AI Paper Atlas

Research map / Architectures and efficiency / Quantization

QTALE: Quantization-Robust Token-Adaptive Layer Execution for LLMs

KangHyun Noh, Jinheon Choi, Yulhwa Kim

ICML 2026 · Architectures and efficiency › Quantization · approach: Post-training quantization of LLMs · 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 entropy regularization on router outputs and adjusted execution threshold (QTALE) over D-LLMD-LLM under 3-bit AWQ · CSQA avg, LLaMA2-7Baccuracy: 70.57 -> 72.79Improved
QTALE over D-LLM under 4-bit AWQD-LLM · CSQA avg, LLaMA3.2-3Baccuracy: 73.96 -> 78.41Improved
QTALE over D-LLM under 4-bit AWQD-LLM · Alpaca PPL, LLaMA3.1-8Bperplexity: 5.47 -> 4.11Improved
QTALE over D-LLM under 3-bit AWQD-LLM · CSQA avg, LLaMA3.2-3Baccuracy: 59.70 -> 62.82Improved
add L_entropy (theta 0.5) (ablation)D-LLM · CSQA avg 4-bit, LLaMA2-7Baccuracy: 77.32 -> 78.86Improved
add L_entropy (theta 0.5) (ablation)D-LLM · CSQA avg 4-bit, LLaMA3.1-8Baccuracy: 77.68 -> 78.86Improved
add execution ratio adjustment theta (without L_entropy) (ablation)D-LLM · CSQA avg 4-bit, LLaMA2-7Baccuracy: 77.32 -> 78.48Improved
add theta adjustment on top of L_entropy (ablation)QTALE · CSQA avg 4-bit, LLaMA3.1-8Baccuracy: 78.86 -> 79.17Improved
add theta adjustment on top of L_entropy (ablation)QTALE · Alpaca PPL 4-bit, LLaMA3.1-8Bperplexity: 5.01 -> 4.11Improved
quantize token-adaptive model (4-bit) instead of 16-bitQTALE · Latency CSQAspeedup: 1.28x -> 1.26xNo clear effect
swap AWQ for MagR+GPTQQTALE · CSQA avg 4-bit, LLaMA3.2-3Baccuracy: 73.31 -> 77.62Improved
add 50% Wanda unstructured sparsity to QTALE-trained modelQTALE · CSQA avg, LLaMA3.2-3Baccuracy: 63.56 -> 68.34Improved

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