Embeddings of CohereForAI/aya_dataset, produced with amkdg/Qwen3-Embedding-8B-NVFP4 — 4096-d,
L2-normalized float16 (cosine = dot product).
190,385 conversations → 191,650 vectors
emb.npy — float16 [191650, 4096]
meta.parquet — one row per vector, aligned with emb.npy: id, uuid, tag, chunk, n_chunks, count, source_ref
manifest.json — counts and provenance
import numpy as np, pyarrow.parquet as pq
emb = np.load("emb.npy"… See the full description on the dataset page:
https://huggingface.co/datasets/amkdg/aya-dataset-Embeddings.
