Dense vectors for every passage of English Wikipedia, ready for retrieval-
augmented generation. Publishing these saves ~63 GPU-hours of embedding.
Pairs with the passage text at
wikipedia-en-2026-07-01-passages
and prebuilt FAISS indexes at
wikipedia-en-2026-07-01-faiss.
17,473,199 passage vectors, 1024-dim, float16, L2-normalized.
34 GB across 57 .npy shards (vecs_gpu*.npy), one… See the full description on the dataset page:
https://huggingface.co/datasets/Sherlock-Comms/wikipedia-en-2026-07-01-qwen3-embed-4b.
