AI Paper Atlas

Research map / Architectures and efficiency / Quantization

LittleBit-2: Maximizing the Spectral Energy Gain in Sub-1-Bit LLMs via Latent Geometry Alignment

Banseok Lee, Youngmin Kim

ICML 2026 · Architectures and efficiency › Quantization · approach: KV cache quantization · 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 Joint-ITQ rotation to LittleBit initialization (ablation)LittleBit · WikiText-2, 1.0 bppperplexity: 16.30 -> 11.53Improved
add random internal rotation only (ablation)LittleBit · WikiText-2, 1.0 bppperplexity: 16.30 -> 12.63Improved
replace random rotation with Joint-ITQ rotation (ablation)LittleBit + random rotation · WikiText-2, 1.0 bppperplexity: 12.63 -> 11.53Improved
add random rotation only (ablation)LittleBit · WikiText-2, 0.1 bppperplexity: 26.11 -> 24.16Improved
replace random rotation with Joint-ITQ (ablation)LittleBit + random rotation · WikiText-2, 0.1 bppperplexity: 24.16 -> 23.74Improved
low-rank binary instead of tiny-rank FP16 (ablation)Llama-3 8B at 0.1 bpp · WikiText-2perplexity: 59.44 -> 26.11Improved
apply Joint-ITQ geometric alignmentLittleBit · Llama-3 8B zero-shot avg, 0.1 bppaccuracy: 41.80 -> 41.64No clear effect
apply Joint-ITQ geometric alignmentLittleBit · Llama-2 13B zero-shot avg, 1.0 bppaccuracy: 51.49 -> 60.53Improved
apply Joint-ITQ geometric alignmentLittleBit vs OneBit · Llama-2 7B, 1.0 bppperplexity: 9.08 -> 8.27Improved
apply Joint-ITQLittleBit · Gemma-3 27B, 0.1 bppperplexity: 17.72 -> 16.38Improved
apply internal rotation (ablation)LittleBit · Llama-2 7B synthetic spectrum break-even gammabreak-even gamma: 0.36 -> 0.41Improved
apply Joint-ITQ on top of rotation (ablation)LittleBit · synthetic spectrum break-even gammabreak-even gamma: 0.41 -> 0.51Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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