frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

AI discovers a 5x faster MoE load balancing algorithm than human experts

https://adrs-ucb.notion.site/moe-load-balancing
99•melissapan•5h ago

Comments

accheng•4h ago
As an author of the blog, I'll note that this was one of the easiest applications of ADRS. Bowen, who was leading this effort, got things running within a day or two and the initial runs were with free Google credits! It was exciting to see how quickly these kinds of frameworks could be applied to real-world engineering and algorithmic challenges.
logicallee•3h ago
did AI explain its thinking, or could it have just stumbled upon the solution without designing it or understanding why it worked? i.e. could it have just been a hallucination that happened to work?
accheng•3h ago
This is a great question! By analyzing the logs of OpenEvolve with the full model outputs, we observed how the AI got its ideas (seemed to be pulling from literature in the space) and how it tried to apply them. So in some sense, it "reasoned" about how to get better algorithms. And we saw this process proceed systematically via the ADRS framework to converge to a significantly better algorithm
logicallee•2h ago
very interesting, thank you.
Izikiel43•1h ago
Can you confirm if this generated code is the same as https://arxiv.org/pdf/2402.02447 ?
liamYC•24m ago
What does ADRS stand for?
_--__--__•3h ago
Nice result, but the snake pattern is pretty obvious and intuitive even for a human who just glances over the problem. It kinda breaks if there is huge variance (if the top load expert is orders of magnitude higher than #2 it probably should just get its own GPU), but I'm not familiar enough with MoE to know if that's a realistic possibility.
abmfy•2h ago
Thanks! In realistic workloads, the differences won’t be orders of magnitude.

I agree that this is a fairly simple problem. Experienced engineers—or anyone who has faced similar challenges—can quickly come up with such solutions. The key point, however, is that others might get stuck in their research simply because they don’t realize these quick solutions exist (“I don’t know what I don’t know”). AI helps bridge that gap by making expert-level knowledge accessible to every researcher, allowing them to focus more on exploring the truly unknown parts.

bgwalter•2h ago
Except that "AI" steals and mostly does not do citations.

EDIT: The chutzpah of downvoting this is striking. The paper says "surpasses highly optimized algorithms engineered by human experts to achieve a 5.0x speedup" and https://news.ycombinator.com/item?id=45689663 links to a 2024 paper where humans discovered a 4.2x speedup using a snake pattern. The 2024 paper is not cited.

pakt1•2h ago
that's true for any application of AI :(
dzogchen•3h ago
Alternate title: “Human experts discover a 5x faster MoE load balancing algorithm using AI”
anonymous908213•1h ago
Better title: "Clueless humans use AI to plagiarise an algorithm they didn't know existed, assume they discovered it".

We are in the absolute worst timeline.

taqpos•2h ago
i wonder how hard it is to get the setup for AI to evolve on?
abmfy•2h ago
I spent 2~3 hours setting up, most of the time was spent on writing the evaluator

Actually I think the evaluator will be the most important part for the whole pipeline to work

accheng•2h ago
Yes, getting the right workloads and ensuring correctness are crucial parts of the process
qat321•2h ago
Really cool to see the AI-discovered algorithm is not just a theoretical result but is actually in a PR for vLLM. My question is about the code itself. Was the Python/PyTorch generated by OpenEvolve directly usable, or did it require significant human cleanup to make it readable, maintainable, and conform to the project's coding standards? I'm curious about how close we are to AI generating production-ready, human-editable code for complex algorithms.
abmfy•2h ago
It's directly usable, since it need to pass the evaluator first; also it contains clear comments about the intent
nomel•1h ago
I assume this means it still went through human review, more than the evaluator was complete enough to not require it?
bgwalter•2h ago
I'm not sure if this is the exact same thing, but a load balancing paper reported a 4.2x speedup by applying a "snake pattern" in 2024:

https://arxiv.org/pdf/2402.02447

letitgo12345•2h ago
Seems the same tbh
abmfy•1h ago
Thanks for letting us know! While we’re tackling different problems, the core idea around load balancing is quite similar.

The pattern might be a familiar trick to those experienced with this kind of problem — you can see my thoughts on it here: https://news.ycombinator.com/item?id=45688236#45689440

