zhuang-li/SCAR

[ACL 2025 main] SCAR: Data Selection via Style Consistency-Aware Response Ranking for Efficient Instruction-Tuning of Large Language Models

27
/ 100
Experimental

This tool helps machine learning engineers and researchers select the best training data for fine-tuning large language models. You provide a list of instructions and their corresponding answers, and SCAR scores how 'style consistent' and beneficial each pair is. The output is a ranked list of instruction-answer pairs, allowing you to efficiently choose the most impactful data to improve your LLM's performance.

No commits in the last 6 months.

Use this if you are a machine learning engineer or researcher looking to improve the performance of your large language models by selecting the highest quality instruction-tuning data from a larger dataset.

Not ideal if your training data includes non-English examples or has duplicate entries, as these are not currently supported and require manual cleaning beforehand.

large-language-models LLM-fine-tuning data-curation machine-learning-research natural-language-processing
No License Stale 6m No Package No Dependents
Maintenance 2 / 25
Adoption 7 / 25
Maturity 8 / 25
Community 10 / 25

How are scores calculated?

Stars

39

Forks

4

Language

Python

License

Last pushed

Aug 06, 2025

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/transformers/zhuang-li/SCAR"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.