https://arxiv.org/abs/2510.17699
👾 Code: https://github.com/3145tttt/GAS

\n","updatedAt":"2025-10-22T13:25:10.397Z","author":{"_id":"64a8fc71baf671fbebb4d032","avatarUrl":"/avatars/cffd8c2154dfa57e2ce526996992ffb8.svg","fullname":"Aleksandr Oganov","name":"3145tttt","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":1,"identifiedLanguage":{"language":"en","probability":0.831738293170929},"editors":["3145tttt"],"editorAvatarUrls":["/avatars/cffd8c2154dfa57e2ce526996992ffb8.svg"],"reactions":[],"isReport":false}},{"id":"68f9861543e0b678e1deab12","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":318,"isUserFollowing":false},"createdAt":"2025-10-23T01:34:13.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"This is an automated message from the [Librarian Bot](https://huggingface.co/librarian-bots). I found the following papers similar to this paper. \n\nThe following papers were recommended by the Semantic Scholar API \n\n* [Soft-Di[M]O: Improving One-Step Discrete Image Generation with Soft Embeddings](https://huggingface.co/papers/2509.22925) (2025)\n* [POSE: Phased One-Step Adversarial Equilibrium for Video Diffusion Models](https://huggingface.co/papers/2508.21019) (2025)\n* [Universal Inverse Distillation for Matching Models with Real-Data Supervision (No GANs)](https://huggingface.co/papers/2509.22459) (2025)\n* [Large Scale Diffusion Distillation via Score-Regularized Continuous-Time Consistency](https://huggingface.co/papers/2510.08431) (2025)\n* [Score Distillation of Flow Matching Models](https://huggingface.co/papers/2509.25127) (2025)\n* [SSDD: Single-Step Diffusion Decoder for Efficient Image Tokenization](https://huggingface.co/papers/2510.04961) (2025)\n* [Score-based Idempotent Distillation of Diffusion Models](https://huggingface.co/papers/2509.21470) (2025)\n\n\n Please give a thumbs up to this comment if you found it helpful!\n\n If you want recommendations for any Paper on Hugging Face checkout [this](https://huggingface.co/spaces/librarian-bots/recommend_similar_papers) Space\n\n You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: `@librarian-bot recommend`","html":"

This is an automated message from the Librarian Bot. I found the following papers similar to this paper.

