{"work":{"id":"fd04f498-ff85-4de3-bcc7-31ef072b2ceb","openalex_id":null,"doi":null,"arxiv_id":"2202.00512","raw_key":null,"title":"Progressive Distillation for Fast Sampling of Diffusion Models","authors":null,"authors_text":"Tim Salimans, Jonathan Ho","year":2022,"venue":"cs.LG","abstract":"Diffusion models have recently shown great promise for generative modeling, outperforming GANs on perceptual quality and autoregressive models at density estimation. A remaining downside is their slow sampling time: generating high quality samples takes many hundreds or thousands of model evaluations. Here we make two contributions to help eliminate this downside: First, we present new parameterizations of diffusion models that provide increased stability when using few sampling steps. Second, we present a method to distill a trained deterministic diffusion sampler, using many steps, into a new diffusion model that takes half as many sampling steps. We then keep progressively applying this distillation procedure to our model, halving the number of required sampling steps each time. On standard image generation benchmarks like CIFAR-10, ImageNet, and LSUN, we start out with state-of-the-art samplers taking as many as 8192 steps, and are able to distill down to models taking as few as 4 steps without losing much perceptual quality; achieving, for example, a FID of 3.0 on CIFAR-10 in 4 steps. Finally, we show that the full progressive distillation procedure does not take more time than it takes to train the original model, thus representing an efficient solution for generative modeling using diffusion at both train and test time.","external_url":"https://arxiv.org/abs/2202.00512","cited_by_count":null,"metadata_source":"pith","metadata_fetched_at":"2026-05-23T07:57:43.017431+00:00","pith_arxiv_id":"2202.00512","created_at":"2026-05-09T06:25:40.398629+00:00","updated_at":"2026-05-23T07:57:43.017431+00:00","title_quality_ok":true,"display_title":"Progressive Distillation for Fast Sampling of Diffusion Models","render_title":"Progressive Distillation for Fast Sampling of Diffusion Models"},"hub":{"state":{"work_id":"fd04f498-ff85-4de3-bcc7-31ef072b2ceb","tier":"super_hub","tier_reason":"100+ Pith inbound or 10,000+ external citations","pith_inbound_count":108,"external_cited_by_count":null,"distinct_field_count":16,"first_pith_cited_at":"2022-08-12T09:54:11+00:00","last_pith_cited_at":"2026-05-21T05:25:41+00:00","author_build_status":"needed","summary_status":"needed","contexts_status":"needed","graph_status":"needed","ask_index_status":"needed","reader_status":"not_needed","recognition_status":"not_needed","updated_at":"2026-05-25T02:45:20.818306+00:00","tier_text":"super_hub"},"tier":"super_hub","role_counts":[{"context_role":"background","n":18},{"context_role":"method","n":5},{"context_role":"baseline","n":2}],"polarity_counts":[{"context_polarity":"background","n":18},{"context_polarity":"use_method","n":5},{"context_polarity":"baseline","n":2}],"runs":{"ask_index":{"job_type":"ask_index","status":"succeeded","result":{"title":"Progressive Distillation for Fast Sampling of Diffusion Models","claims":[{"claim_text":"Diffusion models have recently shown great promise for generative modeling, outperforming GANs on perceptual quality and autoregressive models at density estimation. A remaining downside is their slow sampling time: generating high quality samples takes many hundreds or thousands of model evaluations. Here we make two contributions to help eliminate this downside: First, we present new parameterizations of diffusion models that provide increased stability when using few sampling steps. Second, we present a method to distill a trained deterministic diffusion sampler, using many steps, into a