
In addition while building agentic workflow using langgraph for tool invocation, bind_tools seems to be a requirement.

Can you please advise if/when tool-calling can be supported/integrated.

agent_executor = create_sql_agent(llm, toolkit=toolkit, agent_type="tool-calling", verbose=True)

ValueError: This function requires a .bind_tools method be implemented on the LLM.