Skip to main content
Datasetinstruction-tuningv1.0

LIMA

by Meta AI · open-source · Last verified 2026-03-17

LIMA (Less Is More for Alignment) is a carefully curated dataset of 1,000 high-quality instruction-response pairs demonstrating that alignment quality matters more than quantity. Sourced from StackExchange, wikiHow, and manually written prompts, LIMA-tuned models rival GPT-4 on many benchmarks.

https://huggingface.co/datasets/GAIR/lima
B
BAbove Average
Adoption: B+Quality: A+Freshness: BCitations: AEngagement: F

Specifications

License
CC-BY-NC-4.0
Pricing
open-source
Capabilities
instruction-tuning, alignment, supervised-fine-tuning
Integrations
huggingface-datasets
Use Cases
fine-tuning, alignment-research, data-efficiency
API Available
No
Tags
quality-over-quantity, instruction-following, meta, curated, 1000-samples
Added
2026-03-17
Completeness
100%

Index Score

69.1
Adoption
74
Quality
90
Freshness
69
Citations
86
Engagement
0

Put AI to work for your business

Deploy this dataset alongside autonomous AaaS agents that handle tasks end-to-end — no babysitting required.

Explore the full AI ecosystem on Agents as a Service