mistral-hackaton-2026/ecotopia-citizens-ministral-8b

Ecotopia Citizens - Ministral 8B LoRA

Fine-tuned LoRA adapter for generating realistic citizen dialogue and reactions in the Ecotopia political simulation game. Citizens respond to mayor policies with contextual opinions based on their demographic profiles.

Model Details

• Base Model: mistralai/Ministral-8B-Instruct-2410
• Method: LoRA (r=16, alpha=32) + 4-bit QLoRA (NF4)
• Framework: TRL SFT + PEFT
• Training: HuggingFace Jobs (A10G GPU), 3 epochs
• Dataset: 340 examples (272 train / 68 val)

Usage

This is a LoRA adapter. Load with PEFT:

from peft import PeftModel
from transformers import AutoModelForCausalLM

base = AutoModelForCausalLM.from_pretrained("mistralai/Ministral-8B-Instruct-2410")
model = PeftModel.from_pretrained(base, "mistral-hackaton-2026/ecotopia-citizens-ministral-8b")

Links

• W&B Project
• GitHub Repo
• Extract Model

Team

Built for the Mistral Worldwide Hackathon 2026 (London) by Nolan Cacheux and Kyle Kreuter.