Huawei releases an open weight model trained on Huawei Ascend GPUs
arxiv.org
arxiv.org
I was sceptical of the US sanctions but this seems like a real win if this can be taken all the way to its logical conclusions.
It's already technically feasible: https://www.primeintellect.ai/blog/intellect-2
GPT‑4.1 scores 54.6% on SWE-bench Verified, improving by 21.4%abs over GPT‑4o and 26.6%abs over GPT‑4.5—making it a leading model for coding.
People sang praise from the roof for Google's Gemini 2.5 models, but in many things for me they can't even beat Deepseek V3.
In terms of code and science, Gemini is way, way too verbose in its output, and because of that it ends up getting confused by itself and hurting the quality of longer windows.
R1 does this too, but it poisons itself in the reasoning loop. You can see it during the streaming, literally criss-crossing its thoughts and thinking itself into loops before it finally arrives at an answer.
On top of that, both R1 and Gemini Pro / Flash are mediocre at anything creative. I can accept that from R1, since it's mainly meant as more of a "hard sciences" model, but Gemini is meant to be an all-purpose model.
If you pit Gemini, Deepseek R1 and Deepseek V3 against each other in a writing contest, V3 will blow both of them out of the water.
But in general 2.5 Pro is an extremely strong model. It may lose out in some respects to o3-pro, but o3-pro is so much slower that its utility tends to be limited by my own attention span. I don't think either would have much to fear from V3, though, except possibly in the area of short fiction composition.
I wonder whether you're actually running the proper DeepSeek-R1 model, or one of those lesser finetunes?
Markets used to be places to make money more smart (efficient allocation of capital) but have somehow degraded to index fund buys that track average economic growth of a few hot stocks that are expected to at least not get cold anytime soon.
There are much better ways to increase diversity
Diversity is not pouring oil into water and using the polluted oil-water in lieu of oil and also in lieu of oil. If you want actual diversity you need differences that are separated from each other. It is precisely what has been collapsing for the last 80+ years, actual real diversity, precisely because unique separate groups and clusters have been shattered, scattered, mixed, and polluted.
Even AI is now accelerating this collapse of what is really a form of human biodiversity, or should it be called cultural diversity, as AI is causing a conformity of thought. There are several reports and papers on that phenomenon already.
It’s absolutely ridiculous to claim that somehow those factors will increase over the prior situation simply because we increase actual, real diversity of unique things; not this fake, fraudulent, delusional diversity that has forced on us like a toxic sludge dump that has destroyed human diversity as everyone increasingly consumes the same “content” slop and eats the same food slop, and has the same cultural and musical slop.
https://blog.lambdaclass.com/introducing-demo-decoupled-mome...
Edited several times, I should add: IANAL, but this sounds similar to meta releasing llama weights. I think that the spirit of the European law is to control concrete uses of AI and not a broad distribution of weights and architecture. So my question is: Does the EU AI act ban this distribution?, I think it provides more competition and options for Europeans.
Edited: Thinking a little more, installing open weights could allow backdoors (in the form of a way to manipulate intelligent agents via specials prompts designated to control the system), so perhaps from a national security point of view some care should be taken (but I personally hate that). So another question: Is there a way to control if open weights can create back doors (via prompt injection)?, I recall a paper in which prompt by symbols like 0?,#2! could put the system in a state in which one can read information that the LLM is asked to hide (that is a well known attack available to those that know the weights).
Another question: Is fine tuning or Lora a way to eliminate o amilliorate such prompt attacks?, is there any python library to defend against such attacks. Download - install - modify by fine tune or lora - now you are protected.
"AI systems should fall within the scope of this Regulation even when they are neither placed on the market, nor put into service, nor used in the Union."
I don't really understand the limits of it's scope e.g. the difference between making a system available vs. controlling how it's used is not clear to me. I don't think you can escape the regulation of high-risk uses by offering a "general purpose" AI with no controls on how it's used.
