krzysztof/landing_page
1
0
Fork 0
mirror of https://github.com/qdrant/landing_page.git synced 2026-10-09 21:08:31 +02:00
Code Issues Packages Projects Releases Wiki Activity

Add link to visual reprentation of TQ

Browse Source
This commit is contained in:
jojii
2026-05-13 11:03:19 +02:00
parent f9d760c05e
commit cb477c16c5
1 changed files with 2 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
qdrant-landing/content/articles/turboquant-quantization.md
+2
View File
@@ -74,6 +74,8 @@ TurboQuant ([Zandieh et al., 2026](https://arxiv.org/abs/2504.19874)) is a rotat
The elegance: **no per-dataset training, no calibration set, no codebooks to persist**. The codebook is derived once from the standard normal distribution and is universal. The same lookup table works for every dataset and every dimensionality. By contrast, PQ requires a learned codebook trained on representative data and shipped alongside the index. The elegance: **no per-dataset training, no calibration set, no codebooks to persist**. The codebook is derived once from the standard normal distribution and is universal. The same lookup table works for every dataset and every dimensionality. By contrast, PQ requires a learned codebook trained on representative data and shipped alongside the index.
For a visual explanation of TurboQuant, see [this interactive walkthrough](https://arkaung.github.io/interactive-turboquant/).
### MSE vs PROD: Picking the Variant ### MSE vs PROD: Picking the Variant
The original paper proposes two variants. **MSE** is the literal recipe above: scalar Lloyd-Max quantization, score by codebook lookup. **PROD** adds a second QJL random projection on top of the indices to cancel the per-vector length bias that MSE inherits from rounding to a finite codebook. The original paper proposes two variants. **MSE** is the literal recipe above: scalar Lloyd-Max quantization, score by codebook lookup. **PROD** adds a second QJL random projection on top of the indices to cancel the per-vector length bias that MSE inherits from rounding to a finite codebook.