Jellyfish-7B

Model Details

Jellyfish-7B is a large language model equipped with 7 billion parameters.
We fine-tuned the mistralai/Mistral-7B-Instruct-v0.2 model using the datasets pertinent to data preprocessing tasks. The training data include two parts:

• Jellyfish-13B training data
• GPT4 generated reasoning data for data preprocessing tasks.

Jellyfish-7B vs GPT-3.5-turbo wining rate by GPT4 evaluation is 56.36%.

More details about the model can be found in the Jellyfish paper.

• Developed by: Haochen Zhang, Yuyang Dong, Chuan Xiao, Masafumi Oyamada
• Contact: dongyuyang@nec.com
• Funded by: NEC Corporation, Osaka University
• Language(s) (NLP): English
• License: Non-Commercial Creative Commons license (CC BY-NC-4.0)
• Finetuned from model: mistralai/Mistral-7B-Instruct-v0.2

Citation

If you find our work useful, please give us credit by citing:

@article{zhang2023jellyfish,
  title={Jellyfish: A Large Language Model for Data Preprocessing},
  author={Zhang, Haochen and Dong, Yuyang and Xiao, Chuan and Oyamada, Masafumi},
  journal={arXiv preprint arXiv:2312.01678},
  year={2023}
}

Performance on seen tasks

Task	Type	Dataset	Non-LLM SoTA1	GPT-3.52	GPT-42	Jellyfish-13B	Jellyfish-7B
Entity Matching	Seen	Fodors-Zagats	100	100	100	100	100
Entity Matching	Seen	Beer	94.37	96.30	100	96.77	96.55
Entity Matching	Seen	iTunes-Amazon	97.06	96.43	100	98.11	96.30
Entity Matching	Seen	DBLP-ACM	98.99	96.99	97.44	98.98	98.88
Entity Matching	Seen	DBLP-GoogleScholar	95.60	76.12	91.87	98.51	95.15
Entity Matching	Seen	Amazon-Google	75.58	66.53	74.21	81.34	80.83
Entity Matching	Unseen	Walmart-Amazon	86.76	86.17	90.27	89.42	85.64
Entity Matching	Unseen	Abt-Buy	89.33	--	92.77	89.58	82.38
Data Imputation	Seen	Restaurant	77.20	94.19	97.67	94.19	88.37
Data Imputation	Seen	Buy	96.50	98.46	100	100	96.62
Data Imputation	Unseen	Filpkart	68.00	--	89.94	81.68	79.44
Data Imputation	Unseen	Phone	86.70	--	90.79	87.21	85.00
Error Detection	Seen	Hosptial	94.40	90.74	90.74	95.59	96.27
Error Detection	Seen	Adult	99.10	92.01	92.01	99.33	91.96
Error Detection	Unseen	Flights	81.00	--	83.48	82.52	66.92
Error Detection	Unseen	Rayyan	79.00	--	81.95	90.65	69.82
Schema Matching	Seen	Sythea	38.50	57.14	66.67	36.36	44.44
Schema Matching	Seen	MIMIC	20.00	--	40.00	40.00	40.00
Schema Matching	Unseen	CMS	50.00	--	19.35	59.29	13.79

For GPT-3.5 and GPT-4, we used the few-shot approach on all datasets. However, for Jellyfish-13B and Jellyfish-Interpreter, the few-shot approach is disabled on seen datasets and enabled on unseen datasets.
Accuracy as the metric for data imputation and the F1 score for other tasks.

Performance on unseen tasks

Column Type Annotation

Dataset	RoBERTa (159 shots)1	GPT-3.51	GPT-4	Jellfish-13B	Jellyfish-7B
SOTAB	79.20	89.47	91.55	82.00	80.89

Few-shot is disabled for Jellyfish-13B.

1. Results from Column Type Annotation using ChatGPT

Attribute Value Extraction

Dataset	Stable Beluga 2 70B1	SOLAR 70B1	GPT-3.51	GPT-4 1	Jellfish-13B	Jellyfish-7B
AE-110k	52.10	49.20	61.30	55.50	58.12	76.85
OA-Mine	50.80	55.20	62.70	68.90	55.96	76.04

Prompt Template

[INST]:

<prompt> (without the <>)

[\INST]]