In terms of the open-source nature - I can see it being treated like giving away any other regulated product e.g. medication, cars, safety equipment etc. The lack of cost won't transfer the liability from the supplier to the consumer.
https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:52...
Not to sound too snarky (just a little snarky), I’m just curious how it all works.
As an aside: in general a sovereign country can do whatever they want in their own territory, this includes the right of the country to bind itself to treaties. So in your hypothetical,
> Does your home country observe the EU’s violation of its sovereignty and throw you in a home country jail?
This doesn’t look like a violation of sovereignty to me; the non-EU has decided to enforce an EU law. Why? I don’t know, maybe it makes business easier for the multinational companies of the non-EU country.
Countries can also do things like apply secondary sanctions to an entity. So, again hypothetically, the EU doesn’t need to be able to enforce a ruling against you. They can make you toxic to anyone who wants to do business in the EU.
Not much
It's usually up to the the member states to implement the EU regulation.
As you can see from Hungary with Orban in recent years though the EU's response is slow and lacking.
When there is consensus things can move a lot faster though.
Already happening to Huawei and presumably the EU market is significant to them.
> for example of an operator established in the Union that contracts certain services to an operator established outside the Union in relation to an activity to be performed by an AI system that would qualify as high-risk and whose effects impact natural persons located in the Union.
> this Regulation should also apply to providers and users of AI systems that are established in a third country, to the extent the output produced by those systems is used in the Union
Otherwise it seems to reach way beyond what it actually is.
Explicitly prohibiting EU usage in the license is probably a move to reduce liability under the eyes of those “used in the Union” clauses.
The passage continues:
"To prevent the circumvention of this Regulation and to ensure an effective protection of natural persons located in the Union, this Regulation should also apply to providers and users of AI systems that are established in a third country, to the extent the output produced by those systems is used in the Union."
An AI would come under this regulation even it's just the outputs that are used in the EU. Interesting to think about what that could lead to.
Conversely, I as an individual don't need to worry about it since I don't live there (similar stories for various other overly broad laws).
I agree with you that the usefulness of that clause is suspect given how broad the wording of that law is. How do other companies publishing open models deal with this? For example Meta.
"Protect" ourselves against whom? I'm a EU citizen (unfortunately), and I'm fully on board with China against Brussels. Which is to say, don't try to speak for everyone in this God-forsaken so-called union.
Apply for russian citizenship at their consulate then, they run programs for people of your mindset.
Maybe not the exact thing you're talking about, but that description reminds me of the Alliance for Open Media -- their codec licenses are royalty-free, but the same terms revoke your usage rights if you sue anyone for the use of these formats.
> 3. Conditions for License Grant. You represent and warrant that You will not, access, download, install, run, deploy, integrate, modify, or otherwise use the Model, directly or indirectly, within the European Union.
[1] https://gitcode.com/ascend-tribe/pangu-pro-moe-model/blob/ma...
[0]: https://dionwiggins.substack.com/p/llama-4-is-banned-in-the-...
Edit: it was 8B, not 7B.
Is it something these companies do that they worry violates it?
This would plausibly include anyone developing an LLM, even if they aren't selling access to it or building applications based on it. There are several exemptions, and the Act obstensibly avoids creating burdens for most general-purpose LLMs, but the point is that Huawei wants to avoid any worry by not "plac[ing] it on the market" in the first place.
Probably safe enough on your own computer, but could have consequences if it’s a work computer.
consequences for employer who "might" get a license audit done on their machines.
does it really happen so often that a random employer in the eu would have to be concerned?
If run locally, why?
To answer your question, he modified my comment (see the parentheses):
"> The point is, who gives a damn about (doing an illegal thing) in reality, on their (private property where nobody is likely to see that)?"
