Dataset · reasonir
Synthetic data for ReasonIR reasoning search
reasonir/reasonir-data
English dataset with complex queries and texts of varying lengths used to train the ReasonIR-8B retriever.
What's inside
A collection ranging from 100 thousand to 1 million English text examples created to train the ReasonIR-8B model. It contains synthetic data with texts of varying lengths and complex queries (note that original positive documents for complex queries must be downloaded and merged separately using an included script due to legal restrictions).
What you can do with it
- Training neural retrievers for complex search with reasoning
- Improving document retrieval in RAG pipelines
- Testing search models on queries of varying lengths
- Tasks
- Text retrieval
- Languages
- English
- Size
- 100K–1M rows
- Downloads this month
- 493
- Likes
- 36
Models trained on it
- lightonai/Reason-ModernColBERTSentence similarity 6.7K
- mradermacher/Diver-Retriever-4B-1020-GGUF 972
- AQ-MedAI/Diver-Retriever-4BText ranking 792
- mradermacher/Diver-Retriever-1.7B-i1-GGUF 586
- AQ-MedAI/Diver-Retriever-4B-1020Text ranking 454
- mradermacher/Diver-Retriever-1.7B-GGUF 283
- mradermacher/Diver-Retriever-0.6B-GGUF 263
- AQ-MedAI/Diver-Retriever-0.6BText ranking 197
How to load it
from datasets import load_dataset
ds = load_dataset("reasonir/reasonir-data")