ne","claim_type":"abstract","evidence_strength":"source_metadata"},{"claim_text":"0 3.30 RoboTron-Mani [40] ICCV, 2025 3D input 1 94.7 80.3 65.1 51.4 39.0 3.31 DaDu-Corki-SW [9] ISCA, 2025 - 1 92.3 80.0 67.4 56.6 45.8 3.42 RoboUniView (default) [23] arXiv, 2024 - 195.4 82.768.5 56.4 46.1 3.49 DTP [7] RA-L, 2025 Plan 1 92.4 81.970.2 60.3 50.9 3.55 NFE = 4 methods LaDi-WM [10] arXiv, 2025 WM≈4 92.7 83.1 72.1 61.2 54.1 3.63 LightDP [38] ICCV, 2025 Distill 493.7 84.5 74.1 64.4 55.6 3.72 NFE≥10 methods RoboTron-Mani (DiT) [40] arXiv, 2025 3D input>1096.9 83.068.1 56.5 46.8 3.51 𝜋 ","claim_type":"baseline","confidence":0.95,"evidence_strength":"citation_context"},{"claim_text":"\"# steps\" refers to the number of neural network runs.Δ denotes values taken from prior publications.∗ indicates usage of extra training data.𝑔 denotes use of classifier-free guidance [28]. Note that𝐿 in (𝑁×𝐿 ) notation is inference loop count per sampling step. Model AR FID↓IS↑# params # steps # Gflops BigGAN-deepΔ [5] 7.0 171.4 160M 1 - StyleGAN-XLΔ𝑔 [60] 2.3 265.1 166M 1 - Improved DDPMΔ [52] 12.3 - 280M 250>150k ADM + UpsampleΔ𝑔 [13] 3.9 215.8 554M 250 371k LDM-4Δ𝑔∗ [56] 3.6 247.7 400M 250 5","claim_type":"baseline","confidence":0.95,"evidence_strength":"citation_context"},{"claim_text":"the IEEE/CVF conference on computer vision and pattern recognition, pages 10684-10695, 2022. 1, 3, 7 [53] Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch, Michael Rubinstein, and Kfir Aberman. Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 22500- 22510, 2023. 1 [54] Tim Salimans and Jonathan Ho. Progressive distillation for fast sampling of diffusion models.arX","claim_type":"background","confidence":0.85,"evidence_strength":"citation_context"},{"claim_text":"in as few as 50 steps. A recent method [50] introduces molecule-specific trajectory diagnosis, which accelerates sampling and achieves strong performance with as few as 12 steps. Distillation and Distribution Matching for Accelerated SamplingEfforts to accelerate diffusion models generally focus on two strategies. The first involves progressive distillation [29], consistency models [33, 32], shortcut models [23, 6, 21, 52], and flow map distillation [3, 37], where a student model is trained to r","claim_type":"background","confidence":0.85,"evidence_strength":"citation_context"},{"claim_text":"locity E[𝑥 1 −𝑥 0 |𝑥 𝑡, 𝑡, 𝑐] must average over all sub-modes within class 𝑐. By further conditioning on a sub-mode index 𝑘, each sub- distribution (𝑐, 𝑘) becomes approximately unimodal, and the con- ditional mean E[𝑥 1 −𝑥 0 |𝑥 𝑡 , 𝑡, 𝑐, 𝑘] accurately points to a specific mode with no averaging distortion. Specifically, we first leverage a pre-trained vision foundation model (i.e., DINOv3 [46]) to extract semantic features and partition the target data manifold into 𝐾 distinct sub-manifolds, eff","claim_type":"background","confidence":0.85,"evidence_strength":"citation_context"},{"claim_text":"Meanwhile, reparameterized policy gradient methods require backpropagation through time [16], making them computationally expensive and prone to unstable optimization. To improve training efficiency and stability, recent works have explored one-step distillation tech- niques that compress iterative generative processes into single-step generators [17, 18]. Building on this idea, recent state-of-the-art methods such as flow Q-learning [19] and its variants [20, 21] distill iterative flow models i","claim_type":"background","confidence":0.85,"evidence_strength":"citation_context"}],"why_cited":"Pith tracks Progressive Distillation for Fast Sampling of Diffusion Models because it crossed a citation-hub threshold. Current citing contexts most often use it as background evidence (18 