Librarian Bot. I found the following papers similar to this paper.

The following papers were recommended by the Semantic Scholar API

AnimateLCM: Accelerating the Animation of Personalized Diffusion Models and Adapters with Decoupled Consistency Learning (2024)
Enhanced Distribution Alignment for Post-Training Quantization of Diffusion Models (2024)
Scalable High-Resolution Pixel-Space Image Synthesis with Hourglass Diffusion Transformers (2024)
Make a Cheap Scaling: A Self-Cascade Diffusion Model for Higher-Resolution Adaptation (2024)
Progressive Knowledge Distillation Of Stable Diffusion XL Using Layer Level Loss (2024)

Please give a thumbs up to this comment if you found it helpful!

If you want recommendations for any Paper on Hugging Face checkout this Space

You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: \n\n@librarian-bot\n\t recommend

\n","updatedAt":"2024-02-24T01:20:59.918Z","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":264}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.7313646674156189},"editors":["librarian-bot"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2402.14167","authors":[{"_id":"65d807403381a5be2777e499","user":{"_id":"629d55b53a3221bb2117422c","avatarUrl":"/avatars/e39f25cd7190d282478cdac20a7700ee.svg","isPro":false,"fullname":"Zizheng Pan","user":"zizhpan","type":"user"},"name":"Zizheng Pan","status":"admin_assigned","statusLastChangedAt":"2024-02-23T12:59:11.640Z","hidden":false},{"_id":"65d807403381a5be2777e49a","user":{"_id":"65a8a828ef14f9e6037dfad2","avatarUrl":"/avatars/55fdc0393b2e37ac998ddbdb1e1ff636.svg","isPro":false,"fullname":"Bohan Zhuang","user":"BohanZ","type":"user"},"name":"Bohan Zhuang","status":"extracted_confirmed","statusLastChangedAt":"2024-02-23T09:07:36.541Z","hidden":false},{"_id":"65d807403381a5be2777e49b","user":{"_id":"641d1c5ec3983aa94915c162","avatarUrl":"/avatars/127985b837ecf61e43c835deee578b5e.svg","isPro":false,"fullname":"De-An Huang","user":"deahuang","type":"user"},"name":"De-An Huang","status":"admin_assigned","statusLastChangedAt":"2024-02-23T12:59:19.059Z","hidden":false},{"_id":"65d807403381a5be2777e49c","user":{"_id":"64c1a69e226e016da8450ae2","avatarUrl":"/avatars/54c161e8b8543244ed13cbe47017624e.svg","isPro":false,"fullname":"Weili Nie","user":"xiaoli08","type":"user"},"name":"Weili Nie","status":"admin_assigned","statusLastChangedAt":"2024-02-23T12:59:27.287Z","hidden":false},{"_id":"65d807403381a5be2777e49d","user":{"_id":"66c8037c737ba92ae3fe0322","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/66c8037c737ba92ae3fe0322/WR_Yh5DWOVVh7IFlF24NM.jpeg","isPro":false,"fullname":"Zhiding Yu","user":"Zhiding","type":"user"},"name":"Zhiding Yu","status":"claimed_verified","statusLastChangedAt":"2024-08-29T07:25:21.964Z","hidden":false},{"_id":"65d807403381a5be2777e49e","name":"Chaowei Xiao","hidden":false},{"_id":"65d807403381a5be2777e49f","name":"Jianfei Cai","hidden":false},{"_id":"65d807403381a5be2777e4a0","user":{"_id":"6532920b3e385cfc6002938d","avatarUrl":"/avatars/cb9cc6d2733031582c83f56dc6cd1dd5.svg","isPro":false,"fullname":"Anima Anandkumar","user":"animakumar","type":"user"},"name":"Anima Anandkumar","status":"admin_assigned","statusLastChangedAt":"2024-02-23T12:59:58.141Z","hidden":false}],"publishedAt":"2024-02-21T23:08:54.000Z","submittedOnDailyAt":"2024-02-23T00:17:31.384Z","title":"T-Stitch: Accelerating Sampling in Pre-Trained Diffusion Models with\n Trajectory Stitching","submittedOnDailyBy":{"_id":"60f1abe7544c2adfd699860c","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674929746905-60f1abe7544c2adfd699860c.jpeg","isPro":false,"fullname":"AK","user":"akhaliq","type":"user"},"summary":"Sampling from diffusion probabilistic models (DPMs) is often expensive for\nhigh-quality image generation and typically requires many steps with a large\nmodel. In this paper, we introduce sampling Trajectory Stitching T-Stitch, a\nsimple yet efficient technique to improve the sampling efficiency with little\nor no generation degradation. Instead of solely using a large DPM for the\nentire sampling trajectory, T-Stitch first leverages a smaller DPM in the\ninitial steps as a cheap drop-in replacement of the larger DPM and switches to\nthe larger DPM at a later stage. Our key insight is that different diffusion\nmodels learn similar encodings under the same training data distribution and\nsmaller models are capable of generating good global structures in the early\nsteps. Extensive experiments demonstrate that T-Stitch is training-free,\ngenerally applicable for different architectures, and complements most existing\nfast sampling techniques with flexible speed and quality trade-offs. On DiT-XL,\nfor example, 40% of the early timesteps can be safely replaced with a 10x\nfaster DiT-S without performance drop on class-conditional ImageNet generation.\nWe further show that our method can also be used as a drop-in technique to not\nonly accelerate the popular pretrained stable diffusion (SD) models but also\nimprove the prompt alignment of stylized SD models from the public model zoo.\nCode is released at https://github.com/NVlabs/T-Stitch","upvotes":12,"discussionId":"65d807433381a5be2777e502","ai_summary":"T-Stitch, a technique for efficient sampling from diffusion probabilistic models, leverages smaller models early in the process to reduce computational cost without degrading image quality.","ai_keywords":["diffusion probabilistic models","sampling Trajectory Stitching T-Stitch","sampling efficiency","diffusion models","training-free","fast sampling techniques","DiT-XL","DiT-S","pretrained stable diffusion","prompt alignment"]},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"630c2ddb86b8b9904c3860a6","avatarUrl":"/avatars/9b6cec2e9e269ccac1533eb7bf1ac2c5.svg","isPro":false,"fullname":"Igor Melnyk","user":"imelnyk","type":"user"},{"_id":"629d55b53a3221bb2117422c","avatarUrl":"/avatars/e39f25cd7190d282478cdac20a7700ee.svg","isPro":false,"fullname":"Zizheng Pan","user":"zizhpan","type":"user"},{"_id":"6538119803519fddb4a17e10","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6538119803519fddb4a17e10/ffJMkdx-rM7VvLTCM6ri_.jpeg","isPro":false,"fullname":"samusenps","user":"samusenps","type":"user"},{"_id":"620783f24e28382272337ba4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/620783f24e28382272337ba4/zkUveQPNiDfYjgGhuFErj.jpeg","isPro":false,"fullname":"GuoLiangTang","user":"Tommy930","type":"user"},{"_id":"648eb1eb59c4e5c87dc116e0","avatarUrl":"/avatars/c636cea39c2c0937f01398c94ead5dad.svg","isPro":false,"fullname":"fdsqefsgergd","user":"T-representer","type":"user"},{"_id":"6101c620900eaa0057c2ce1d","avatarUrl":"/avatars/bd282166c120711c65b5409dc860ac58.svg","isPro":false,"fullname":"Abdel-Dayane Marcos","user":"admarcosai","type":"user"},{"_id":"65a8a828ef14f9e6037dfad2","avatarUrl":"/avatars/55fdc0393b2e37ac998ddbdb1e1ff636.svg","isPro":false,"fullname":"Bohan Zhuang","user":"BohanZ","type":"user"},{"_id":"63c5d43ae2804cb2407e4d43","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1673909278097-noauth.png","isPro":false,"fullname":"xziayro","user":"xziayro","type":"user"},{"_id":"663ccbff3a74a20189d4aa2e","avatarUrl":"/avatars/83a54455e0157480f65c498cd9057cf2.svg","isPro":false,"fullname":"Nguyen Van Thanh","user":"NguyenVanThanhHust","type":"user"},{"_id":"64b695dcd3df8086e5ed7c89","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64b695dcd3df8086e5ed7c89/06Toh65jDEz3WJbIM6ZmZ.jpeg","isPro":false,"fullname":"Adam Fields","user":"adamelliotfields","type":"user"},{"_id":"678e587c7673ebe9207de9f8","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/HsvG_fJWRMVqjIUpcWl-s.png","isPro":false,"fullname":"Jayanta debnath","user":"jayant9800","type":"user"},{"_id":"67c07dc7a261b07d0f511cd8","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/67c07dc7a261b07d0f511cd8/o4UmaHp8v9V6kNCChihT3.jpeg","isPro":false,"fullname":"KIMMMM","user":"Donghjae","type":"user"}],"acceptLanguages":["*"],"dailyPaperRank":0}">

