AI Paper Atlas

Research map / Architectures and efficiency / Quantization

Pushing the Limits of Block Rotations in Post-Training Quantization

Sai Sanjeet, Ian Colbert, Pablo Monteagudo-Lago, Giuseppe Franco, Yaman Umuroglu, Nicholas J. Fraser

ICML 2026 · Architectures and efficiency › Quantization · approach: Post-training quantization of ViTs · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add MassDiff permutation before block rotation (ablation)Llama3 1B block 16 Qronos INT4 · WikiText2WikiText2 perplexity: 35.8 -> 18.2Improved
add MassDiff permutation before block rotation (ablation)Llama3 8B block 32 Qronos INT4 · WikiText2WikiText2 perplexity: 12.8 -> 8.5Improved
add MassDiff permutation before block rotation (ablation)Llama3 1B block 32 RTN INT4 · WikiText2WikiText2 perplexity: 4e3 -> 38.0Improved
add MassDiff permutation before block rotation (ablation)Llama3 8B block 32 RTN INT4 · WikiText2WikiText2 perplexity: 6e3 -> 12.0Improved
add MassDiff permutation to MR-Qronos on GSM8K (ablation)Llama3 8B INT4 · GSM8Kaccuracy: 1.7 -> 75.4Improved
use MassDiff instead of random permutation (ablation)PeRQ* Llama3 1B · WikiText2WikiText2 perplexity: 27.8 -> 16.9Improved
use MassDiff instead of absmax permutation (ablation)PeRQ* Llama3 1B · WikiText2WikiText2 perplexity: 20.1 -> 16.9Improved
use MassDiff instead of ZigZag permutation (ablation)PeRQ* Llama3 1B · WikiText2WikiText2 perplexity: 17.1 -> 16.9Improved
use MassDiff instead of ZigZag with 128K calibration tokens (ablation)PeRQ* Llama3 1B block 32 · WikiText2WikiText2 perplexity: 17.3 -> 16.9Improved
add MassDiff with C4 calibration (ablation)PeRQ* Llama3 1B · WikiText2WikiText2 perplexity: 29.2 -> 18.9Improved
replace GPTQ with Qronos rounding (ablation)MassDiff + QuaRot Llama3 1B · WikiText2WikiText2 perplexity: 28.4 -> 16.9Improved
replace RTN with GPTQ rounding when learning rotations (ablation)MassDiff + SpinQuant Llama3 1B · WikiText2WikiText2 perplexity: 15.9 -> 19.5Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy