llm-roleplay 2.0.5

LLM Roleplay: Simulating Human-Chatbot Interaction

LLM Roleplay: Simulating Human-Chatbot Interaction

Roleplay is a Python package that provides an easy method for generating goal-oriented, persona-based multi-turn dialogues, simulating diverse human-chatbot interactions. This repository contains the code and data for the LLM Roleplay method, as presented in the paper LLM Roleplay: Simulating Human-Chatbot Interaction. It includes all the experiment codes and the necessary data to replicate them, as described in the paper.

More About Roleplay

The development of chatbots requires collecting a large number of human-chatbot dialogues to reflect the breadth of users' sociodemographic backgrounds and conversational goals. However, the resource requirements to conduct the respective user studies can be prohibitively high and often only allow for a narrow analysis of specific dialogue goals and participant demographics. In this paper, we propose LLM-Roleplay: a goal-oriented, persona-based method to automatically generate diverse multi-turn dialogues simulating human-chatbot interaction. LLM-Roleplay can be applied to generate dialogues with any type of chatbot and uses large language models (LLMs) to play the role of textually described personas. To validate our method we collect natural human-chatbot dialogues from different sociodemographic groups and conduct a human evaluation to compare real human-chatbot dialogues with our generated dialogues. We compare the abilities of state-of-the-art LLMs in embodying personas and holding a conversation and find that our method can simulate human-chatbot dialogues with a high indistinguishability.

The LLM Roleplay (llm-roleplay) codebase is built upon the UrarTU framework (version 2). For detailed insights into its structure, please refer to the Getting Started Guide.

Installation

Getting started with llm-roleplay is a breeze! 💨 Just follow these steps to set up the necessary packages and create a local package called llm-roleplay:

• Clone the repository: git clone [e-mail]:UKPLab/llm-roleplay.git
• Navigate to the project directory: cd llm-roleplay
• Execute the magic command: pip install .

🪄 After running the previous command, llm-roleplay will install the required packages including the latest version of urartu (>=2.0) and make it ready to use. Plus, an alias will be created, allowing you to access llm-roleplay from any directory on your operating system effortlessly:

urartu --help

Now, to register llm-roleplay under the corresponding name in urartu we need to run the following command by providing the path where the module is located, for more info refere to UrarTU's documentation:

urartu register --name=llm_roleplay --path=PATH_TO_ROLEPLAY/llm_roleplay

After this you can run urartu -h again to see the available modules under launch command and make sure that llm_roleplay is present there.

Exploring the Experiments

Before diving into using llm-roleplay, let's set up Aim. This tool will track our experiment metadata and generated dialogues, storing them locally on our system.

Let's start the Aim server to store all the metadata and dialogues of our experiments. By default, it will run on port 53800. Use this command to get it running:

aim server

Since we are running the Aim server on our local machine, we will use the address: aim://0.0.0.0:53800. For remote tracking, refer to Track experiments with aim remote server.

To explore the wealth of metrics that Aim captures effortlessly, follow these steps:

• Navigate to the directory containing the .aim repository.
• Run the command that sparks the magic:

aim up

Usage

Let's get started with generating dialogues using the llm-roleplay action. The process is simple: just provide the name of the configuration file containing the action, followed by the action name itself. For the llm-roleplay action, we'll initiate it by using the Mistral 8x7B model as the inquirer. 🎇

urartu launch --name=llm_roleplay action_config=dialogue_generator +action_config/task/model_inquirer=mixtral +action_config/task/model_responder=llama action_config.task.model_inquirer.api_token="YOUR_TOKEN"

The action_config parameter specifies which configuration file to use to run the action. Afterward, we define the configuration file for the inquirer using the model_inquirer argument and set the configuration for the responder with the model_responder argument.

To execute the command on a Slurm cluster, modify the llm_roleplay/configs/action_config/dialogue_generator.yaml file with the corresponding fields, and then use the same command to run the job. For more details on how to edit the configuration files, please refer to the upcoming sections.

Huggingface Authentication You might need to log in to HuggingFace to authenticate your use of Mistral 8x7B. To do this, use the huggingface-cli login command and provide your access token. To obtain a HuggingFace access token, refer to User access tokens.

Configs: Tailoring Your Setup

The default configs which shape the way of configs are defined in urartu under urartu/config directory:

• urartu/config/main.yaml: This core configuration file sets the foundation for default settings, covering all available keys within the system.
• urartu/config/action_config Directory: A designated space for specific action configurations. For more see the structure of UrarTU.

Crafting Customizations

You have two flexible options for tailoring your configurations in llm-roleplay.

1. Custom Config Files: To simplify configuration adjustments, llm-roleplay provides a dedicated configs directory where you can store personalized configuration files. These files seamlessly integrate with Hydra's search path. The directory structure mirrors that of urartu/config. You can define project-specific configurations in specially named files. The dialogue_generator.yaml file within the configs directory houses all the configurations specific to our llm-roleplay project, with customized settings.

   - **Personalized User Configs**: To further tailor configurations for individual users, create a directory named `configs_{username}` at the same level as the `configs` directory, where `{username}` represents your operating system username (check out `configs_tamoyan` for an example). The beauty of this approach is that there are no additional steps required. Your customizations will smoothly load and override the default configurations, ensuring a seamless and hassle-free experience. ✨
   
   The order of precedence for configuration overrides is as follows: `urartu/config`, `llm_roleplay/configs`, `llm_roleplay/configs_{username}`, giving priority to user-specific configurations.
   

2. CLI Approach: For those who prefer a command-line interface (CLI) approach, urartu offers a convenient method. You can enhance your commands with specific key-value pairs directly in the CLI. For example, modifying your working directory path is as simple as:

   urartu launch --name=llm_roleplay action_config=dialogue_generator action_config.workdir=PATH_TO_WORKDIR
   

Choose the method that suits your workflow best and enjoy the flexibility urartu provides for crafting custom configurations.

Effortless Launch

With urartu, launching actions is incredibly easy, offering you two options. 🚀

• Local Marvel: This option allows you to run jobs on your local machine, right where the script is executed.
• Cluster Voyage: This choice takes you on a journey to the Slurm cluster. By adjusting the slurm.use_slurm setting in llm_roleplay/configs/action_config/dialogue_generator.yaml, you can easily switch between local and cluster execution.

Enjoy the flexibility to choose the launch adventure that best suits your needs and goals!

---

You're all set to dive into goal-oriented, persona-based, diverse, and multi-turn dialogue generation with llm-roleplay! 🌟 If you encounter any issues or have suggestions, feel free to open an issue for assistance. 😊

Cite

Please use the following citation:

@misc{tamoyan2024llmroleplaysimulatinghumanchatbot,
  title={LLM Roleplay: Simulating Human-Chatbot Interaction},
  author={Hovhannes Tamoyan and Hendrik Schuff and Iryna Gurevych},
  year={2024},
  eprint={2407.03974},
  archivePrefix={arXiv},
  primaryClass={cs.CL},
  url={https://arxiv.org/abs/2407.03974},
}

Contacts

Hovhannes Tamoyan, Hendrik Schuff

Please feel free to contact us if you have any questions or need to report any issues.

Links

UKP Lab Homepage | TU Darmstadt Website

Disclaimer

This repository contains experimental software and is published for the sole purpose of giving additional background details on the respective publication.