Research map / Architectures and efficiency / Quantization
Post-training weight quantization
Compresses weights by quantization with sharpness control, loss-aware objectives, and quantization noise.
100 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Quantization, keyword cluster “post quantization · ptq · compression”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Quiet None of the tracked benchmarks has a report from 2025–26.
16 papers in 2023–24 and 34 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 13.6% to 10.0%.
No longer reported by papers here: ImageNet (top-1 accuracy, last in 2023).
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2016 · Affinity Preserving Quantization for Hashing: A Vector Quantization Approach to Learning Compact Binary Codes (AAAI 2016 · 11 citations)
- 2017 · Soft-to-Hard Vector Quantization for End-to-End Learning Compressible Representations (NeurIPS 2017 · 358 citations)
- 2018 · Model compression via distillation and quantization (ICLR 2018 · 252 citations)
- 2019 · Post training 4-bit quantization of convolutional networks for rapid-deployment (NeurIPS 2019 · 261 citations)
- 2020 · Accelerating Large-Scale Inference with Anisotropic Vector Quantization (ICML 2020 · 109 citations)
- 2021 · Training with Quantization Noise for Extreme Model Compression (ICLR 2021 · 114 citations)
- 2022 · Nonuniform-to-Uniform Quantization: Towards Accurate Quantization via Generalized Straight-Through Estimation (CVPR 2022 · 116 citations)
- 2023 · LVQAC: Lattice Vector Quantization Coupled With Spatially Adaptive Companding for Efficient Learned Image Compression (CVPR 2023 · 25 citations)
- 2024 · MagR: Weight Magnitude Reduction for Enhancing Post-Training Quantization (NeurIPS 2024 · 3 citations)
- 2025 · Automatic Joint Structured Pruning and Quantization for Efficient Neural Network Training and Compression (CVPR 2025 · 21 citations)
- 2026 · Robust Training of Neural Networks at Arbitrary Precision and Sparsity (ICLR 2026 · 1 citation)
Earliest papers here
- Affinity Preserving Quantization for Hashing: A Vector Quantization Approach to Learning Compact Binary Codes (AAAI 2016 · 11 citations)
- Soft-to-Hard Vector Quantization for End-to-End Learning Compressible Representations (NeurIPS 2017 · 358 citations)
- Towards the Limit of Network Quantization (ICLR 2017 · 60 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
ImageNet · top-1 accuracy (higher is better)
6 papers, 2019–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 74.99 | Focused Quantization for Sparse CNNs (NeurIPS 2019 · 11 citations) |
| 2020 | 76.12 | And the Bit Goes Down: Revisiting the Quantization of Neural Networks (ICLR 2020 · 65 citations) |
| 2021 | 77.15 | Permute, Quantize, and Fine-Tune: Efficient Compression of Neural Networks (CVPR 2021 · 25 citations) |
| 2023 | 78.1 | Probabilistic Weight Fixing: Large-scale training of neural network weight uncertainties for quantisation. (NeurIPS 2023 · 1 citation) |
What has been tried
77 papers here report 857 measured changes (a component added, removed or swapped, compared with a baseline): helped 72%, no clear effect 6%, hurt 18%, mixed 3%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Finetune QwT modules and classification head for one epoch | ImageNet-1k val, 4/4 · Top-1 accuracy | 61.4 -> 64.8 | Helped | Quantization without Tears (CVPR 2025 · 6 citations) |
| add RS | CIFAR-10, 10%, 20%, and 30% keep ratios · accuracy | 84.1 -> 85.1; 87.6 -> 88.8; 91.0 -> 93.0 | Helped | Adaptive Dataset Quantization (AAAI 2025 · 5 citations) |
| Include the full set of proposed components rather than an ablated configuration | ImageNet, 4w4a · top-1 accuracy | 69.14 -> 71.20 | Helped | Enhancing Diversity for Data-free Quantization (CVPR 2025 · 4 citations) |
| add MagR preprocessing to QuIP | WikiText2 · perplexity | 27.13 -> 13.31 | Helped | MagR: Weight Magnitude Reduction for Enhancing Post-Training Quantization (NeurIPS 2024 · 3 citations) |
| Use post-training quantization rather than quantizing during S→Q sparse fine-tuning | WikiText2 validation · perplexity | 33.51 -> 30.46; 36.20 -> 32.51 | Helped | Effective Interplay between Sparsity and Quantization: From Theory to Practice (ICLR 2025 · 2 citations) |
| add data-free holistic distillation to input-agnostic binarization | WikiText2, Phi-2 2.7B · perplexity | 12.32 -> 10.48 | Helped | QBB: Quantization with Binary Bases for LLMs (NeurIPS 2024 · 2 citations) |
| Add the orthogonal constraint | Bmshj2018 and SwinT-ChARM, all four context models · BD-rate savings over scalar quantizer | -18.2% -> -22.60%; -7.94% -> -11.94%; -6.34% -> -10.61%; -4.20% -> -8.31%; -8.09% -> -12.44%; -2.25% -> -6.37%; -1.22% - | Helped | Learning Optimal Lattice Vector Quantizers for End-to-end Neural Image Compression (NeurIPS 2024 · 2 citations) |
| Quantize only the embedding layer rather than the position-wise feed-forward or attention layer | WikiText-103 test · perplexity | Hurt | Layer-wise Quantization for Quantized Optimistic Dual Averaging (ICML 2025 · 0 citations) | |
| replace LSQ QAT with post-training OBC quantization | PACS · OOD accuracy | 87.8 ± 0.3 -> 83.7 ± 0.4 | Hurt | QT-DoG: Quantization-Aware Training for Domain Generalization (ICML 2025 · 0 citations) |
| Increase optimization steps beyond 50,000 | ImageNet, data-free pseudo 2-bit · top-1 accuracy | +1.99% | Helped | Binary Quadratic Quantization: Beyond First-Order Quantization for Real-Valued Matrix Compression (NeurIPS 2025 · 0 citations) |
| Replace one 4096-entry codebook with 16 1024-entry codebooks | MaskGit-VQGAN, 256 tokens · MeanAcc (%) | 93.51 -> 92.23 | Hurt | Switchable Token-Specific Codebook Quantization For Face Image Compression (NeurIPS 2025 · 0 citations) |
| Increase calibration batch size from 1 to 8 | NeRV-3M, Beauty/Jockey/ReadyS · PSNR (dB) | 29.02 -> 29.09 | Helped | On Quantizing Neural Representation for Variable-Rate Video Coding (ICLR 2025 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 17 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- CIFAR-10 · 4 of 17 papers
- CIFAR-100 · 3 of 17 papers
- ImageNet · 3 of 17 papers
- WikiText2 · 3 of 17 papers
- ImageNet-1K · 2 of 17 papers
Baselines
- OBC · 2 of 17 papers
- PSAQ-ViT · 2 of 17 papers
- RepQ-ViT · 2 of 17 papers
- PTQ4ViT · 2 of 17 papers
- GPTQ · 2 of 17 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 50 papers; from OpenAlex and the papers' affiliations).
- Zhejiang University · 2 papers
- Samsung · 2 papers
- McMaster University · 2 papers
- East China Normal University · 2 papers
- Sorbonne University · 2 papers
- Tsinghua University · 2 papers
- EPFL · 2 papers
Use it
Download BibTeX of 40 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Post-training quantization of LLMs (136 papers)
- Post-training quantization of ViTs (105 papers)
- KV cache quantization (72 papers)
Elsewhere in the topic
- Mixed-precision quantization (92 papers)
- Low-precision training formats (58 papers)
- Ultra-low-bit LLM quantization (56 papers)
- Binary neural networks (27 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).