So... at best what he said is purely theoretical. He admitted it himself: "nobody is likely to see that". Though I am not sure I agree with it, but then again, in reality, no one gives a fuck, at least not in Europe.
breaking license will do what? whats up with licenses and violations? you and me are random people on internet
The same thing breaking any license does. If you do it in your basement, nothing by definition. If you incorporate it in a service or distribute it as part of a project, well then you're on the hook. (and that is what license holders tend to care about)
The fallout? This AI’s sneakier than a two-headed president—it could snitch to its creators quicker than you can say “Don’t Panic.” If they spot your EU coordinates, you’re in for a galactic stink-eye, with your setup potentially bricked or your data hitchhiking to a dodgy server at the edge of the galaxy. Worse, if the code’s got a nasty streak, your PC could end up a smoking crater, reciting bad poetry in binary.
nord is suggesting it's possible that the physical computer running this model could be used as a "hub" for potential spyware, or be overloaded with workloads that are not related to the actual task of running the model (and instead may be some form of malware performing other computational tasks). It could potentially perform data exfiltration, or act discriminatorily based on your percieved location (such as if you're located within the EU). At worst, data loss or firmware corruption/infection may be of concern in case of license violation.
I'm not sure I would outright disagree that this as possible, but with some caveats. I would think the reason that the license stipulates that usage within the EU is forbidden due to the EU AI Act (here is a resource to read through it: https://artificialintelligenceact.eu/ai-act-explorer/).
same. i call bull on this.
remember how they convinced huawei was public enemy without evidence because nokia and others were unable to compete with them?
can anyone give a technical answer how will weights get to know this fact?
Literally the same for all other open weights, this is just legal ass covering where most others don’t even do that.
FYI, this is not intended to be offensive to Europeans, I am European myself. That is not the point. The point is, who gives a damn about the LICENSE in reality, on their PERSONAL computer? Serious question.
Also any company of a serious size will have lawyers interested in licences of everything you're running.
I know that companies would probably not. But individuals?
I'm not sure which part of that you find confusing. Some people will estimate benefit>risk and won't care.
So what is your answer? Mostly companies only? That is a fair answer, but you are the one who said this:
> You'll be both breaking their licence and potentially your local European data laws.
Again, who cares, dude? Companies might, but individuals probably give a rat's ass. So why leave that comment?
And just for the record, if you quote someone, quote them verbatim, otherwise it is not a quote.
That said, I agree that it is my fault that self-contradicting virtue signaling hypocrites always find a way to irk me.
And I think it is good for the world to know that the LICENSE often means jack shit, unless when companies of significant size are involved.
Again, we all agree that they put it there to cover their own asses, not that Europeans cannot download, install, and run their product, right?
they might license it to companies in the US, but don't want to have to deal with the changes and bureucracy needed to support individuals.
The statement's purpose is to say the equivalent "if you're a European and do run it, it's on you, this is not a product we release or support for the European market, don't expert support, liability, etc".
I mean, this other commenter literally said:
> You'll be both breaking their licence and potentially your local European data laws.
But loyalties don't change facts and China is where serious hackers are rising on merit, doing a lot with limited resourves, giving zero fucks about empty slick talk.
If we wanted to hobble the PRC's technical rise we should have subsidized wasteful NVIDIA use and had Altman/YC be in charge: they'd still be gladhanding about how to pump their portfolio companies sticker price and avoid "systemic shocks" to the stock market anchored on NVDA.
Some just are narc types.
It's not supposed to make them appear as plausibly denying that some European can download and use this.
It's role is to signal that if someone does, it's on him, not them, and he wont have any support, liability claims, etc as if they could if it was a product intended for their use.
It's to make it a matter of legal record that you stated they should abstain.
Copyright warnings on music and DVDs never stopped people pirating them either.
Also it is not rational for any individual to buy the hardware for running a serious LLM and then let it idle 99.9% of the day.
But of course the world changes so rapidly that what is now is irrelevant tomorrow.
There were serious laws limiting the export a "modern" cryptography software from the USA.
Some of us had to face up to the serious challenge of connecting to an FTP server and downloading PGP and risking violating US law to download a software package.
A few years later we had to decide "Do you want the secure Netscape, or the insecure Netscape?".
I'm sure we all chose the ethical choice.
https://en.wikipedia.org/wiki/Export_of_cryptography_from_th...