Jweb_Guru•11m ago
It's okay to acknowledge that you missed something in your literature search. Everyone does. It's not okay to sweep it under the rug or pretend that it's novel after having having the prior work pointed out to you, especially when a central part of your thesis is that "AI" discovered a novel algorithm and it's very likely that this algorithm was part of the LLM's training data.
coliveira•45m ago
Most probably the AI was secretly tested on this data and is just stealing the algorithm.
pos456•2h ago
this feels less like Copilot and more like AlphaGo for systems programming. it's not just finding patterns in existing code, but discovering novel and more efficient strategies in a given problem space. Very cool.
quc1k•2h ago
The final code might be fast, but is it understandable? The evolution process shows it tried a bunch of things that didn't work. The final result is a heuristic that won out based on a specific simulator and fitness function.
accheng•2h ago
The code was quite short and easy to read. Specifying the right scoring function and scoping the problem are key parts of getting good results with ADRS.
snklt•1h ago
does this only work for vLLM or is generally applicable?
mavt6•1h ago
i'm skeptical this generalizes beyond problems that can be expressed as "rearrange tensors faster". it feels like a solution that only works for a very narrow and convenient class of problems.
maven5t•1h ago
getting a 5x speedup for less than $10 and in just five hours is insane. the roi on this approach is going to be hard to beat.
joaohaas•1h ago
So, if I got this right, this is just about re-implementing an existing load balancing algorithm faster...? If so, this is really dumb. As you guys checked out, yes most load balancing algorithms are slow/dumb:

>First, we evaluate DeepSeek's open-source EPLB implementation. This employs a greedy bin-packing strategy: experts are sorted by load in descending order, and each is placed onto the least-loaded GPU that has capacity (Figure 3a, Example 1). While simple, the solution is slow because it written in Python and uses a for-loop to performs linear search for finding the best-fit GPU choice.

This is because when considering a load balancing algorithm, unless the work being done (in this case by the GPU) lasts only a few ms, the load balancing algorithm being fast will never be the bottleneck. The post does not mention whether this is the case at all.

Also, I don't want to sound rude, but if all they managed to get is a 5x increase over a simple python algorithm, I don't think this is impressive at all...? Any rewrite of the 'dumb' algorithm in a language with more memory control and cache continuity should result in much better results.

teunlao•1h ago
Agree. Starting from Python for-loops is embarrassing baseline. Any decent implementation gets you most of that 5x for free. The interesting part isn't the speedup - it's that AI can do routine optimization unsupervised. That's the actual value prop.
abmfy•44m ago
Thanks for commenting! Actually in this case, "the work being done" can be really fast because it can be done asynchronously. For context, here’s how this translates in a real-world application.

The original algorithm was provided by DeepSeek, and our optimized implementation achieves a 92× speedup over it. The 5x number is comparing with another baseline that is undisclosed yet.

When integrating EPLB into vLLM, I discovered—somewhat unexpectedly—that the open-source algorithm consumes nearly half of the total time of a rearrangement step, with the remaining time spent transferring weights across GPUs. To address this, I applied OpenEvolve to the algorithm, setting the primary objective to improve speed while maintaining the same balance factor. It performed remarkably well. With additional optimizations on the weight transferring, the overall overhead has now become almost negligible.

chanwutk•1h ago
Great read.
coliveira•47m ago
The idea that AI can discover anything is ridiculous. It can propose algorithms like it creates any piece of text, but only the human researcher is capable of analyzing the algorithm, proving that it works, understand what it is doing, i.e., pretty much everything that we call a new "discovery". I would have zero confidence in an algorithm "discovered" by an AI in isolation.
PostOnce•42m ago
Theoretically if I were to type into an LLM "Write a novel compression algorithm for images that is at least 25% smaller at the same speed and quality as ___" and it did, and I ran the code (which I didn't understand) and it worked, wouldn't that count?

The odds of that working, though, are of course pretty near 0. But theoretically, it could happen.

malfist•40m ago
You might find that if it did produce something, it might not be _novel_
adrianN•16m ago
That is a problem human researchers face too.
coliveira•30m ago
As you say, the odds of this happening are very close to zero. But suppose for a minute that this was possible. Did you learn anything? Do you really have a discovery? Was this done using a novel method or applying something that already exists? If you give this to somebody else, should they believe it works? Is this result even understandable by human beings? You'd need to answer so many questions that in the end even this would NOT be a discovery by the machine but by yourself.
holografix•29m ago
It can propose algorithms which than it can _itself test and iterate on_
Noumenon72•47m ago
There's nowhere on the page to find out what "ADRS" stands for since the upper left is cut off and isn't a link to your home page.
kristjansson•8m ago
This is quite cool, but I must note that the 5x reported in the headline is the _runtime_ of the load balancing algorithm itself, not the load factor or throughput of the system or what have you.

> On average, it takes about 540 ms to re-balance the experts and achieves a load balance factor of 0.66 (calculated as the ratio of average to maximum tokens generated per GPU).

> ...

> We also consider a non-public reference implementation from a frontier lab that we have access to. This implementation avoids explicit iteration and reduces the rebalancing algorithm runtime to 19.6 ms while achieving the same balance factor as the open-source algorithm.

> ...

> The resulting algorithm matches the load balance factor of the other baselines while reducing runtime to just 3.7 ms, yielding a 5.0x speedup over the internal reference implementation.

