Mistral-7B-Instruct-v0.3 fine-tuned to write Italian crossword clues
Given an Italian text, a keyword and a clue type, this model writes a crossword clue for that keyword. It is a LoRA adapter for Mistral-7B-Instruct-v0.3 trained on Italian-Clue-Instruct.
From Harnessing LLMs for Educational Content-Driven Italian Crossword Generation (CLiC-it 2024). Earlier and follow-up work on Italian educational crosswords: CLiC-it 2023 and IJCoL 2025. Code: Italian-Text-to-Crosswords.
bare_noun_phrase · definite_article_phrase · clitic · copular_sentence — see the
repository for definitions and examples.
The generator script handles prompting, clue types and output parsing:
git clone https://github.com/KamyarZeinalipour/Italian-Text-to-Crosswords && cd Italian-Text-to-Crosswords
pip install -r requirements.txt
python main.py --model-name Kamyar-zeinalipour/Mistral-7B-ITA-Text-to-Cross --input-file input.csv --output-file output.csv --clue-type copular_sentence
ROUGE against GPT-4o reference clues, base → fine-tuned:
| Model | ROUGE-1 | ROUGE-2 | ROUGE-L |
|---|---|---|---|
| Mistral-7B-Instruct-v0.3 | 0.342 → 0.611 | 0.176 → 0.458 | 0.261 → 0.556 |
| Llama-3-8B-Instruct | 0.258 → 0.552 | 0.112 → 0.403 | 0.198 → 0.501 |
| Base model | Adapter |
|---|---|
| Llama-3-8B-Instruct | Llama3-8B-ITA-Text-to-Cross |
| Mistral-7B-Instruct-v0.3 | Mistral-7B-ITA-Text-to-Cross (this model) |
@inproceedings{zeinalipour2024harnessing,
title = {Harnessing LLMs for Educational Content-Driven Italian Crossword Generation},
author = {Kamyar Zeinalipour and Achille Fusco and Asya Zanollo and Marco Maggini and Marco Gori},
booktitle = {Proceedings of the 10th Italian Conference on Computational Linguistics (CLiC-it 2024)},
year = {2024},
url = {https://aclanthology.org/2024.clicit-1.110/}
}
Released under the base model's license (apache-2.0).
adapter_model.safetensors 623 MBrng_state_0.pth 15 kBrng_state_1.pth 15 kBrng_state_2.pth 15 kBrng_state_3.pth 15 kBtraining_args.bin 6.1 kBscheduler.pt 1.1 kBmistralai/Mistral-7B-Instruct-v0.3