They will certainly not violate EU laws and also probably not the licence.
My question: is open weight even interesting? What does that really offer? Does it allow one to peer into the biases (or lack thereof) of a model? Does it allow one to train a competing model?
Would open source be something different and preferable — or are "weights the new source" in this LLM world we are finding ourselves in?
I'm trying to educate myself.
I use Gemini to extract structured data from images and the flash model is great at this. I wonder how much effort it would be to create a smaller model that would run on something like a NUC with an AMD APU that is good enough for that one use case.
Or perhaps you end up with mini external GPU sticks that run use case specific models on them. Might not be much of a market for that, but could be pretty cool.
Its only 108 million params.
Maybe it's not really missing and the APIs for LLMs are just too good and cheap to make homebrew stuff exciting.
but more likely it's going on and we're just not seeing it
in general, though, I think once a certain amount of money is involved, people just start to get rabid and everything becomes a lot less fun
It's possible to run models locally, fidget with temp etc
Being able to change other things on the fly like identify weights most used for a prompt and just changing those to see what happens is much harder.
I've tried both LLMS and image generators on my machine locally and while it's gotten in easier it's a long task just setting up. Especially if you run into driver issues.
All the effort is in productizing foundational models and apps built on top of them, but as that plateaus distilled models and new approaches will probably get more time in the sun. I'm hopeful that if this is the case we will see more weird stuff come available.
just recent discussion on HN: "Small language models are the future of agentic AI"
Instead, cutting research funding and discouraging foreign students/researchers from coming to the US means that there will be depleted US capability just when China finds its groove.
Bravo.
Does this mean that Huawei phone which has been hurt badly by sanction will now stand a fighting chance because of homegrown GPU?
How good or bad these GPU compares to the SOTA GPU in the west?
And does this mean that Huawei has the ability to crank out the GPU commercially?
I've been using phones without Google Play for years.
My point is that YMMV based on where you are.
oh, man, "stand a fighting chance"? huawei phone sales has already been back and surpassed apple in china.
https://www.counterpointresearch.com/insight/china-smartphon...
Anyway, they would need to duplicate ASML first which will probably not happen in the foreseeable future.
And looking at how the recent wars and skirmishes in Ukraine, Israel and Pakistan/India have gone I think western military superiority is no longer real. In a conventional war of today US will lose and most likely nuke the other country so I think its best for the world if China gets so powerful that USA accepts that it no longer is the sole super power and we can avoid a nuclear war. As that is where we are heading either a multi polar world or a nuclear holocaust.
Can you elaborate on why this will not happen in the foreseeable future?
Because in my version of the foreseeable future I see it happening quite readily.
EUV is not the magic that everyone believes it to be. It can be replicated by us, (the US), at our convenience. It can even be replicated by the Chinese and Japanese. (Personally, I'd throw the Koreans and the Russians into that pool as well.)
But that ain't even the point. The point is that, in that particular game, there is always more than one way to skin a cat so to speak. It's not at all a foregone conclusion that EUV is the best way to skin a cat, and it'd certainly be a bad bet to assume it is the only way to skin a cat. Those are the questions I would hope that we, (the US), are focusing research efforts on, and we should assume that China is also focusing research efforts in that direction.
PS - Please no one bring up Trump gutting research. Here I'm only speaking of clear strategic research priorities in an ideal, (ie - collaborative), political environment. Obviously, politically erected structural concerns impact the viability of any research strategy we want to implement. I'm just talking about what I think would be ideal.
Here is the GIT https://github.com/Tencent-Hunyuan/Hunyuan-A13B
I always thought sceptical of the US sanctions, but that they backfire so fast is insane.
Out of China's perspective it might make sense to take out the wests AI capabilities soon.
Guess what a telco equipment company is good at :p
https://community.cadence.com/cadence_blogs_8/b/breakfast-by...
It has nothing to do with the size of anything.
Despite all the anti-EU propaganda in these parts, the entire AI stack is powered by us.
