Research map / Architectures and efficiency / Quantization
QTALE: Quantization-Robust Token-Adaptive Layer Execution for LLMs
KangHyun Noh, Jinheon Choi, Yulhwa Kim
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add entropy regularization on router outputs and adjusted execution threshold (QTALE) over D-LLM | D-LLM under 3-bit AWQ · CSQA avg, LLaMA2-7B | accuracy: 70.57 -> 72.79 | Improved |
| QTALE over D-LLM under 4-bit AWQ | D-LLM · CSQA avg, LLaMA3.2-3B | accuracy: 73.96 -> 78.41 | Improved |
| QTALE over D-LLM under 4-bit AWQ | D-LLM · Alpaca PPL, LLaMA3.1-8B | perplexity: 5.47 -> 4.11 | Improved |
| QTALE over D-LLM under 3-bit AWQ | D-LLM · CSQA avg, LLaMA3.2-3B | accuracy: 59.70 -> 62.82 | Improved |
| add L_entropy (theta 0.5) (ablation) | D-LLM · CSQA avg 4-bit, LLaMA2-7B | accuracy: 77.32 -> 78.86 | Improved |
| add L_entropy (theta 0.5) (ablation) | D-LLM · CSQA avg 4-bit, LLaMA3.1-8B | accuracy: 77.68 -> 78.86 | Improved |
| add execution ratio adjustment theta (without L_entropy) (ablation) | D-LLM · CSQA avg 4-bit, LLaMA2-7B | accuracy: 77.32 -> 78.48 | Improved |
| add theta adjustment on top of L_entropy (ablation) | QTALE · CSQA avg 4-bit, LLaMA3.1-8B | accuracy: 78.86 -> 79.17 | Improved |
| add theta adjustment on top of L_entropy (ablation) | QTALE · Alpaca PPL 4-bit, LLaMA3.1-8B | perplexity: 5.01 -> 4.11 | Improved |
| quantize token-adaptive model (4-bit) instead of 16-bit | QTALE · Latency CSQA | speedup: 1.28x -> 1.26x | No clear effect |
| swap AWQ for MagR+GPTQ | QTALE · CSQA avg 4-bit, LLaMA3.2-3B | accuracy: 73.31 -> 77.62 | Improved |
| add 50% Wanda unstructured sparsity to QTALE-trained model | QTALE · CSQA avg, LLaMA3.2-3B | accuracy: 63.56 -> 68.34 | Improved |
Closest papers by meaning
- Quantization Error Propagation: Revisiting Layer-Wise Post-Training Quantization (NeurIPS 2025 · similarity 0.90)
- DP-LLM: Runtime Model Adaptation with Dynamic Layer-wise Precision Assignment (NeurIPS 2025 · similarity 0.90)
- Any-Precision LLM: Low-Cost Deployment of Multiple, Different-Sized LLMs (ICML 2024 · similarity 0.89)
- KVTuner: Sensitivity-Aware Layer-Wise Mixed-Precision KV Cache Quantization for Efficient and Nearly Lossless LLM Inference (ICML 2025 · similarity 0.89)
- What Makes Quantization for Large Language Model Hard? An Empirical Study from the Lens of Perturbation (AAAI 2024 · similarity 0.89)
- Agile-Quant: Activation-Guided Quantization for Faster Inference of LLMs on the Edge (AAAI 2024 · similarity 0.89)
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.
