Project code of Human Machine Dialogue

Muraro Nicola 248449

This project is part of the course "Human Machine Dialogue" at the University of Trento.

The system presented here is a conversational agent based on Llama, created to be used within the gym domain.

Set up the environment

In order to create the conda environment with all the required dependencies, run the following command:

conda env create -f environment.yml

Execution of the code

The main program can be runned with the default configuration, by executing:

python3 ./main.py

Optional configuration flags

• --model_type: Source model to use, either 'hugginface' (standard) or 'ollama' (for fast local execution)
• --device: Device to use for the model, default is 'cuda:0'
• --logging: Enable logging for debugging purposes, logs will be saved in './out/log.txt'

Extrinsic evaluation

Local execution

To execute the instrinsic evaluation of the program, you can run the following command:

python3 ./evaluation.py

Execution on the cluster

The configuration file for the cluster is already set up. To run the evaluation on the cluster, you can use the following command:

sbatch evaluation.sbatch