Forked from the official implementation of this ACL 2024 paper:
Nian Li, Chen Gao, et al. EconAgent: Large Language Model-Empowered Agents for Simulating Macroeconomic Activities, ACL 2024.
It is used in the Economics Course (走进经济学) at Ocean University of China for the final assignment of Fall 2024.
Considering that the original implementation is based on chatGPT API, and it is relatively expensive for students (who may not be a professional researcher but just a amateur). And the original work used GPT-3.5 API to be specific, which is not in service anymore. So we use Ollama to replace it.
Install Ollama using one command:
curl -fsSL https://ollama.com/install.sh | shGet llama3.1 model by:
ollama run install llama3.1
Simulate with Ollama, 100 agents, and 240 months:
python simulate_ollama.py --num_agents 100 --episode_length 240
You can modify --num_agents, --episode_length, --dialog_len, --max_price_inflation, and --max_wage_inflation to change the parameters of the simulation.
You can change the model by modifying ollama_model in simulate_utils.py.
Generated file will be saved in data/ directory, you can change it in simulate_utils.py.
Pickle file can be loaded as follows:
import pickle
file_name = 'Your .pkl file name'
with open(file_name, 'rb') as f:
data = pickle.load(f)