{"url":"/sota/quantization-on-imagenet","task":{"name":"Quantization","url":"/task/quantization","note":null},"dataset":{"name":"ImageNet","url":"/dataset/imagenet"},"category":"Computer Vision","categories":["Computer Vision","Methodology"],"category_note":null,"description":"**Quantization** is a promising technique to reduce the computation cost of neural network training, which can replace high-cost floating-point numbers (e.g., float32) with low-cost fixed-point numbers (e.g., int8/int16).\r\n\r\n\r\n<span class=\"description-source\">Source: [Adaptive Precision Training: Quantify Back Propagation in Neural Networks with Fixed-point Numbers ](https://arxiv.org/abs/1911.00361)</span>","description_from":"task","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","rank":"the archive's row order at snapshot; not re-ranked","rows_end_at":"2025-07-28","rows_withheld_as_spam":0,"metric_values":"the archive's strings, untouched"},"metrics":["Top-1 Accuracy (%)","Weight bits","Activation bits"],"metric_direction":{"note":"inferred from the metric name only (the archive records no direction); null = not inferred, chart draws points only","by_metric":{"Top-1 Accuracy (%)":"higher","Weight bits":null,"Activation bits":null}},"counts":{"rows":27,"rows_with_code":24,"rows_with_paper_page":27,"rows_dated":27,"rows_using_additional_data":0},"rows":[{"rank_in_archive_order":1,"model":"FQ-ViT (ViT-L)","metrics":{"Activation bits":"8","Top-1 Accuracy (%)":"85.03","Weight bits":"8"},"uses_additional_data":false,"paper_date":"2021-11-27","paper":"/paper/fq-vit-fully-quantized-vision-transformer","paper_url":"https://arxiv.org/abs/2111.13824v4","paper_title":"FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer","code":"https://github.com/megvii-research/FQ-ViT","n_code_links":1,"syntology":{"n_ran":3,"n_unverified":0,"n_samples":3,"n_pointer_only_licence":1}},{"rank_in_archive_order":2,"model":"FQ-ViT (ViT-B)","metrics":{"Activation bits":"8","Top-1 Accuracy (%)":"83.31","Weight bits":"8"},"uses_additional_data":false,"paper_date":"2021-11-27","paper":"/paper/fq-vit-fully-quantized-vision-transformer","paper_url":"https://arxiv.org/abs/2111.13824v4","paper_title":"FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer","code":"https://github.com/megvii-research/FQ-ViT","n_code_links":1,"syntology":{"n_ran":3,"n_unverified":0,"n_samples":3,"n_pointer_only_licence":1}},{"rank_in_archive_order":3,"model":"FQ-ViT (Swin-B)","metrics":{"Activation bits":"8","Top-1 Accuracy (%)":"82.97","Weight bits":"8"},"uses_additional_data":false,"paper_date":"2021-11-27","paper":"/paper/fq-vit-fully-quantized-vision-transformer","paper_url":"https://arxiv.org/abs/2111.13824v4","paper_title":"FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer","code":"https://github.com/megvii-research/FQ-ViT","n_code_links":1,"syntology":{"n_ran":3,"n_unverified":0,"n_samples":3,"n_pointer_only_licence":1}},{"rank_in_archive_order":4,"model":"FQ-ViT (Swin-S)","metrics":{"Activation bits":"8","Top-1 Accuracy (%)":"82.71","Weight bits":"8"},"uses_additional_data":false,"paper_date":"2021-11-27","paper":"/paper/fq-vit-fully-quantized-vision-transformer","paper_url":"https://arxiv.org/abs/2111.13824v4","paper_title":"FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer","code":"https://github.com/megvii-research/FQ-ViT","n_code_links":1,"syntology":{"n_ran":3,"n_unverified":0,"n_samples":3,"n_pointer_only_licence":1}},{"rank_in_archive_order":5,"model":"FQ-ViT (DeiT-B)","metrics":{"Activation bits":"8","Top-1 Accuracy (%)":"81.20","Weight bits":"8"},"uses_additional_data":false,"paper_date":"2021-11-27","paper":"/paper/fq-vit-fully-quantized-vision-transformer","paper_url":"https://arxiv.org/abs/2111.13824v4","paper_title":"FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer","code":"https://github.com/megvii-research/FQ-ViT","n_code_links":1,"syntology":{"n_ran":3,"n_unverified":0,"n_samples":3,"n_pointer_only_licence":1}},{"rank_in_archive_order":6,"model":"FQ-ViT (Swin-T)","metrics":{"Activation bits":"8","Top-1 Accuracy (%)":"80.51","Weight bits":"8"},"uses_additional_data":false,"paper_date":"2021-11-27","paper":"/paper/fq-vit-fully-quantized-vision-transformer","paper_url":"https://arxiv.org/abs/2111.13824v4","paper_title":"FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer","code":"https://github.com/megvii-research/FQ-ViT","n_code_links":1,"syntology":{"n_ran":3,"n_unverified":0,"n_samples":3,"n_pointer_only_licence":1}},{"rank_in_archive_order":7,"model":"FQ-ViT (DeiT-S)","metrics":{"Activation bits":"8","Top-1 Accuracy (%)":"79.17","Weight bits":"8"},"uses_additional_data":false,"paper_date":"2021-11-27","paper":"/paper/fq-vit-fully-quantized-vision-transformer","paper_url":"https://arxiv.org/abs/2111.13824v4","paper_title":"FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer","code":"https://github.com/megvii-research/FQ-ViT","n_code_links":1,"syntology":{"n_ran":3,"n_unverified":0,"n_samples":3,"n_pointer_only_licence":1}},{"rank_in_archive_order":8,"model":"Xception W8A8","metrics":{"Activation bits":"8","Top-1 Accuracy (%)":"78.972","Weight bits":"8"},"uses_additional_data":false,"paper_date":"2021-09-19","paper":"/paper/hptq-hardware-friendly-post-training","paper_url":"https://arxiv.org/abs/2109.09113v3","paper_title":"HPTQ: Hardware-Friendly Post Training Quantization","code":"https://github.com/sony/model_optimization","n_code_links":1,"syntology":null},{"rank_in_archive_order":9,"model":"ADLIK-MO-ResNet50-W4A4","metrics":{"Activation bits":"4","Top-1 Accuracy (%)":"77.878","Weight bits":"4"},"uses_additional_data":false,"paper_date":"2019-02-21","paper":"/paper/learned-step-size-quantization","paper_url":"https://arxiv.org/abs/1902.08153v3","paper_title":"Learned Step Size Quantization","code":"https://github.com/zhutmost/lsq-net","n_code_links":9,"syntology":{"n_ran":7,"n_unverified":16,"n_samples":23,"n_pointer_only_licence":6}},{"rank_in_archive_order":10,"model":"ADLIK-MO-ResNet50-W3A4","metrics":{"Activation bits":"4","Top-1 Accuracy (%)":"77.34","Weight bits":"3"},"uses_additional_data":false,"paper_date":"2019-02-21","paper":"/paper/learned-step-size-quantization","paper_url":"https://arxiv.org/abs/1902.08153v3","paper_title":"Learned Step Size Quantization","code":"https://github.com/zhutmost/lsq-net","n_code_links":9,"syntology":{"n_ran":7,"n_unverified":16,"n_samples":23,"n_pointer_only_licence":6}},{"rank_in_archive_order":11,"model":"EfficientNet-B0 ReLU W8A8","metrics":{"Activation bits":"8","Top-1 Accuracy (%)":"77.092","Weight bits":"8"},"uses_additional_data":false,"paper_date":"2021-09-19","paper":"/paper/hptq-hardware-friendly-post-training","paper_url":"https://arxiv.org/abs/2109.09113v3","paper_title":"HPTQ: Hardware-Friendly Post Training Quantization","code":"https://github.com/sony/model_optimization","n_code_links":1,"syntology":null},{"rank_in_archive_order":12,"model":"ResNet50-W4A4 (paper)","metrics":{"Activation bits":"4","Top-1 Accuracy (%)":"76.7","Weight bits":"4"},"uses_additional_data":false,"paper_date":"2019-02-21","paper":"/paper/learned-step-size-quantization","paper_url":"https://arxiv.org/abs/1902.08153v3","paper_title":"Learned Step Size Quantization","code":"https://github.com/zhutmost/lsq-net","n_code_links":9,"syntology":{"n_ran":7,"n_unverified":16,"n_samples":23,"n_pointer_only_licence":6}},{"rank_in_archive_order":13,"model":"EfficientNet-B0-W8A8","metrics":{"Activation bits":"8","Top-1 Accuracy (%)":"76.4","Weight bits":"8"},"uses_additional_data":false,"paper_date":"2020-07-20","paper":"/paper/hmq-hardware-friendly-mixed-precision","paper_url":"https://arxiv.org/abs/2007.09952v1","paper_title":"HMQ: Hardware Friendly Mixed Precision Quantization Block for CNNs","code":"https://github.com/sony-si/ai-research","n_code_links":2,"syntology":{"n_ran":0,"n_unverified":4,"n_samples":4,"n_pointer_only_licence":0}},{"rank_in_archive_order":14,"model":"EfficientNet-B0-W4A4","metrics":{"Activation bits":"4","Top-1 Accuracy (%)":"76","Weight bits":"4"},"uses_additional_data":false,"paper_date":"2020-07-20","paper":"/paper/hmq-hardware-friendly-mixed-precision","paper_url":"https://arxiv.org/abs/2007.09952v1","paper_title":"HMQ: Hardware Friendly Mixed Precision Quantization Block for CNNs","code":"https://github.com/sony-si/ai-research","n_code_links":2,"syntology":{"n_ran":0,"n_unverified":4,"n_samples":4,"n_pointer_only_licence":0}},{"rank_in_archive_order":15,"model":"ResNet50-W3A4","metrics":{"Activation bits":"4","Top-1 Accuracy (%)":"75.45","Weight bits":"3"},"uses_additional_data":false,"paper_date":"2020-07-20","paper":"/paper/hmq-hardware-friendly-mixed-precision","paper_url":"https://arxiv.org/abs/2007.09952v1","paper_title":"HMQ: Hardware Friendly Mixed Precision Quantization Block for CNNs","code":"https://github.com/sony-si/ai-research","n_code_links":2,"syntology":{"n_ran":0,"n_unverified":4,"n_samples":4,"n_pointer_only_licence":0}},{"rank_in_archive_order":16,"model":"EfficientNet-B0 W8A8","metrics":{"Activation bits":"8","Top-1 Accuracy (%)":"74.216","Weight bits":"8"},"uses_additional_data":false,"paper_date":"2021-09-19","paper":"/paper/hptq-hardware-friendly-post-training","paper_url":"https://arxiv.org/abs/2109.09113v3","paper_title":"HPTQ: Hardware-Friendly Post Training Quantization","code":"https://github.com/sony/model_optimization","n_code_links":1,"syntology":null},{"rank_in_archive_order":17,"model":"MPT (80) +BN","metrics":{"Top-1 Accuracy (%)":"74.03"},"uses_additional_data":false,"paper_date":"2021-03-17","paper":"/paper/multi-prize-lottery-ticket-hypothesis-finding-1","paper_url":"https://arxiv.org/abs/2103.09377v1","paper_title":"Multi-Prize Lottery Ticket Hypothesis: Finding Accurate Binary Neural Networks by Pruning A Randomly Weighted Network","code":"https://github.com/chrundle/biprop","n_code_links":1,"syntology":null},{"rank_in_archive_order":18,"model":"EfficientNet-W4A4","metrics":{"Activation bits":"4","Top-1 Accuracy (%)":"73.8","Weight bits":"4"},"uses_additional_data":false,"paper_date":"2020-04-20","paper":"/paper/lsq-improving-low-bit-quantization-through","paper_url":"https://arxiv.org/abs/2004.09576v1","paper_title":"LSQ+: Improving low-bit quantization through learnable offsets and better initialization","code":"https://github.com/ZouJiu1/LSQplus","n_code_links":4,"syntology":{"n_ran":3,"n_unverified":7,"n_samples":10,"n_pointer_only_licence":10}},{"rank_in_archive_order":19,"model":"DenseNet-121 W8A8","metrics":{"Activation bits":"8","Top-1 Accuracy (%)":"73.356","Weight bits":"8"},"uses_additional_data":false,"paper_date":"2021-09-19","paper":"/paper/hptq-hardware-friendly-post-training","paper_url":"https://arxiv.org/abs/2109.09113v3","paper_title":"HPTQ: Hardware-Friendly Post Training Quantization","code":"https://github.com/sony/model_optimization","n_code_links":1,"syntology":null},{"rank_in_archive_order":20,"model":"MixNet-W4A4","metrics":{"Activation bits":"4","Top-1 Accuracy (%)":"71.7","Weight bits":"4"},"uses_additional_data":false,"paper_date":"2020-04-20","paper":"/paper/lsq-improving-low-bit-quantization-through","paper_url":"https://arxiv.org/abs/2004.09576v1","paper_title":"LSQ+: Improving low-bit quantization through learnable offsets and better initialization","code":"https://github.com/ZouJiu1/LSQplus","n_code_links":4,"syntology":{"n_ran":3,"n_unverified":7,"n_samples":10,"n_pointer_only_licence":10}},{"rank_in_archive_order":21,"model":"FQ-ViT (DeiT-T)","metrics":{"Activation bits":"8","Top-1 Accuracy (%)":"71.61","Weight bits":"8"},"uses_additional_data":false,"paper_date":"2021-11-27","paper":"/paper/fq-vit-fully-quantized-vision-transformer","paper_url":"https://arxiv.org/abs/2111.13824v4","paper_title":"FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer","code":"https://github.com/megvii-research/FQ-ViT","n_code_links":1,"syntology":{"n_ran":3,"n_unverified":0,"n_samples":3,"n_pointer_only_licence":1}},{"rank_in_archive_order":22,"model":"UniQ (Ours)","metrics":{"Top-1 Accuracy (%)":"71.5"},"uses_additional_data":false,"paper_date":"2021-04-01","paper":"/paper/training-multi-bit-quantized-and-binarized","paper_url":"https://arxiv.org/abs/2104.00210v1","paper_title":"Training Multi-bit Quantized and Binarized Networks with A Learnable Symmetric Quantizer","code":"https://github.com/phuocphn/uniq","n_code_links":1,"syntology":null},{"rank_in_archive_order":23,"model":"MobileNetV2 W8A8","metrics":{"Activation bits":"8","Top-1 Accuracy (%)":"71.46","Weight bits":"8"},"uses_additional_data":false,"paper_date":"2021-09-19","paper":"/paper/hptq-hardware-friendly-post-training","paper_url":"https://arxiv.org/abs/2109.09113v3","paper_title":"HPTQ: Hardware-Friendly Post Training Quantization","code":"https://github.com/sony/model_optimization","n_code_links":1,"syntology":null},{"rank_in_archive_order":24,"model":"MobileNetV2","metrics":{"Top-1 Accuracy (%)":"70.9"},"uses_additional_data":false,"paper_date":"2020-07-20","paper":"/paper/hmq-hardware-friendly-mixed-precision","paper_url":"https://arxiv.org/abs/2007.09952v1","paper_title":"HMQ: Hardware Friendly Mixed Precision Quantization Block for CNNs","code":"https://github.com/sony-si/ai-research","n_code_links":2,"syntology":{"n_ran":0,"n_unverified":4,"n_samples":4,"n_pointer_only_licence":0}},{"rank_in_archive_order":25,"model":"MobileNet-v1 + EWGS + R2Loss","metrics":{"Top-1 Accuracy (%)":"69.79","Weight