Datasetpublic

s1ky/CROP-dataset

Introduction Crop-dataset is a large-scale open-source instruction fine-tuning dataset for LLMs in crop science, which includes over 210K high-quality question-answer pairs in Chinese and English. Basic Information Currently, Crop-dataset primarily includes two types of grains: rice and corn. The dataset contains a sufficient amount of single-turn and multi-turn question-answer pairs. Composition of the Single-round Dialogue Dataset… See the full description on the dataset page: https://huggingface.co/datasets/s1ky/CROP-dataset.

sourceHugging Facecc-by-nc-4.0updated 6mo agoView on Hugging Face

0likes4downloads

Dataset Card

Introduction

Crop-dataset is a large-scale open-source instruction fine-tuning dataset for LLMs in crop science, which includes over 210K high-quality question-answer pairs in Chinese and English. <div style="text-align: center;"> <img src="./Figures/datasetpredistribution.png" alt="Dataset Framework" width="100%"/> </div>

Basic Information

Currently, Crop-dataset primarily includes two types of grains: rice and corn. The dataset contains a sufficient amount of single-turn and multi-turn question-answer pairs.

Composition of the Single-round Dialogue Dataset

Cereal	Type	Task	English QA	Chinese QA	Total
Rice	Closed QA	In-domain knowledge QA	42951	83396	126347
	Open QA	In-domain knowledge QA	2430	2037	4467
		Event extraction	1891	1030	2921
		Named Entity Recognition	2003	1604	3607
		Summary	1586	1628	3214
Corn	Closed QA	In-domain knowledge QA	25259	27667	52926
	Open QA	In-domain knowledge QA	3202	3047	6249
		Event extraction	2245	1322	3567
		Named Entity Recognition	2008	1316	3324
		Summary	1559	1857	3416
Others*	---	---	---	---	<1000
Overall	---	---	85134	124904	210038

*The 'Others' category includes data from other grains like wheat, which is less than 1000 QA pairs in total.

Please note that despite our data-cleaning efforts, the final QA pairs inevitably contain a small amount of data (<0.5%) from other grains like wheat. Because we think this data does not dominantly influence the fine-tuning results, it is included in the rice and maize QA pairs, and we have listed it separately to avoid any misleading counts.

Composition of the Multi-round Dialogue Dataset

Cereal	Scenario	Task	English QA	Chinese QA	Total
Rice	Problem Solving	Pest Control	14 (3-turn) + 71 (4-turn)	8 (3-turn) + 37 (4-turn)	130
		Nutrient Supplementation	19 (3-turn) + 93 (4-turn)	2 (3-turn) + 90 (4-turn) + 1 (5-turn)	205
		Disease Containment	19 (3-turn) + 60 (4-turn)	4 (3-turn) + 39 (4-turn)	122
	Personalized Recommendation	Crop Variety Selection	12 (3-turn) + 53 (4-turn)	9 (3-turn) + 9 (4-turn)	83
		Resource Management	4 (3-turn) + 110 (4-turn) + 1 (5-turn)	5 (3-turn) + 50 (4-turn)	170
	Knowledge Interpretation	Research Interpretation	3 (3-turn) + 125 (4-turn) + 1 (5-turn)	8 (3-turn) + 85 (4-turn)	222
Corn	Problem Solving	Pest Control	20 (3-turn) + 84 (4-turn)	7 (3-turn) + 77 (4-turn)	188
		Nutrient Supplementation	24 (3-turn) + 56 (4-turn)	8 (3-turn) + 30 (4-turn)	118
		Disease Containment	21 (3-turn) + 64 (4-turn)	2 (3-turn) + 19 (4-turn) + 1 (5-turn)	107
	Personalized Recommendation	Crop Variety Selection	19 (3-turn) + 75 (4-turn)	46 (3-turn) + 47 (4-turn)	187
		Resource Management	8 (3-turn) + 94 (4-turn)	1 (3-turn) + 69 (4-turn)	172
	Knowledge Interpretation	Research Interpretation	5 (3-turn) + 94 (4-turn) + 1 (5-turn)	6 (3-turn) + 61 (4-turn)	167
Overall	---	---	1150	721	1871

How to Use

We have released our dataset in Alpaca format. All question-answer pairs are organized by task. The codes and prompts related to this dataset are released at https://github.com/RenqiChen/The_Crop.

BibTeX & Citation

If you find our codes and datasets useful, please consider citing our work:

bibtex

@inproceedings{zhangempowering,
  title={Empowering and Assessing the Utility of Large Language Models in Crop Science},
  author={Zhang, Hang and Sun, Jiawei and Chen, Renqi and Liu, Wei and Yuan, Zhonghang and Zheng, Xinzhe and Wang, Zhefan and Yang, Zhiyuan and Yan, Hang and Zhong, Han-Sen and others},
  booktitle={The Thirty-eight Conference on Neural Information Processing Systems Datasets and Benchmarks Track}
}