contexts).","role_counts":[{"n":18,"context_role":"background"},{"n":5,"context_role":"method"},{"n":2,"context_role":"baseline"}]},"error":null,"updated_at":"2026-05-21T22:03:10.740270+00:00"},"author_expand":{"job_type":"author_expand","status":"succeeded","result":{"authors_linked":[{"id":"94c5dbc0-da43-4fd4-abc6-96347f32f39c","orcid":null,"display_name":"Tim Salimans"},{"id":"c91ab32b-1b57-42d9-ae75-38fd5dc6cd26","orcid":null,"display_name":"Jonathan Ho"}]},"error":null,"updated_at":"2026-05-21T22:03:11.098758+00:00"},"context_extract":{"job_type":"context_extract","status":"succeeded","result":{"enqueued_papers":25},"error":null,"updated_at":"2026-05-14T14:41:44.775600+00:00"},"graph_features":{"job_type":"graph_features","status":"succeeded","result":{"co_cited":[{"title":"Score-Based Generative Modeling through Stochastic Differential Equations","work_id":"d9110e53-a5d4-4794-a4c5-a575e91c31ad","shared_citers":25},{"title":"Classifier-Free Diffusion Guidance","work_id":"acf2c588-c088-4a6c-938e-150ad7c666d7","shared_citers":19},{"title":"Flow Matching for Generative Modeling","work_id":"6edb71c4-5d64-40af-a394-9757ea051a36","shared_citers":16},{"title":"Denoising Diffusion Implicit Models","work_id":"8fa2128b-d18c-405c-ac92-0e669cf89ac0","shared_citers":15},{"title":"Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow","work_id":"a1989e1b-d66d-4533-be3a-fb9c5fd62290","shared_citers":11},{"title":"Auto-Encoding Variational Bayes","work_id":"97d95295-30e1-42b4-bbf6-85f0fa4edb44","shared_citers":10},{"title":"SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis","work_id":"8034c587-fba6-4941-87ba-c98f2ac962cb","shared_citers":10},{"title":"Decoupled Weight Decay Regularization","work_id":"07ef7360-d385-4033-83f7-8384a6325204","shared_citers":8},{"title":"Mean Flows for One-step Generative Modeling","work_id":"07a52ad5-0f82-4095-9a66-559b09fea1ae","shared_citers":8},{"title":"Building Normalizing Flows with Stochastic Interpolants","work_id":"83654cbd-832b-4269-bec9-26faa87ececd","shared_citers":7},{"title":"Consistency Models","work_id":"502bf494-8fcd-434f-828f-0566ab606719","shared_citers":6},{"title":"Human Preference Score v2: A Solid Benchmark for Evaluating Human Preferences of Text-to-Image Synthesis","work_id":"40702548-f094-4c67-a5db-a62f426f852e","shared_citers":6},{"title":"Latent Consistency Models: Synthesizing High-Resolution Images with Few-Step Inference","work_id":"53b1d836-7feb-402c-97c7-87b9bc51196f","shared_citers":6},{"title":"Adam: A Method for Stochastic Optimization","work_id":"1910796d-9b52-4683-bf5c-de9632c1028b","shared_citers":5},{"title":"Back to Basics: Let Denoising Generative Models Denoise","work_id":"37973de8-a5e6-4d92-897b-a98fa9f7f2f3","shared_citers":5},{"title":"Flow-GRPO: Training Flow Matching Models via Online RL","work_id":"bf1e8e81-ff31-401a-a5dc-d9c49df168ab","shared_citers":5},{"title":"High-Resolution Image Synthesis with Latent Diffusion Models","work_id":"f0270d36-2952-47fb-84c1-95e3ec341126","shared_citers":5},{"title":"Imagen Video: High Definition Video Generation with Diffusion Models","work_id":"bb20d241-dc6f-4b0a-b071-fd43a2cbd57f","shared_citers":5},{"title":"Improved techniques for training consistency models","work_id":"f730a570-4b0e-4910-87bb-ed1ec909fbc6","shared_citers":5},{"title":"Make-A-Video: Text-to-Video Generation without Text-Video Data","work_id":"52a801fc-a707-45a1-a8cd-0d6702f124ab","shared_citers":5},{"title":"Simplifying, Stabilizing and Scaling Continuous-Time Consistency Models","work_id":"1151c0b6-2be7-436c-ab83-843d92dea769","shared_citers":5},{"title":"Training Diffusion Models with Reinforcement Learning","work_id":"67684dda-3930-452a-b91a-36cbb8e2e219","shared_citers":5},{"title":"AnimateDiff: Animate Your Personalized Text-to-Image