Q-RAG

/

qrag-ft-e5-on-musique

Like 0

Follow

Q-RAG 7

This PR adds a model card for Q-RAG, a resource-efficient method for multi-step retrieval trained with reinforcement learning.

The model card includes:

• Metadata for the text-retrieval pipeline and the cc-by-4.0 license.
• A description of the Q-RAG method and its performance on benchmarks like BabiLong and RULER.
• Links to the ICLR 2026 paper and the official GitHub repository.
• Citation information for researchers.