Papers

arxiv:2402.14167

T-Stitch: Accelerating Sampling in Pre-Trained Diffusion Models with Trajectory Stitching

Published on Feb 21, 2024

· Submitted by

AK on Feb 23, 2024

Upvote

Authors:

Zizheng Pan ,

Bohan Zhuang ,

De-An Huang ,

Weili Nie ,

Zhiding Yu ,

Anima Anandkumar

Abstract

T-Stitch, a technique for efficient sampling from diffusion probabilistic models, leverages smaller models early in the process to reduce computational cost without degrading image quality.

AI-generated summary

Sampling from diffusion probabilistic models (DPMs) is often expensive for high-quality image generation and typically requires many steps with a large model. In this paper, we introduce sampling Trajectory Stitching T-Stitch, a simple yet efficient technique to improve the sampling efficiency with little or no generation degradation. Instead of solely using a large DPM for the entire sampling trajectory, T-Stitch first leverages a smaller DPM in the initial steps as a cheap drop-in replacement of the larger DPM and switches to the larger DPM at a later stage. Our key insight is that different diffusion models learn similar encodings under the same training data distribution and smaller models are capable of generating good global structures in the early steps. Extensive experiments demonstrate that T-Stitch is training-free, generally applicable for different architectures, and complements most existing fast sampling techniques with flexible speed and quality trade-offs. On DiT-XL, for example, 40% of the early timesteps can be safely replaced with a 10x faster DiT-S without performance drop on class-conditional ImageNet generation. We further show that our method can also be used as a drop-in technique to not only accelerate the popular pretrained stable diffusion (SD) models but also improve the prompt alignment of stylized SD models from the public model zoo. Code is released at https://github.com/NVlabs/T-Stitch