Diffusion Models without Specific Tuning","work_id":"1f9d1d3b-a6d6-45a9-9f13-51393c03be8a","shared_citers":4},{"title":"DanceGRPO: Unleashing GRPO on Visual Generation","work_id":"7404dd36-8f9c-478f-b089-ef9f8189c711","shared_citers":4}],"time_series":[{"n":3,"year":2023},{"n":3,"year":2024},{"n":2,"year":2025},{"n":39,"year":2026}],"dependency_candidates":[]},"error":null,"updated_at":"2026-05-14T14:51:47.491857+00:00"},"identity_refresh":{"job_type":"identity_refresh","status":"succeeded","result":{"items":[{"title":"Qwen3 Technical Report","outcome":"unchanged","work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e","resolver":"local_arxiv","confidence":0.98,"old_work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e"}],"counts":{"fixed":0,"merged":0,"unchanged":1,"quarantined":0,"needs_external_resolution":0},"errors":[],"attempted":1},"error":null,"updated_at":"2026-05-14T14:41:42.907075+00:00"},"role_polarity":{"job_type":"role_polarity","status":"succeeded","result":{"title":"Progressive Distillation for Fast Sampling of Diffusion Models","claims":[{"claim_text":"Diffusion models have recently shown great promise for generative modeling, outperforming GANs on perceptual quality and autoregressive models at density estimation. A remaining downside is their slow sampling time: generating high quality samples takes many hundreds or thousands of model evaluations. Here we make two contributions to help eliminate this downside: First, we present new parameterizations of diffusion models that provide increased stability when using few sampling steps. Second, we present a method to distill a trained deterministic diffusion sampler, using many steps, into a ne","claim_type":"abstract","evidence_strength":"source_metadata"},{"claim_text":"0 3.30 RoboTron-Mani [40] ICCV, 2025 3D input 1 94.7 80.3 65.1 51.4 39.0 3.31 DaDu-Corki-SW [9] ISCA, 2025 - 1 92.3 80.0 67.4 56.6 45.8 3.42 RoboUniView (default) [23] arXiv, 2024 - 195.4 82.768.5 56.4 46.1 3.49 DTP [7] RA-L, 2025 Plan 1 92.4 81.970.2 60.3 50.9 3.55 NFE = 4 methods LaDi-WM [10] arXiv, 2025 WM≈4 92.7 83.1 72.1 61.2 54.1 3.63 LightDP [38] ICCV, 2025 Distill 493.7 84.5 74.1 64.4 55.6 3.72 NFE≥10 methods RoboTron-Mani (DiT) [40] arXiv, 2025 3D input>1096.9 83.068.1 56.5 46.8 3.51 𝜋 ","claim_type":"baseline","confidence":0.95,"evidence_strength":"citation_context"},{"claim_text":"\"# steps\" refers to the number of neural network runs.Δ denotes values taken from prior publications.∗ indicates usage of extra training data.𝑔 denotes use of classifier-free guidance [28]. Note that𝐿 in (𝑁×𝐿 ) notation is inference loop count per sampling step. Model AR FID↓IS↑# params # steps # Gflops BigGAN-deepΔ [5] 7.0 171.4 160M 1 - StyleGAN-XLΔ𝑔 [60] 2.3 265.1 166M 1 - Improved DDPMΔ [52] 12.3 - 280M 250>150k ADM + UpsampleΔ𝑔 [13] 3.9 215.8 554M 250 371k LDM-4Δ𝑔∗ [56] 3.6 247.7 400M 250 5","claim_type":"baseline","confidence":0.95,"evidence_strength":"citation_context"},{"claim_text":"the IEEE/CVF conference on computer vision and pattern recognition, pages 10684-10695, 2022. 1, 3, 7 [53] Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch, Michael Rubinstein, and Kfir Aberman. Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 22500- 22510, 2023. 