Use wiki18_100w as the database with the E5-base-v2 embedder.
The retrieval corpus is large and must be downloaded manually. Download wiki18_100w from ModelScope:
https://www.modelscope.cn/datasets/hhjinjiajie/FlashRAG_Dataset/tree/master/retrieval_corpus
we have prepared the download script, after it, put the corpus under tests/data/retrieval_corpus, including wiki18_100w.jsonl and e5_flat_inner.index
the corpus has prebuilt FAISS index with… See the full description on the dataset page:
https://huggingface.co/datasets/Ecoka/Echo-datasets.