HuggingFace Text Generation License No Longer Open-Source
github.com
github.com
People make mistakes when choosing a license and should be OK if they course-correct fairly quickly.
Regardless, if the library is worth anything (I am not familiar), I would suspect the pre 1.0 version to be forked and sucked up by AWS/Azure/etc similar to ElasticSearch.
If they wanted to profit from their ideas, they could have started with selling the product from the start.
No, no and no. I can want to publish the project to the community but at the same time want to restrict corporations from using it. It goes against the OSI open-source definition but.... we should maybe think about how much that definition is fit for purpose then.
This is like saying "if you don't want someone to steal your work, just don't publish it"
Personally, I will continue to release any code I open source as Apache 2, and if I intend to sell it, code will not be open source. You are free to think I am "wrong", but fortunately we need not agree with each other.
AGPL is not a full solution to this because the SaaS loophole.
Huggingface is entirely in the right here - there is zero reason for them to effectively subsidise their competitors who will obviously not contribute back in any meaningful way, but at the same time, want to publish the source, either for altruistic reasons or for publicity reasons.
"SaaS loophole" is the GPL. AGPL is the thing that closes it.
People using free software for free is not a loophole.
The technicalities of actually using AGPLv3 is... well, if your only goal is to keep large businesses from touching your code, sure. But you can also do that with the weird funky postmodern licenses that technically confer no rights and could easily be used to set up copyleft trolling. If you just want rules that are fair and easy to comply with, this license has... problems. To make a long story short, AGPLv3 only makes sense for web development in interpreted languages that make it easy to list out the code of the site. In literally any other context[0] you're creating traps to sue people with.
MongoDB wanted to extend AGPLv3 even further with a requirement to release all supporting software - e.g. management UIs and things like that. OSI wouldn't put their blessing on it. Amazon just reimplemented the MongoDB API rather than complying with SSPL.
No, I don't think we should be entertaining the notion of API copyright. Hell, FOSS Patents thought Oracle was in the right purely because "oh it'd let Oracle assert GPL", even though it would completely pull the foundation out of the commons.
If you want to be even stricter, there's also the OpenWatcom license, which trips the copyleft whenever you use the software. So no private forking. This is actually less of a compliance headache than AGPLv3 - you don't need to allow source code download over the network, you just have to publish your modifications somewhere. The FSF refuses to touch it because they consider private forking to be a human right.
The problem with stopping "the bourgeoise" from hoovering up all the innovation from the commons is that doing so takes the software out of the commons. Its worse than AGPLv3, SSPL, or OpenWatcom. Anyone who is trying to do this is not trying to protect the commons, they're trying to join the bourgeoise. Because saying that you're not allowed to host the software for others, period, is the language of the proprietary world. Proprietary software licensing is so lucrative specifically because of usage restrictions - it lets you look into each user's wallet and extract the maximum amount of money from it.
[0] Hector Martin found AGPLV3 on the source code for an Ethernet PHY - how the hell do you put a "view source" link on an Ethernet packet?
He's also pointed out that if you just link to a Git repo as your compliance, that's not good enough, because you need to have the specific source code currently executing be accessible. You can't point to a commit inside of itself[1] The original idea was that AGPLv3 source would have quines in them that listed all the source code, and then AGPLv3 would prohibit removing the quines, but you really can only do that in interpreted languages (e.g. PHP, Python, or Ruby).
[1] Cryptographic hash algorithms would be no good if you could do this
If that bill goes through. This argument is nullified at the global scale. And it behooves open source authors to create licensing agreements Commercially.
Open Source simply means Open Source. Everything after the fact is up to the license and bad actors will always ignore such.
(then this discussion wouldn't exist)
And so the only options are for HuggingFace to eat the cost of R&D, possibly to improve a direct competitor, or to limit commercial use in this way.
And nothing stops you from selling them a separate license, in exchange for money.
- X presumably being someone without any potential profitability path
- Y someone who can make money from it?
I can get the rationale of not wanting Amazon and Microsoft to profit from you work, but I am unsure how you thread this licensing needle without a GPL-similar license from the start. Either you put it out in the open without restrictions (and eat the potential consequences), encumber it in such a way that no corporation wants to touch it, or you never share it.