1 [54] Tim Salimans and Jonathan Ho. Progressive distillation for fast sampling of diffusion models.arX","claim_type":"background","confidence":0.85,"evidence_strength":"citation_context"},{"claim_text":"in as few as 50 steps. A recent method [50] introduces molecule-specific trajectory diagnosis, which accelerates sampling and achieves strong performance with as few as 12 steps. Distillation and Distribution Matching for Accelerated SamplingEfforts to accelerate diffusion models generally focus on two strategies. The first involves progressive distillation [29], consistency models [33, 32], shortcut models [23, 6, 21, 52], and flow map distillation [3, 37], where a student model is trained to r","claim_type":"background","confidence":0.85,"evidence_strength":"citation_context"},{"claim_text":"locity E[𝑥 1 −𝑥 0 |𝑥 𝑡, 𝑡, 𝑐] must average over all sub-modes within class 𝑐. By further conditioning on a sub-mode index 𝑘, each sub- distribution (𝑐, 𝑘) becomes approximately unimodal, and the con- ditional mean E[𝑥 1 −𝑥 0 |𝑥 𝑡 , 𝑡, 𝑐, 𝑘] accurately points to a specific mode with no averaging distortion. Specifically, we first leverage a pre-trained vision foundation model (i.e., DINOv3 [46]) to extract semantic features and partition the target data manifold into 𝐾 distinct sub-manifolds, eff","claim_type":"background","confidence":0.85,"evidence_strength":"citation_context"},{"claim_text":"Meanwhile, reparameterized policy gradient methods require backpropagation through time [16], making them computationally expensive and prone to unstable optimization. To improve training efficiency and stability, recent works have explored one-step distillation tech- niques that compress iterative generative processes into single-step generators [17, 18]. Building on this idea, recent state-of-the-art methods such as flow Q-learning [19] and its variants [20, 21] distill iterative flow models i","claim_type":"background","confidence":0.85,"evidence_strength":"citation_context"}],"why_cited":"Pith tracks Progressive Distillation for Fast Sampling of Diffusion Models because it crossed a citation-hub threshold. Current citing contexts most often use it as background evidence (18 contexts).","role_counts":[{"n":18,"context_role":"background"},{"n":5,"context_role":"method"},{"n":2,"context_role":"baseline"}]},"error":null,"updated_at":"2026-05-21T22:03:10.736924+00:00"},"summary_claims":{"job_type":"summary_claims","status":"succeeded","result":{"title":"Progressive Distillation for Fast Sampling of Diffusion Models","claims":[{"claim_text":"Diffusion models have recently shown great promise for generative modeling, outperforming GANs on perceptual quality and autoregressive models at density estimation. A remaining downside is their slow sampling time: generating high quality samples takes many hundreds or thousands of model evaluations. Here we make two contributions to help eliminate this downside: First, we present new parameterizations of diffusion models that provide increased stability when using few sampling steps. Second, we present a method to distill a trained deterministic diffusion sampler, using many steps, into a ne","claim_type":"abstract","evidence_strength":"source_metadata"}],"why_cited":"Pith tracks Progressive Distillation for Fast Sampling of Diffusion Models because it crossed a citation-hub threshold.","role_counts":[]},"error":null,"updated_at":"2026-05-14T14:51:38.958368+00:00"}},"summary":{"title":"Progressive Distillation for Fast Sampling of Diffusion Models","claims":[{"claim_text":"Diffusion models have recently shown great promise for generative modeling, outperforming GANs on perceptual quality and autoregressive models at density estimation. A remaining downside is their slow sampling time: generating high quality samples takes many hundreds or thousands of model evaluations. Here we make two contributions to help eliminate this downside: First, we present new parameterizations of diffusion models that provide increased stability when using few sampling steps. Second, we present a method to distill a trained deterministic diffusion sampler, using many steps, into a ne","claim_type":"abstract","evidence_strength":"source_metadata"}],"why_cited":"Pith tracks Progressive Distillation for Fast