Research map / Architectures and efficiency / Quantization
DLLMQuant: A Post-Training Quantization Framework Tailored for Diffusion-Based Large Language Models
Chen Xu, Zhixuan Chen, 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 TMAS, IA-AQ and CGQ (ablation) | QuaRot · LLaDA-8B W4A4 average | accuracy: 51.03 -> 54.29 | Improved |
| add TMAS, IA-AQ and CGQ (ablation) | AWQ · LLaDA-8B W4A4 average | accuracy: 48.09 -> 49.26 | Improved |
| add the three DLLMQuant techniques | QuaRot · GSM8K, LLaDA-8B W4A4 | accuracy: 44.57 -> 56.25 | Improved |
| add the three DLLMQuant techniques | QuaRot · HumanEval, LLaDA-1.5-8B W4A4 | accuracy: 37.33 -> 44.58 | Improved |
| add the three DLLMQuant techniques | QuaRot · Dream-7B W4A4 average | accuracy: 58.85 -> 61.90 | Improved |
| add the three DLLMQuant techniques | AWQ · LLaDA-1.5-8B W4A4 average | accuracy: 58.49 -> 59.94 | Improved |
| add the three DLLMQuant techniques | QuaRot · LLaDA-8B C-EVAL W4A4 | accuracy: 56.32 -> 61.22 | Improved |
| add IA-AQ (leave out quantizing matmul after softmax naively) (ablation) | QuaRot · LLaDA-8B PIQA, cumulative MSE over steps | MSE: MSE nearly 0.1 without IA-AQ | Improved |
| add CGQ weighting by both mask state and score (ablation) | QuaRot · GSM8K, 4-bit | accuracy: 44.57 -> 45.58 | Improved |
| add CGQ weighting by both mask state and score (ablation) | QuaRot · HumanEval, 4-bit | accuracy: 25.33 -> 26.85 | Improved |
| use mask-only weighting in CGQ vs score & mask (ablation) | QuaRot · GSM8K, 4-bit | accuracy: 45.07 -> 45.58 | Improved |
| quantize LLaDA to 4 bit weights with DLLMQuant | LLaDA-8B · inference vs FP16 | speedup: 1.71x speedup | 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.
- QuaRot — QuaRot: Outlier-Free 4-Bit Inference in Rotated LLMs (NeurIPS 2024) · better on 5 of 5 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
- Evaluating Quantized Large Language Models (ICML 2024 · similarity 0.91)
- Exploring Post-training Quantization in LLMs from Comprehensive Study to Low Rank Compensation (AAAI 2024 · similarity 0.91)
- QLLM: Accurate and Efficient Low-Bitwidth Quantization for Large Language Models (ICLR 2024 · similarity 0.90)
- Norm Tweaking: High-Performance Low-Bit Quantization of Large Language Models (AAAI 2024 · similarity 0.90)
- CBQ: Cross-Block Quantization for Large Language Models (ICLR 2025 · similarity 0.90)
- Quantization Error Propagation: Revisiting Layer-Wise Post-Training Quantization (NeurIPS 2025 · similarity 0.90)
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.
