PaddlePaddle/RocketQA

Based on the README, here's a technical summary: Provides dual-encoder and cross-encoder architectures built on ERNIE pre-trained models for dense retrieval, with separate encoding pipelines for queries and passages enabling efficient vector similarity matching at scale. Integrates with vector search frameworks like FAISS and the JINA neural search platform to build end-to-end retrieval pipelines, while supporting fine-tuning on custom datasets through PaddlePaddle's training API.

785 stars and 22 monthly downloads. No commits in the last 6 months. Available on PyPI.