Okay, so then irrational people get to shoot themselves in the foot by not using quality free software. What problem is? Presumably those people will be out-competed by ones who make more sensible decisions, or at least make less money.
> And so saying something like "you can sell a separate license" doesn't change that dynamic.
But it does though? If you don't want to use software licensed under the GPL then you can go pay them money to get a different license. Which they like because they get your money. Getting money from irrational people is one of the most profitable things in the world.
And if you're neither paying them nor contributing back changes in the spirit of the GPL, why should they care if you don't use it?
You can still:
+ use it for personal use
+ use it as part of a commercial project
+ sell a hosted service of <v1.0
You cannot:
+ wrap an API around the library (v1.0+) and sell that, without a license from HF.
This is less restrictive in practice than some of the extreme Open Source copyleft licenses. It’s fine.
[1]: https://en.wikipedia.org/wiki/The_Open_Source_Definition
This reminds me of that language model coming out of the PRC about a year ago that claimed to be open, yet it turned out that it was only the code that was open and not the model itself. Which is fine (your labour on your terms), but use a different term in that case as I can assure you most people have an interest in code and weights, not just the former.
Actually it's now totally OK to loooove megacorps and their lovely and caring ways to do things, after all it's totally a GOOD thing only a few select megacorps can do cutting edge training otherwise some closet nazi may build a hitler bot in his basent!
Exploitation of labour was, very literally, the main argument ESR put forward in favour of the Bazaar model -- and it is the reason Open Source as a term exists.
I am also free to call it open source.
‘The question is,’ said Alice, ‘whether you can make words mean different things–that’s all.’
‘The question is,’ said Humpty Dumpty, ‘which is to be master–that’s all’"
-- Humpty Dumpty to Alice, _Through the Looking Glass_
See https://www.arp242.net/open-source.html for an actual historical definition of open source, free of corporate propaganda.
---
My take is (I think) pragmatic. I know for a fact that both OSI and FSF had a desire to delineate terms for their own agendas (both of which have been stated in the open). Both worked hard over decades to explore the implications of their definitions and we as a society also worked with them to form some kind of consensus of what these terms mean. Heck, it is not like Free Software is not problematic as a term due to the ambiguity (and prior usage…) of the word free and FSF has discussed this at great length. Likewise, Open Source is of course flawed as a term in its own way, but it is not useful to start muddying the waters around the term at this point as there is a consensus in place that allows us to have a discussion about Open Source and its flaws to begin with.
The Ethical Source movement for example object to both Open Source and Free Software licenses, while still adhering to freely sharing code. Great! I happen to disagree with them, but I think their reasoning is fine and they decided to invent their own (obviously faulty in some ways) term to delineate themselves from those they disagree with. Similarly, I wish this anti-corporate, source-available, whatever proto-movement that keeps popping up would get together, write a manifesto, and come up with a name rather than dive into discussions such as this and try to somehow argue that when the great majority of us used Open Source as a term for the last twenty or so years what we actually had in mind included licenses which restricted fields of use and discriminated against persons and/or groups. Sorry, I do not buy that, not even for a second.
As something mildly tangential. I find it confusing that members of this proto-movement seems to both support venture capital-fueled start-ups such as Hugging Face and at the same time frequently uses Marxist-esque terms. Heck, I have also seen attacks on Open Source authors from people using alt-right terms. Maybe there is more than one new movement out there? Regardless, I look forward to seeing what term they end up using and reading their reasoning and definition when they finally stop arguing semantics.
In light of this, maybe you should consider the possibility that you're wrong and you're backing the wrong team.
No, because as I pointed out there are more than two camps and there are some of us that find even what OSI approves too restrictive. Just like OSI does not own the culture of open and free software, neither does FSF. Thus you can not claim that OSI watered down and corrupted the “true” spirit of free and open software. What they did was to formalise and create (some say appropriate) a term to serve their purposes and what they formalised inarguably captures a subset of what was and is open and free software. Regardless of whether we agree with their agenda or not, I do think it is fair to let them have the term they defined as it makes it easier to have a discourse around software licensing without all of the sudden including source-available licenses and the likes.