Things like ASML are the tattered remnants of the EU being a powerful force in this market. China might not be able to replicate ASML, we'll see. But if the EU can compete meaningfully with the Asians at semiconductor manufacturing that'd be a shocking development.
All modern semiconductors need ASML machines.
All modern AI is based on research by Sepp Hochreitner.
The www is based on work done by CERN.
Most factories are run on Siemens automation systems.
We've moved on from lower-level concerns like building chips to a higher level, directionally controlling the way economies and societies develop.
Economic growth is an enabling factor for a higher level metric, quality of live. It succeeds spectacularly there.
You've got a bunch of poor people starving and set up a factory to produce a couple of bombs and pull some farmers off the land to have them march up and down the parade grounds. That isn't going to put more food on the table. Sure it might be a good idea regardless because being invaded sucks, but it isn't wealth creation without a very creative understanding of wealth. A society can't be better off without creating new wealth.
I suppose there could be some sort of plowshares-to-swords program where they build tractors and bulldozers that can be repurposed into military gear on short notice. But I havn't heard of that working anywhere before and my read of modern militarys is that the gear gets pretty specialised to military use. And the soldiers need to actually focus on soldiering rather than being part timers.
At some level a country needs an army because without one they don't get to have an economy, but it is a dead weight in terms of prosperity. Military expenditure comes directly out of what people would otherwise use to improve their own lifestyles.
If the only thing you needed to make cutting edge chips were some machines from ASML every country in the world would be making their own chips right now.
China built their own.
I've gotten to the point where I no longer even listen to people who underestimate the Chinese. Whatever points they're making can be safely ignored. Indeed, strategic sense demands we ignore them. The time for underestimating the Chinese is long past. That's not the reality we live in any longer.
Now it is more like throwing pebbles at someone who is stronger while hoping he doesn't come over to punch back.
I've seen this argument before with some pro-China voices and it's the same circular reasoning in arguments like Roko's Basilisk. It is is futile and unwise to oppose China's ascent at the risk of incurring their future wrath when said rise is inveitable, better that we pledge loyalty now for future favours". It's nothing more than intimidation.
The difference is that future of China's hegemony is far from inevitable. They are strong, but they're not stronger just yet. Their innovation hasn't really broken through in red oceans when paired against competive incumbents, and they are just as prone to hype bubbles as we are. And their massive investment is coming at the cost of a resulting maligned consumption balance and destructive price wars. So there still moves to be made that can radically alter the trajectory of events.
Which argument is right really depends on a realistic evaluation of the situation.
To me I just have a bad gut feeling when we (the west) force China to oppose us.
I also had a bad gut feeling when we outsourced our technology there. We let the buy our companies while China only allowed joint ventures with state controlled companies.
Why didn't we pick up the fight 10 years earlier if they are so evil that we cannot let them win?
If I would have to bet my money I would bet it on China.
I live in a country which has experienced some hard and soft embargoes over the years, and let's look what has it done.
- We wanted to buy drones, and denied. Now we are one of the biggest drone manufacturers in the world.
- We denied air defense systems. We are developing a whole arsenal of missiles and rockets now, incl. standoff/cruise missiles.
- We denied planes. Our 4.5th generation fighter program got a great speed boost.
- We denied advanced naval technology. We built stealth ships, fast coastguard boats and all navigational systems which goes inside them.
- We denied optical pods for drones and aerial vehicles. We built our own in 6 months.
etc. etc...
Sanctions and embargoes are the biggest catalyst for a country to advance their tech at tremendous velocity.
That highly depends on the size and natural resource available to the country.
The Ford Motor Company's dealings in Nazi Germany and the Soviet Union significantly helped the adoption of mass production in those countries.
Typically it actually looks like the opposite. When I look at, eg, North Korea or Iran - if I were going to try and make them wealthy it is mostly internal policies that are the problem and not external ones.
If North Korea set itself up with single digit % company and income tax combined with a strong rule of law, local education programs and a liberal economy it would barely matter what sanctions were imposed on them. A tide of money would flow in and they'd eventually be wealthy under their own power anyway if not. Although it isn't obvious why anyone would sanction a small well run country; there is a correlation between sanctions and incompetent governance.
