Value Error when instantiating LLMGraphTransformer object using AzureChatOpenAI related to Native Function Calling Support

[vincentsyha]

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

from langchain_experimental.graph_transformers import LLMGraphTransformer
from langchain_openai import AzureChatOpenAI

# Set up chat language model
chat_model_azure = AzureChatOpenAI(
    azure_endpoint=AZURE_LLM_ENDPOINT,
    deployment_name="gpt-4o",
    openai_api_version=AZURE_API_VERSION,
    openai_api_key=AZURE_LLM_API_KEY,
    temperature=0,
)

llm_transformer = LLMGraphTransformer(
  llm=chat_model_azure, 
  node_properties=["description"],
  relationship_properties=["description"]
)

Description

Using gpt-4o deployed version is 2024-05-13

When attempting to instantiate a LLMGraphTransformer object using an AzureChatOpenAI object I am met with the following error:

ValueError: The 'node_properties' and 'relationship_properties' parameters cannot be used in combination with a LLM that doesn't support native function calling.

According to the OpenAI documentation this version of GPT 4o does support function calling. Could this at all be related to Microsoft deprecating the function_calling parameter that the LLMGraphTransformer() checks for?

If so is there any workaround at all?

System Info

System Information

OS: Darwin
OS Version: Darwin Kernel Version 23.5.0: Wed May 1 20:12:58 PDT 2024; root:xnu-10063.121.3~5/RELEASE_ARM64_T6000
Python Version: 3.11.5 | packaged by conda-forge | (main, Aug 27 2023, 03:35:23) [Clang 15.0.7 ]

Package Information

langchain_core: 0.2.22
langchain: 0.2.9
langchain_community: 0.2.7
langsmith: 0.1.92
langchain_experimental: 0.0.62
langchain_google_vertexai: 0.0.5
langchain_openai: 0.0.3
langchain_text_splitters: 0.2.2

Packages not installed (Not Necessarily a Problem)

The following packages were not found:

langgraph
langserve

[dosubot[bot]]

Hey there, @vincentsyha!

I'm here to help you out with any bugs, questions, and even becoming a contributor. Let's see what we can do about that issue you're facing.

To work around the ValueError when instantiating an LLMGraphTransformer object using an AzureChatOpenAI object, you can set the node_properties and relationship_properties parameters to False. This avoids the need for native function calling support.

Here is an example of how to instantiate the LLMGraphTransformer object with the workaround:

from langchain_experimental.graph_transformers import LLMGraphTransformer
from langchain_openai import AzureChatOpenAI

# Set up chat language model
chat_model_azure = AzureChatOpenAI(
    azure_endpoint=AZURE_LLM_ENDPOINT,
    deployment_name="gpt-4o",
    openai_api_version=AZURE_API_VERSION,
    openai_api_key=AZURE_LLM_API_KEY,
    temperature=0,
)

# Instantiate the LLMGraphTransformer object with node_properties and relationship_properties set to False
llm_transformer = LLMGraphTransformer(
    llm=chat_model_azure,
    allowed_nodes=["Person", "Company", "Product", "Award", "Characteristic"],
    allowed_relationships=["WORKS_FOR", "HAS_AWARD", "PRODUCED_BY", "HAS_CHARACTERISTIC"],
    node_properties=False,
    relationship_properties=False
)

By setting node_properties and relationship_properties to False, you can avoid the ValueError and successfully instantiate the LLMGraphTransformer object ^[1][2].

To continue talking to Dosu, mention @dosu.

[rathcoding]

Same problem.

I'm trying to use the LLMGraphTransformer with "llama3-groq-tool-use" via Ollama.

Tried node_properties=True and node_properties=['name', 'document_id'] but got the same error.

EDIT: Nevermind... just found out that the "langchain_community.llms.Ollama" class does not implement the "with_structured_output()" method needed for function calling. =(

[vincentsyha]

In my case the problem was that the version of langchain-openai in my environment was too low.

Upgrading to 0.1.17 resolved this issue for me.