amkdg/

toxic-chat-Embeddings

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

10,165 conversations → 10,165 vectors emb.npy — float16 [10165, 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", mmap_mode="r")… See the full description on the dataset page: https://huggingface.co/datasets/amkdg/toxic-chat-Embeddings.