Sampling of Diffusion Models because it crossed a citation-hub threshold.","role_counts":[]},"graph":{"co_cited":[{"title":"Score-Based Generative Modeling through Stochastic Differential Equations","work_id":"d9110e53-a5d4-4794-a4c5-a575e91c31ad","shared_citers":25},{"title":"Classifier-Free Diffusion Guidance","work_id":"acf2c588-c088-4a6c-938e-150ad7c666d7","shared_citers":19},{"title":"Flow Matching for Generative Modeling","work_id":"6edb71c4-5d64-40af-a394-9757ea051a36","shared_citers":16},{"title":"Denoising Diffusion Implicit Models","work_id":"8fa2128b-d18c-405c-ac92-0e669cf89ac0","shared_citers":15},{"title":"Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow","work_id":"a1989e1b-d66d-4533-be3a-fb9c5fd62290","shared_citers":11},{"title":"Auto-Encoding Variational Bayes","work_id":"97d95295-30e1-42b4-bbf6-85f0fa4edb44","shared_citers":10},{"title":"SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis","work_id":"8034c587-fba6-4941-87ba-c98f2ac962cb","shared_citers":10},{"title":"Decoupled Weight Decay Regularization","work_id":"07ef7360-d385-4033-83f7-8384a6325204","shared_citers":8},{"title":"Mean Flows for One-step Generative Modeling","work_id":"07a52ad5-0f82-4095-9a66-559b09fea1ae","shared_citers":8},{"title":"Building Normalizing Flows with Stochastic Interpolants","work_id":"83654cbd-832b-4269-bec9-26faa87ececd","shared_citers":7},{"title":"Consistency Models","work_id":"502bf494-8fcd-434f-828f-0566ab606719","shared_citers":6},{"title":"Human Preference Score v2: A Solid Benchmark for Evaluating Human Preferences of Text-to-Image Synthesis","work_id":"40702548-f094-4c67-a5db-a62f426f852e","shared_citers":6},{"title":"Latent Consistency Models: Synthesizing High-Resolution Images with Few-Step Inference","work_id":"53b1d836-7feb-402c-97c7-87b9bc51196f","shared_citers":6},{"title":"Adam: A Method for Stochastic Optimization","work_id":"1910796d-9b52-4683-bf5c-de9632c1028b","shared_citers":5},{"title":"Back to Basics: Let Denoising Generative Models Denoise","work_id":"37973de8-a5e6-4d92-897b-a98fa9f7f2f3","shared_citers":5},{"title":"Flow-GRPO: Training Flow Matching Models via Online RL","work_id":"bf1e8e81-ff31-401a-a5dc-d9c49df168ab","shared_citers":5},{"title":"High-Resolution Image Synthesis with Latent Diffusion Models","work_id":"f0270d36-2952-47fb-84c1-95e3ec341126","shared_citers":5},{"title":"Imagen Video: High Definition Video Generation with Diffusion Models","work_id":"bb20d241-dc6f-4b0a-b071-fd43a2cbd57f","shared_citers":5},{"title":"Improved techniques for training consistency models","work_id":"f730a570-4b0e-4910-87bb-ed1ec909fbc6","shared_citers":5},{"title":"Make-A-Video: Text-to-Video Generation without Text-Video Data","work_id":"52a801fc-a707-45a1-a8cd-0d6702f124ab","shared_citers":5},{"title":"Simplifying, Stabilizing and Scaling Continuous-Time Consistency Models","work_id":"1151c0b6-2be7-436c-ab83-843d92dea769","shared_citers":5},{"title":"Training Diffusion Models with Reinforcement Learning","work_id":"67684dda-3930-452a-b91a-36cbb8e2e219","shared_citers":5},{"title":"AnimateDiff: Animate Your Personalized Text-to-Image Diffusion Models without Specific Tuning","work_id":"1f9d1d3b-a6d6-45a9-9f13-51393c03be8a","shared_citers":4},{"title":"DanceGRPO: Unleashing GRPO on Visual Generation","work_id":"7404dd36-8f9c-478f-b089-ef9f8189c711","shared_citers":4}],"time_series":[{"n":3,"year":2023},{"n":3,"year":2024},{"n":2,"year":2025},{"n":39,"year":2026}],"dependency_candidates":[]},"authors":[{"id":"c91ab32b-1b57-42d9-ae75-38fd5dc6cd26","orcid":null,"display_name":"Jonathan Ho","source":"manual","import_confidence":0.72},{"id":"94c5dbc0-da43-4fd4-abc6-96347f32f39c","orcid":null,"display_name":"Tim Salimans","source":"manual","import_confidence":0.72}]}}