New paper finds TurboQuant performs worse than RaBitQ, citing reproducibility issues

A new technical note revisits the RaBitQ and TurboQuant quantization methods, comparing them under a unified framework. The analysis found that TurboQuant performed worse than RaBitQ in most tested settings for inner-product estimation, nearest-neighbor search, and KV cache quantization. Furthermore, the note documents reproducibility issues with the runtime and recall results reported in the original TurboQuant paper, indicating that some reported outcomes could not be replicated from the released implementation. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Highlights potential reproducibility issues in quantization research, urging careful validation of experimental results.

RANK_REASON This is a research paper published on arXiv that analyzes and compares existing methods, including reporting reproducibility issues.

Read on arXiv cs.AI →

• RaBitQ
• TurboQuant
• Jianyang Gao
• arXiv

• paper
• other