Skip to content

Commit 64baef4

Browse files
committed
deploy: 03ef597
1 parent 939f114 commit 64baef4

File tree

2 files changed

+100
-5
lines changed

2 files changed

+100
-5
lines changed

references/index.html

Lines changed: 95 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -4233,6 +4233,101 @@ <h2 id="year2024" class="year">2024 <span class="count-stat"></span></h2>
42334233
</div>
42344234
</div>
42354235

4236+
</div>
4237+
</li>
4238+
<li><div class="bibtex-entry-container">
4239+
<div>
4240+
<div class="col-md-3 col-sm-12 bibtex-ref-meta hidden">
4241+
<div class="row">
4242+
<div class="col-md-12 ref-label label label-default">
4243+
SelvamEtAl2024
4244+
</div>
4245+
</div>
4246+
</div>
4247+
4248+
<div class="bibtex-ref-entry">
4249+
<span id="SelvamEtAl2024">N. R. Selvam, A. Merchant, and S. Ermon, “Self-Refining Diffusion Samplers: Enabling Parallelization via Parareal Iterations,” arXiv:2412.08292v1 [cs.LG], 2024 [Online]. Available at: <a href="http://arxiv.org/abs/2412.08292v1" target="_blank">http://arxiv.org/abs/2412.08292v1</a></span>
4250+
</div>
4251+
4252+
4253+
<div class="row mt-4">
4254+
<div class="btn-group btn-group-xs btn-group-justified" role="group">
4255+
<div class="btn-group btn-group-xs" role="group">
4256+
4257+
<button class="button" data-toggle="modal" data-target="#modalSelvamEtAl2024Bibtex" aria-expanded="false"
4258+
aria-controls="modalSelvamEtAl2024Bibtex">
4259+
BibTeX
4260+
</button>
4261+
4262+
</div>
4263+
<div class="btn-group btn-group-xs" role="group">
4264+
4265+
<button class="button" data-toggle="modal" data-target="#modalSelvamEtAl2024Abstract" aria-expanded="false"
4266+
aria-controls="modalSelvamEtAl2024Abstract">
4267+
Abstract
4268+
</button>
4269+
4270+
</div>
4271+
</div>
4272+
</div>
4273+
</div>
4274+
4275+
<div class="modal" id="modalSelvamEtAl2024Bibtex" tabindex="-1" role="dialog"
4276+
aria-labelledby="modalSelvamEtAl2024BibtexTitle" aria-hidden="true">
4277+
<div class="modal-background">
4278+
<div class="modal-card">
4279+
<div class="modal-content">
4280+
<header class="modal-card-head">
4281+
<div class="modal-card-head-content">
4282+
<div>
4283+
<p class="modal-card-title" id="modalSelvamEtAl2024BibtexTitle">
4284+
BibTeX entry <code>SelvamEtAl2024</code>
4285+
</p>
4286+
</div>
4287+
<button type="button" class="delete" data-dismiss="modal" aria-label="Close"><span
4288+
aria-hidden="true">&times;</span></button>
4289+
</div>
4290+
</header>
4291+
<section class="modal-card-body">
4292+
<pre class="abstract">@unpublished{SelvamEtAl2024,
4293+
author = {Selvam, Nikil Roashan and Merchant, Amil and Ermon, Stefano},
4294+
howpublished = {arXiv:2412.08292v1 [cs.LG]},
4295+
title = {Self-Refining Diffusion Samplers: Enabling Parallelization via Parareal Iterations},
4296+
url = {http://arxiv.org/abs/2412.08292v1},
4297+
year = {2024}
4298+
}
4299+
</pre>
4300+
</section>
4301+
</div>
4302+
</div>
4303+
</div>
4304+
</div>
4305+
4306+
4307+
<div class="modal" id="modalSelvamEtAl2024Abstract" tabindex="-1" role="dialog"
4308+
aria-labelledby="modalSelvamEtAl2024AbstractTitle" aria-hidden="true">
4309+
<div class="modal-background">
4310+
<div class="modal-card">
4311+
<div class="modal-content">
4312+
<header class="modal-card-head">
4313+
<div class="modal-card-head-content">
4314+
<div>
4315+
<p class="modal-card-title" id="modalSelvamEtAl2024AbstractTitle">
4316+
Abstract for BibTeX entry <code>SelvamEtAl2024</code>
4317+
</p>
4318+
</div>
4319+
<button type="button" class="delete" data-dismiss="modal" aria-label="Close"><span
4320+
aria-hidden="true">&times;</span></button>
4321+
</div>
4322+
</header>
4323+
<section class="modal-card-body">
4324+
In diffusion models, samples are generated through an iterative refinement process, requiring hundreds of sequential model evaluations. Several recent methods have introduced approximations (fewer discretization steps or distillation) to trade off speed at the cost of sample quality. In contrast, we introduce Self-Refining Diffusion Samplers (SRDS) that retain sample quality and can improve latency at the cost of additional parallel compute. We take inspiration from the Parareal algorithm, a popular numerical method for parallel-in-time integration of differential equations. In SRDS, a quick but rough estimate of a sample is first created and then iteratively refined in parallel through Parareal iterations. SRDS is not only guaranteed to accurately solve the ODE and converge to the serial solution but also benefits from parallelization across the diffusion trajectory, enabling batched inference and pipelining. As we demonstrate for pre-trained diffusion models, the early convergence of this refinement procedure drastically reduces the number of steps required to produce a sample, speeding up generation for instance by up to 1.7x on a 25-step StableDiffusion-v2 benchmark and up to 4.3x on longer trajectories.
4325+
</section>
4326+
</div>
4327+
</div>
4328+
</div>
4329+
</div>
4330+
42364331
</div>
42374332
</li>
42384333
<li><div class="bibtex-entry-container">

sitemap.xml

Lines changed: 5 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -354,16 +354,19 @@
354354
<loc>http://parallel-in-time.org/about/imprint.html</loc>
355355
</url>
356356
<url>
357-
<loc>http://parallel-in-time.org/codes/</loc>
357+
<loc>http://parallel-in-time.org/references/</loc>
358358
</url>
359359
<url>
360360
<loc>http://parallel-in-time.org/groups/</loc>
361361
</url>
362362
<url>
363+
<loc>http://parallel-in-time.org/methods/</loc>
364+
</url>
365+
<url>
363366
<loc>http://parallel-in-time.org/projects/</loc>
364367
</url>
365368
<url>
366-
<loc>http://parallel-in-time.org/references/</loc>
369+
<loc>http://parallel-in-time.org/codes/</loc>
367370
</url>
368371
<url>
369372
<loc>http://parallel-in-time.org/events/upcoming/</loc>
@@ -375,9 +378,6 @@
375378
<loc>http://parallel-in-time.org/events/</loc>
376379
</url>
377380
<url>
378-
<loc>http://parallel-in-time.org/methods/</loc>
379-
</url>
380-
<url>
381381
<loc>http://parallel-in-time.org/about/</loc>
382382
</url>
383383
<url>

0 commit comments

Comments
 (0)