configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: test
path: data/test-*
dataset_info:
features:
- name: instruction
dtype: string
- name: response
dtype: string
- name: category
dtype: string
splits:
- name: train
num_bytes: 40388948.1573514
num_examples: 76011
- name: test
num_bytes: 2125957.8426486026
num_examples: 4001
download_size: 19780999
dataset_size: 42514906
Dataset Card: MTG-Eval
Dataset Summary
The MTG-Eval dataset is designed to enhance and evaluate the understanding of Magic: the Gathering (MTG) rules and card interactions by large language models. It consists of synthetic question-answer pairs generated from MTG card descriptions, official rulings, and card interactions. This dataset aims to reduce the hallucinations of language models and improve their performance in MTG-related tasks such as deck generation and in-game decision-making.
Dataset Details
- Name: MTG-Eval
- Number of Instances: 80,032
- Card Descriptions: 26,702
- Rules Questions: 27,104
- Card Interactions: 26,226
- Source: Data derived from MTGJSON and Commander Spellbook
Data Generation Methodology
- Card Descriptions: Reformatted rules text from MTG cards.
- Rules Questions: Reformatted rulings from the MTG Rules Committee.
- Card Interactions: Synthetic questions generated from combo descriptions in the Commander Spellbook database.
Use Cases
- Training Language Models: Improve language models' understanding of MTG rules and card interactions.
- Evaluation Benchmark: Assess performance of language models on MTG-related tasks.
Acknowledgments
Thanks to the MTGJSON project and the team at Commander Spellbook for generously sharing their dataset, without which this dataset would not be possible. All generated data is unofficial Fan Content permitted under the Fan Content Policy. Not approved/endorsed by Wizards. Portions of the materials used are property of Wizards of the Coast. ©Wizards of the Coast LLC.
Resources
- Fine-Tuned Model: MTG-Llama on HuggingFace
- Training Code: GitHub Repository
- Blog Post: boggs.tech