Add TurboQuant vector quantization algorithm - #354

Draft

sroussey wants to merge 26 commits into

mainfrom

claude/integrate-arxiv-paper-VF55c

Commits on Aug 16, 2026

feat(util,ai): add TurboQuant vector quantization algorithm

sroussey
authored and
claude
committed
e766421
Potential fix for pull request finding 'Unused variable, import, function or class'

sroussey
authored and
claude
committed
c0d79a5
fix(util,ai): address PR review comments on TurboQuantize and VectorQuantizeTask

sroussey
authored and
claude
committed
016fa06
feat(util,ai): add turboQuantizeToTypedArray for direct storage-compatible output

sroussey
authored and
claude
committed
abf62ee
chore: apply prettier formatting to TurboQuant files
claude
committed
ab8c9a9
fix(util,ai): restrict TurboQuant to signed targets; harden input validation (#713)

sroussey
and
claude
committed
de8a04b
fix(util,ai): correct TurboQuant quantization grid and harden its decode path
claude
committed
120eec3
fix(util): scale the TurboQuant norm, and repair three tests that could not fail (#757)

sroussey
and
claude
committed
f15bec0
fix(ai): expose TurboQuant padding on VectorQuantizeTask and correct its guidance (#762)

sroussey
and
claude
committed
c475f01
test(util): measure TurboQuant's cosine bias on correlated pairs (currently red)

sroussey
authored and
claude
committed
c027fb5
fix(util): recover the cosine from TurboQuant's 1-bit sign statistic

sroussey
authored and
claude
committed
89796c8
docs(util): state that TurboQuant's packed codec has no storage path (#802)

sroussey
and
claude
committed
f76c298
fix(util): correct TurboQuant's cosine shrinkage at 2-4 bits (#803)

sroussey
and
claude
committed
25902d5
docs(util,ai): correct TurboQuant's accuracy claims against an honest int8 baseline (#801)

sroussey
and
claude
committed
7c525f0
fix(util): reject a TurboQuant record whose norm overflows a double
claude
committed
24e1eb9
test(util): measure TurboQuant's ranking fidelity per bit width
claude
committed
8d45e8f
docs(util): correct TurboQuant's ranking claim to what the measurement shows
claude
committed
b98fe04
docs(util): warn that turboDequantize's output is not shrinkage-corrected
claude
committed
60f9e36
fix(ai): reject a heterogeneous vector batch under method "turbo"
claude
committed
7129524
fix(ai): reject a heterogeneous vector batch under every method
claude
committed
70636ee
Merge pull request #814 from workglow-dev/claude/optimistic-goldberg-d74u5n-turboquant-fixes
sroussey
authored
86780fb

Navigation Menu

Add TurboQuant vector quantization algorithm - #354

Add TurboQuant vector quantization algorithm#354
sroussey wants to merge 26 commits into
mainfrom
claude/integrate-arxiv-paper-VF55c

Commits on Aug 16, 2026

feat(util,ai): add TurboQuant vector quantization algorithm

Potential fix for pull request finding 'Unused variable, import, function or class'

fix(util,ai): address PR review comments on TurboQuantize and VectorQuantizeTask

feat(util,ai): add turboQuantizeToTypedArray for direct storage-compatible output

chore: apply prettier formatting to TurboQuant files

fix(util,ai): restrict TurboQuant to signed targets; harden input validation (#713)

fix(util,ai): correct TurboQuant quantization grid and harden its decode path

fix(util): scale the TurboQuant norm, and repair three tests that could not fail (#757)

fix(ai): expose TurboQuant padding on VectorQuantizeTask and correct its guidance (#762)

test(util): measure TurboQuant's cosine bias on correlated pairs (currently red)

fix(util): recover the cosine from TurboQuant's 1-bit sign statistic

docs(util): state that TurboQuant's packed codec has no storage path (#802)

fix(util): correct TurboQuant's cosine shrinkage at 2-4 bits (#803)

docs(util,ai): correct TurboQuant's accuracy claims against an honest int8 baseline (#801)

fix(util): reject a TurboQuant record whose norm overflows a double

test(util): measure TurboQuant's ranking fidelity per bit width

docs(util): correct TurboQuant's ranking claim to what the measurement shows

docs(util): warn that turboDequantize's output is not shrinkage-corrected

fix(ai): reject a heterogeneous vector batch under method "turbo"

fix(ai): reject a heterogeneous vector batch under every method

Merge pull request #814 from workglow-dev/claude/optimistic-goldberg-d74u5n-turboquant-fixes

Commits on Aug 17, 2026

fix(util): reject a TurboQuant prepared query that cannot span its candidate

test(ai): pin the padded turbo vector's IVectorStorage width contract

docs(util,ai): correct TurboQuant's accuracy advice, bias table and public disclaimer

fix(util): pin TurboQuant's solved loading factors and drop the boxed code array

Merge pull request #834 from workglow-dev/claude/turboquant-review-fixes-354