How to use JSON format equivalent in ChatGroq from langchain_groq?

[dinhvanlinh0610]

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_groq import ChatGroq

llm_json_mode = ChatGroq(model="llama3-8b-8192",temperature=0, format="json")

response = llm_json_mode.invoke("Test")

Description

I am using langchain with the langchain_ollama and langchain_groq integrations to process natural language tasks.

When working with ChatOllama from langchain_ollama, I can use the format="json" parameter like this:

from langchain_ollama import ChatOllama

llm_json_mode = ChatOllama(model=local_llm, temperature=0, format="json")

This ensures that the output is formatted as JSON. However, when using ChatGroq from langchain_groq, I couldn’t find a similar format parameter. Here’s how I currently initialize ChatGroq:

from langchain_groq import ChatGroq

llm_groq_mode = ChatGroq(model=local_llm, temperature=0, format="json")

And Error:
response = self.client.create(messages=message_dicts, **params)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: Completions.create() got an unexpected keyword argument 'format'

I want the output to be in JSON format, but there doesn't seem to be a format argument for ChatGroq. Is there an equivalent parameter or a way to ensure JSON output when using ChatGroq?

Any help or guidance would be appreciated!

System Info

Additional Details
LangChain version: 0.3.10
LangChain_Groq version: 0.2.1
LangChain_Ollama version: 0.2.1
Python version: Python 3.12.0

[feijoes]

@dinhvanlinh0610 You can achieve structured output with ChatGroq by using the with_structured_output method with json_mode

from langchain_groq import ChatGroq
llm = ChatGroq(model="", temperature=0)
structured_llm = llm.with_structured_output(method="json_mode", include_raw=True)

structured_llm.invoke("")

docs