bits":"4"},"uses_additional_data":false,"paper_date":"2023-03-14","paper":"/paper/r-2-range-regularization-for-model","paper_url":"https://arxiv.org/abs/2303.08253v2","paper_title":"R2 Loss: Range Restriction Loss for Model Compression and Quantization","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":26,"model":"MobileNet-v1 + LSQ + R2Loss","metrics":{"Top-1 Accuracy (%)":"69.64"},"uses_additional_data":false,"paper_date":"2023-03-14","paper":"/paper/r-2-range-regularization-for-model","paper_url":"https://arxiv.org/abs/2303.08253v2","paper_title":"R2 Loss: Range Restriction Loss for Model Compression and Quantization","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":27,"model":"ResNet-18 + PACT + R2Loss","metrics":{"Activation bits":"4","Top-1 Accuracy (%)":"68.45","Weight bits":"2"},"uses_additional_data":false,"paper_date":"2023-03-14","paper":"/paper/r-2-range-regularization-for-model","paper_url":"https://arxiv.org/abs/2303.08253v2","paper_title":"R2 Loss: Range Restriction Loss for Model Compression and Quantization","code":null,"n_code_links":0,"syntology":null}],"since_archive":{"claim":"Results that newer papers report for their own method, placed here by Syntology. A model pointed at the cell in the paper's own table; the number was read from that cell and checked against this leaderboard's metric, dataset, split and scale; an independent check that saw this leaderboard's other rows and every other leaderboard on the same dataset accepted it. Not reviewed by the paper's authors or by the archive's editors, and not ranked against the archive rows.","extraction_file_present":true,"measurement":{"test_papers":883,"papers_with_output":881,"judged_true":108,"judged":110,"wilson95_lower":0.9361,"measured_on":"2026-09-24","frozen_commit":"0e3de0df94"},"measurement_note":"blind adjudication of accepted entries on a held-out split of archive papers, rules frozen before the test","coverage":{"sentence":"Syntology has checked 6,264 of the 9,581 papers on this site that are newer than the archive; results from the others appear after they are checked.","complete":false,"papers_newer_than_archive":9581,"papers_checked":6264,"papers_extracted_not_yet_verified":0,"boards_without_verdict":2,"papers_not_yet_extracted":3316},"order":"newest first by month (arXiv date, else the arXiv-id month), then arXiv id descending","columns":[],"entries":[]},"syntology":{"read_at":"2026-09-24T18:15:14+00:00","claim":"Per row: N of M harvested code samples from that row's paper executed on a synthesized fixture; the other M-N are unverified. Not a reproduction of the row's number; not a correctness claim. n_pointer_only_licence counts samples the site points at rather than redistributes (a licence axis, independent of ran/unverified).","rows_with_graph_line":17,"rows_with_any_sample_ran":13,"distinct_papers_with_graph_line":4,"distinct_papers_with_any_sample_ran":3,"samples_over_distinct_papers":{"n_ran":13,"n_unverified":27,"n_samples":40,"n_pointer_only_licence":17,"note":"each paper (arXiv id) counted once, however many rows it is behind; this is the page-level figure"},"samples_row_weighted":{"n_ran":51,"n_unverified":78,"n_samples":129,"n_pointer_only_licence":46,"note":"row-weighted: a paper behind several rows is counted once per row; inflated relative to samples_over_distinct_papers by design, kept for readers summing the per-row syntology blocks"}}}