Does Langchain’s `create_csv_agent` and `create_pandas_dataframe_agent` functions work with non-OpenAl LLMs – Langchain

The Problem:

Langchain provides create_csv_agent and create_pandas_dataframe_agent functions for integrating non-OpenAI LLMs with its platform. However, the documentation only showcases examples using OpenAI APIs. Determine if these functions can work with non-OpenAI LLM models like Llama2 and Vicuna and provide guidance on how to use them with these alternative models.

The Solutions:

Solution 1: Using HuggingFace Hub to Load Non-OpenAl LLMs and Integrate with Langchain’s Agents

To use non-OpenAl Large Language Models (LLMs) with Langchain’s `create_csv_agent` and `create_pandas_dataframe_agent` functions, follow these steps:

1. Load the Non-OpenAl LLM from HuggingFace Hub:

   • Refer to this link for guidance on loading models from the HuggingFace Hub:
     HuggingFace Hub Integration
   • Obtain the repository ID (repo_id) of the desired non-OpenAl LLM model from the HuggingFace Hub.

2. Create the HuggingFaceHub LLM Object:

   • Import the necessary modules and set the HUGGINGFACEHUB_API_TOKEN environment variable with your API token.
   • Instantiate the HuggingFaceHub class with the repo_id and any desired model configuration options, such as temperature and max length.

3. Integrate the Non-OpenAl LLM with Langchain’s Agents:

   • Create a Langchain agent using the create_csv_agent or create_pandas_dataframe_agent function, specifying the HuggingFaceHub LLM object as the argument for the llm parameter.
   • Configure other optional parameters as needed, such as verbose, max_iterations, include_df_in_prompt, and so on.

By following these steps, you can seamlessly integrate non-OpenAl LLMs with Langchain’s agents to perform various natural language processing tasks using your preferred LLM model.

Video Explanation:

The following video, titled "”LangChain", provides additional insights and in-depth exploration related to the topics discussed in this post.

”Colab

(https://drp.li/m1mbM) Load HuggingFace models locally so that you can use models you can’t use …”]