Dataset · Open-Orca
Complete FLAN Collection
Open-Orca/FLAN
Complete FLAN task suite in parquet format—around 300 GB of English texts for instruction tuning.
What's inside
The repository contains the complete FLAN collection, around 300 GB in parquet, generated by the official seqio template engine from Google's FLAN Collection GitHub repository. Data is in English; licenses are inherited from source datasets. Size ranges from 100M to 1B records.
What you can do with it
- Train an instruction-tuned LLM
- Perform LoRA fine-tuning
- Evaluate generalization on FLAN tasks
- Languages
- English
- Size
- 100M–1B rows
- Downloads this month
- 17.5K
- Likes
- 206
Models trained on it
- dmis-lab/UltraBERT-150MFill mask 13.4K
- dmis-lab/UltraBERT-380MFill mask 9.9K
- Sayan01/Llama-Flan-XL2baseText generation 815
- mradermacher/GLM4.7-Distill-LFM2.5-1.2B-i1-GGUFText generation 476
- mradermacher/AlgoMind-1.2B-i1-GGUFText generation 451
- mradermacher/AlgoMind-1.2B-GGUFText generation 270
- mradermacher/GLM4.7-Distill-LFM2.5-1.2B-GGUFText generation 183
- Sayan01/T5-Flan-SmallText generation 109
How to load it
from datasets import load_dataset
ds = load_dataset("Open-Orca/FLAN")