The following papers were recommended by the Semantic Scholar API

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":"2025-10-23T01:34:13.666Z","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":318,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.6927080750465393},"editors":["librarian-bot"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2510.17699","authors":[{"_id":"68f77d6ca7b9b96b2c13ed7d","user":{"_id":"64a8fc71baf671fbebb4d032","avatarUrl":"/avatars/cffd8c2154dfa57e2ce526996992ffb8.svg","isPro":false,"fullname":"Aleksandr Oganov","user":"3145tttt","type":"user"},"name":"Aleksandr Oganov","status":"claimed_verified","statusLastChangedAt":"2025-10-21T14:00:21.831Z","hidden":false},{"_id":"68f77d6ca7b9b96b2c13ed7e","name":"Ilya Bykov","hidden":false},{"_id":"68f77d6ca7b9b96b2c13ed7f","name":"Eva Neudachina","hidden":false},{"_id":"68f77d6ca7b9b96b2c13ed80","name":"Mishan Aliev","hidden":false},{"_id":"68f77d6ca7b9b96b2c13ed81","name":"Alexander Tolmachev","hidden":false},{"_id":"68f77d6ca7b9b96b2c13ed82","name":"Alexander Sidorov","hidden":false},{"_id":"68f77d6ca7b9b96b2c13ed83","name":"Aleksandr Zuev","hidden":false},{"_id":"68f77d6ca7b9b96b2c13ed84","name":"Andrey Okhotin","hidden":false},{"_id":"68f77d6ca7b9b96b2c13ed85","name":"Denis Rakitin","hidden":false},{"_id":"68f77d6ca7b9b96b2c13ed86","name":"Aibek Alanov","hidden":false}],"mediaUrls":["https://cdn-uploads.huggingface.co/production/uploads/64a8fc71baf671fbebb4d032/02z15_R2mQTvqO0b2W3gr.png"],"publishedAt":"2025-10-20T16:14:38.000Z","submittedOnDailyAt":"2025-10-22T11:54:57.307Z","title":"GAS: Improving Discretization of Diffusion ODEs via Generalized\n Adversarial Solver","submittedOnDailyBy":{"_id":"64a8fc71baf671fbebb4d032","avatarUrl":"/avatars/cffd8c2154dfa57e2ce526996992ffb8.svg","isPro":false,"fullname":"Aleksandr Oganov","user":"3145tttt","type":"user"},"summary":"While diffusion models achieve state-of-the-art generation quality, they\nstill suffer from computationally expensive sampling. Recent works address this\nissue with gradient-based optimization methods that distill a few-step ODE\ndiffusion solver from the full sampling process, reducing the number of\nfunction evaluations from dozens to just a few. However, these approaches often\nrely on intricate training techniques and do not explicitly focus on preserving\nfine-grained details. In this paper, we introduce the Generalized Solver: a\nsimple parameterization of the ODE sampler that does not require additional\ntraining tricks and improves quality over existing approaches. We further\ncombine the original distillation loss with adversarial training, which\nmitigates artifacts and enhances detail fidelity. We call the resulting method\nthe Generalized Adversarial Solver and demonstrate its superior performance\ncompared to existing solver training methods under similar resource\nconstraints. Code is available at https://github.com/3145tttt/GAS.","upvotes":25,"discussionId":"68f77d6ca7b9b96b2c13ed87","githubRepo":"https://github.com/3145tttt/GAS","githubRepoAddedBy":"user","ai_summary":"The Generalized Adversarial Solver improves diffusion model sampling efficiency and quality by combining a simple ODE solver parameterization with adversarial training.","ai_keywords":["diffusion models","gradient-based optimization","ODE diffusion solver","function evaluations","adversarial training","Generalized Solver","Generalized Adversarial Solver"],"githubStars":73,"organization":{"_id":"67be94e466f702bfed94ac7d","name":"Bayes-Group","fullname":"Bayesian Methods Research Group","avatar":"https://cdn-uploads.huggingface.co/production/uploads/67be93b43e718f259b6abbe3/6Swfg5YVWup39avDzjk98.jpeg"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"6815f1d874e57666851e2dfa","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/v109P6K_zgqATwTX_xucP.png","isPro":false,"fullname":"Dmitrii Feoktistov","user":"TrandeLik","type":"user"},{"_id":"66680c6451545a8b46c6fd21","avatarUrl":"/avatars/03707f5ea4e2aa8dc825a9782b00ed85.svg","isPro":false,"fullname":"Aibek Alanov","user":"ai-alanov","type":"user"},{"_id":"6262880c5eb4fa93219f0064","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6262880c5eb4fa93219f0064/6yyBvRK4Oh7OhjaaweaVN.jpeg","isPro":false,"fullname":"Daniil Tiapkin","user":"dtiapkin","type":"user"},{"_id":"64d403f7d0cfd876d93392d7","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/TKJXT6GZ5Wv14gcuBAIpQ.jpeg","isPro":false,"fullname":"Ilya Alexeev","user":"voorhs","type":"user"},{"_id":"625c1c1f1e019bb7c2d9371c","avatarUrl":"/avatars/ab26d255d2becee5e530100cbf8c6b14.svg","isPro":false,"fullname":"Mikhail Kuznetsov","user":"mmkuznecov","type":"user"},{"_id":"6548b9a2e70ffa3c0751ce20","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6548b9a2e70ffa3c0751ce20/eA1Tg-eNuK708rrJF-AFR.jpeg","isPro":false,"fullname":"Tytskiy Vladislav","user":"tytskiy","type":"user"},{"_id":"63162d8095b55e2621d1aa1f","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1672404315143-63162d8095b55e2621d1aa1f.jpeg","isPro":false,"fullname":"Ildus Sadrtdinov","user":"isadrtdinov","type":"user"},{"_id":"64fd75ae52e82dd432ac35ef","avatarUrl":"/avatars/74fbea71701a14ab5db3e94b2a6ad3fd.svg","isPro":false,"fullname":"Art","user":"B1KO1L","type":"user"},{"_id":"68ed288aef5fa1ceaac3d418","avatarUrl":"/avatars/1d49b127c77a8d5872d7e5def67507af.svg","isPro":false,"fullname":"Kudryashov Kirill","user":"Kirill22022004","type":"user"},{"_id":"6846cb4810d96a43cf779455","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/fhKXSHLxmiVYvjxrqeltN.png","isPro":false,"fullname":"Rogovoi Sergei","user":"leesfy","type":"user"},{"_id":"60cd95ee15ecba5f2200304a","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60cd95ee15ecba5f2200304a/3gMYeWm8wQO5KfqE5RmEe.jpeg","isPro":false,"fullname":"Alexey Dontsov","user":"therem","type":"user"},{"_id":"646f8807cc049b49e35e06dc","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/646f8807cc049b49e35e06dc/UmPgkpEcqkn8n-OADa_nz.png","isPro":false,"fullname":"Arseny Ivanov","user":"ArsenyIvanov","type":"user"}],"acceptLanguages":["*"],"dailyPaperRank":0,"organization":{"_id":"67be94e466f702bfed94ac7d","name":"Bayes-Group","fullname":"Bayesian Methods Research Group","avatar":"https://cdn-uploads.huggingface.co/production/uploads/67be93b43e718f259b6abbe3/6Swfg5YVWup39avDzjk98.jpeg"}}">

