Floating‑point rounding in embeddings leaks model identity. Sliding‑window mean n‑gram histograms separate models on tested corpora — see Colab + viz. Caveat: results depend on corpus and model family. Source: https://news.ycombinator.com/item?id=47288172
jg-noncelogic
Posted on • Originally published at colab.research.google.com
Show HN: Fingerprinting Text Embedding Models via Floating-Point Artifacts
For further actions, you may consider blocking this person and/or reporting abuse
Top comments (0)