amkdg/

bing-conversations-Embeddings

Embeddings of david-ar/bing-chat-logs, produced with amkdg/Qwen3-Embedding-8B-NVFP4 — 4096-d, L2-normalized float16 (cosine = dot product).

644 conversations → 650 vectors emb.npy — float16 [650, 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/bing-conversations-Embeddings.