- firstly, media attack ("they're dictatorship!", "they're genociding someone", etc), preparing population for stricter measures
- secondly, color revolution, which, if successful, puts puppet malleable government in power, and makes the country ultra-poor (most of the countries, where color revolutions staged by US/West succeeded, became significantly poorer)
- if color revolution didn't work, there's always an option to just bomb the country, because it's always ignored if all the international treaties and laws are ignored if USA or Israel bomb any other nation.
If so, do you think it makes a difference that Turkey is a NATO member, and on (relatively) good terms with the Western powers?
For all the ideological differences and geopolitcal nervousness I don't think the US or EU see themselves as potentially fighting against Turkey, and so they don't feel the need to go to the trouble of strict sanctions or sabotaging local tech.
In recent years the relationship between Turkey and Western countries has been OK-ish (though far from stellar, see the S-400-related tensions or the French-Trukish tension in the Mediterranean).
But if you look at it on a longer perspective, the relationship used to be very tense, first there was the Cyprus crisis leading to pretty harsh western sanctions on military equipment, and then the cold war between Turkey and Greece in the Aegean see, with occasional real fire air combat and casualties.
Turkey’s defense advances took decades and came with major setbacks
Not to mention many NATO incompatibilities
Just look at all the other sanctioned countries
And if you want to look at other sanctioned countries, just look at how NK or Iran's industry fares compared to their economic peers …
It took me ~10 seconds to figure out the guy is Turkish and talking about Turkish defence industry.
Other countries like Iran however, do develop their own drones because of sanctions
https://www.theatlantic.com/international/archive/2018/08/wh...
Huawei was going to work on GPUs anyway, SMIC was going to fab chips anyway. How much of the total GPU compute is the result of increased investment after sanctions, and how much was already planned? And how does it compare to the alternative of importing Nvidia GPUs for the same amount of money?
Unless Huawei is getting better performance per dollar than Nvidia, this is them implementing a costly workaround, which is the point of sanctions: increasing cost.
With sanctions they get free money to develop better chips.
== sanctions backfire
Also the money the spend on top is mostly domestic. They can build some more powerplants to run less efficient chips. They can pay their workers and their developers to build it.
Making it cost more is keeping them from being as competitive — which is the point of sanctions.
Huawei already designed and developed its own chips before sanctions, and SMIC is developing their next node before sanctions.
Whether sanctions work or not is another matter
Sure, but China has started dumping a lot of money into competing with ASML as a result of the sanctions. And no, they are not going the super complicated route of firing lasers at drops of tin to get EUV. They are trying to sidestep that costly complexity, and if they are successful they'll be providing equipment to the top chip producers (around the world or just their own).
It's funny how it's impossible to find English coverage of this book :')
The western media has a narrative with many real evidence to support the narrative so most western people take it for granted. But the reality is more complicated. For many engineer money is an important factor for sure but maybe not all.
It is called hormesis.
*) They're NPUs not GPUs, since they can't render graphics
https://www.bloomberg.com/news/articles/2025-05-13/us-warns-...
The A.I. war is intense, and you can clearly see who is s** their pants here.
Huawei GPUs are illegal worldwide now, according to the US government.
Like, if you're so good at something why are you scared of the unproven newcomer beating you?
(a) alone sounds super negative to ordinary people, but with (b), justified by the existence and achievement of Huawei & Deepseek, the AI cause now sounds a legit jihad silicon valley is carrying.
The semiconductor industry is always a key industry for China. They laid it out in Made in China 2025.
Huawei designed and developed its own chip before any sanctions, and the Chinese government never stopped throwing money into the semiconductor industry.
In the short term, money can go to Nvidia, but it won't be long before China creates its own "Nvidia" like BYD.
The sad part is America elected Trump, and he's gutting American research and cutting out the CHIPS Act.
One country is going for the long term while another country is short sighted