With known covariance, waterfilling improves GPTQ and WaterSIC reaches within 0.25 bit/entry of the rate-distortion limit while being basis-independent.
Available: https://transformer-circuits.pub/2023/ privileged-basis/index.html
2 Pith papers cite this work. Polarity classification is still indexing.
2
Pith papers citing it
years
2026 2verdicts
UNVERDICTED 2representative citing papers
High-rate quantization theory yields accurate approximations for the distortion of absmax INT and FP schemes in generic weight-plus-activation matrix multiplication.
citing papers explorer
No citing papers match the current filters.