Research map / Architectures and efficiency / Quantization
TWLA: Achieving Ternary Weights and Low-Bit Activations for LLMs via Post-Training Quantization
Zhixiong Zhao, Zukang Xu, Zhixuan Chen, Xing Hu, Zhe jiang, Dawei Yang
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 E2M-ATQ (ablation) | TWLA (LLaMA2-13B, A16) · C4 | C4 perplexity: 6e3 -> 18.32 | Improved |
| add KOTMS to E2M-ATQ (ablation) | TWLA (LLaMA2-13B, A16) · C4 | C4 perplexity: 18.32 -> 8.64 | Improved |
| add E2M-ATQ to KOTMS (ablation) | TWLA (LLaMA2-13B, A16) · C4 | C4 perplexity: 57.98 -> 8.64 | Improved |
| add KOTMS to E2M-ATQ (ablation) | TWLA (LLaMA2-13B, A16) · MMLU | MMLU accuracy: 30.15 -> 44.86 | Improved |
| add KOTMS to E2M-ATQ (ablation) | TWLA (Qwen3-14B, A16) · MMLU | MMLU accuracy: 57.64 -> 69.15 | Improved |
| add KOTMS to E2M-ATQ (ablation) | TWLA (Qwen3-14B, A16) · C4 | C4 perplexity: 23.64 -> 17.26 | Improved |
| add KOTMS (no ILA-AMP) to E2M-ATQ at 4-bit activations (ablation) | TWLA (LLaMA2-13B, A4) · C4 | C4 perplexity: 3e3 -> 25.03 | Improved |
| add ILA-AMP to E2M-ATQ + KOTMS (ablation) | TWLA (LLaMA2-13B, A4) · C4 | C4 perplexity: 25.03 -> 10.07 | Improved |
| add ILA-AMP to E2M-ATQ + KOTMS (ablation) | TWLA (LLaMA2-13B, A4) · MMLU | MMLU accuracy: 27.52 -> 38.17 | Improved |
| add ILA-AMP to E2M-ATQ + KOTMS (ablation) | TWLA (Qwen3-14B, A4) · MMLU | MMLU accuracy: 47.60 -> 60.82 | Improved |
| add KOTMS to E2M-ATQ + ILA-AMP (ablation) | TWLA (LLaMA2-13B, A4) · C4 | C4 perplexity: 55.86 -> 10.07 | Improved |
| use TWLA W1.58 instead of PT2-LLM W1.58 | LLaMA3-8B A16 · 7-task avg accuracy | accuracy: 39.04 -> 62.98 | Improved |
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.
- PT2-LLM — PT$^2$-LLM: Post-Training Ternarization for Large Language Models (ICLR 2026) · better on 2 of 2 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
- Exploring Post-training Quantization in LLMs from Comprehensive Study to Low Rank Compensation (AAAI 2024 · similarity 0.93)
- Norm Tweaking: High-Performance Low-Bit Quantization of Large Language Models (AAAI 2024 · similarity 0.93)
- ASER: Activation Smoothing and Error Reconstruction for Large Language Model Quantization (AAAI 2025 · similarity 0.92)
- QLLM: Accurate and Efficient Low-Bitwidth Quantization for Large Language Models (ICLR 2024 · similarity 0.91)
- ABQ-LLM: Arbitrary-Bit Quantized Inference Acceleration for Large Language Models (AAAI 2025 · similarity 0.91)
- CBQ: Cross-Block Quantization for Large Language Models (ICLR 2025 · similarity 0.91)
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.