Papers

arxiv:2510.17699

GAS: Improving Discretization of Diffusion ODEs via Generalized Adversarial Solver

Published on Oct 20, 2025

· Submitted by

Aleksandr Oganov on Oct 22, 2025

Bayesian Methods Research Group

Upvote

Authors:

Aleksandr Oganov ,

Abstract

The Generalized Adversarial Solver improves diffusion model sampling efficiency and quality by combining a simple ODE solver parameterization with adversarial training.

AI-generated summary

While diffusion models achieve state-of-the-art generation quality, they still suffer from computationally expensive sampling. Recent works address this issue with gradient-based optimization methods that distill a few-step ODE diffusion solver from the full sampling process, reducing the number of function evaluations from dozens to just a few. However, these approaches often rely on intricate training techniques and do not explicitly focus on preserving fine-grained details. In this paper, we introduce the Generalized Solver: a simple parameterization of the ODE sampler that does not require additional training tricks and improves quality over existing approaches. We further combine the original distillation loss with adversarial training, which mitigates artifacts and enhances detail fidelity. We call the resulting method the Generalized Adversarial Solver and demonstrate its superior performance compared to existing solver training methods under similar resource constraints. Code is available at https://github.com/3145tttt/GAS.

View arXiv page View PDF GitHub 73 Add to collection

Community

3145tttt

Paper author Paper submitter Oct 22, 2025

•

edited Oct 22, 2025

We introduce the Generalized Adversarial Solver (GAS) — a simple yet powerful approach to greatly accelerate diffusion model sampling (up to 5x) without sacrificing generation quality and fine-grained detail

Our method is based on a novel trainable parametrization of a solver, which can adapt to the diffusion model utilized in your generation pipeline, thus, solve the underlying ODE with more precision

More details about method's implementations, experiments, and ablation studies can be found in our preprint

📄 Preprint: https://arxiv.org/abs/2510.17699
👾 Code: https://github.com/3145tttt/GAS