Computer Science Courses That Don't Exist, but Should (2015)

https://prog21.dadgum.com/210.html
54•wonger_•1h ago•31 comments

Betty White's shoulder bag is a time capsule of World War II (2023)

https://americanhistory.si.edu/explore/stories/betty-white-world-war-ii
144•thunderbong•6d ago•8 comments

Claude Memory

https://www.anthropic.com/news/memory
389•doppp•10h ago•221 comments

/dev/null is an ACID compliant database

https://jyu.dev/blog/why-dev-null-is-an-acid-compliant-database/
189•swills•6h ago•89 comments

How memory maps (mmap) deliver faster file access in Go

https://info.varnish-software.com/blog/how-memory-maps-mmap-deliver-25x-faster-file-access-in-go
81•ingve•5h ago•49 comments

React Flow, open source libraries for node-based UIs with React or Svelte

https://github.com/xyflow/xyflow
66•mountainview•4h ago•9 comments

Cheap DIY solar fence design

https://joeyh.name/blog/entry/cheap_DIY_solar_fence_design/
50•kamaraju•1w ago•15 comments

Can “second life” EV batteries work as grid-scale energy storage?

https://www.volts.wtf/p/can-second-life-ev-batteries-work
117•davidw•9h ago•133 comments

When is it better to think without words?

https://www.henrikkarlsson.xyz/p/wordless-thought
76•Curiositry•6h ago•30 comments

AI discovers a 5x faster MoE load balancing algorithm than human experts

https://adrs-ucb.notion.site/moe-load-balancing
99•melissapan•5h ago•42 comments

Dunkin' Donuts started as Quincy, MA coffee/doughnut shop: Now 10k US locations

https://www.bostonglobe.com/2025/10/10/business/dunkin-donuts-scott-murphy-bain-capital-carlyle-i...
6•bookofjoe•1w ago•1 comments

PyTorch Monarch

https://pytorch.org/blog/introducing-pytorch-monarch/
319•jarbus•17h ago•41 comments

Deepstaria Enigmatica

https://en.wikipedia.org/wiki/Deepstaria_enigmatica
13•handfuloflight•1w ago•1 comments

Apple loses UK App Store monopoly case, penalty might near $2B

https://9to5mac.com/2025/10/23/apple-loses-uk-app-store-monopoly-case-penalty-might-near-2-billion/
190•thelastgallon•5h ago•163 comments

Benchmarking Postgres 17 vs. 18

https://planetscale.com/blog/benchmarking-postgres-17-vs-18
28•bddicken•1w ago•2 comments

Introduction to the concept of likelihood and its applications (2018)

https://journals.sagepub.com/doi/10.1177/2515245917744314
22•sebg•4h ago•3 comments

Zram Performance Analysis

https://notes.xeome.dev/notes/Zram
64•enz•7h ago•19 comments

Summary of the Amazon DynamoDB Service Disruption in US-East-1 Region

https://aws.amazon.com/message/101925/
437•meetpateltech•1d ago•127 comments

OpenAI acquires Sky.app

https://openai.com/index/openai-acquires-software-applications-incorporated
141•meetpateltech•10h ago•82 comments

US probes Waymo robotaxis over school bus safety

https://www.yahoo.com/news/articles/us-investigates-waymo-robotaxis-over-102015308.html
83•gmays•14h ago•122 comments

New updates and more access to Google Earth AI

https://blog.google/technology/research/new-updates-and-more-access-to-google-earth-ai/
133•diogenico•10h ago•40 comments

Pyscripter – Open-source Python IDE written in Delphi

https://github.com/pyscripter/pyscripter
54•peter_d_sherman•4d ago•8 comments

Armed police swarm student after AI mistakes bag of Doritos for a weapon

https://www.dexerto.com/entertainment/armed-police-swarm-student-after-ai-mistakes-bag-of-doritos...
520•antongribok•9h ago•319 comments

Kaitai Struct: declarative binary format parsing language

https://kaitai.io/
87•djoldman•1w ago•32 comments

Trump pardons convicted Binance founder

https://www.wsj.com/finance/currencies/trump-pardons-convicted-binance-founder-7509bd63
775•cowboyscott•11h ago•810 comments

Show HN: Git for LLMs – A context management interface

https://twigg.ai
69•jborland•12h ago•21 comments

I managed to grow countable yeast colonies

https://chillphysicsenjoyer.substack.com/p/i-managed-to-grow-countable-yeast
27•crescit_eundo•1w ago•7 comments

Show HN: OpenSnowcat – A fork of Snowplow to keep open analytics alive

https://opensnowcat.io/
59•joaocorreia•8h ago•14 comments

Glasses-free 3D using webcam head tracking

https://assetstore.unity.com/packages/tools/camera/vr-without-glasses-for-webgl-332314
80•il_nets•5d ago•65 comments

Living Dangerously with Claude

https://simonwillison.net/2025/Oct/22/living-dangerously-with-claude/
153•FromTheArchives•1d ago•76 comments