Hugging Face and AWS partner to make AI more accessible | Hacker News Reader

Hugging Face and AWS partner to make AI more accessible

huggingface.co

102 pointsrjtavares41 comments

My startup computes everything locally. We use our own hardware.

If we'll ever get to the point where we need serious dynamic scaling, sure, we'll use cloud-tooling. But for now, cloud-compute has zero alignment with our vision and strategy. AWS is powerful and surely has its place, but "run it on Bezos' hardware" doesn't exactly spark confidence in "open" or "democratized" AI progress.

One thing I love about Huggingface is that I can simply download a model and do stuff with it. Iirc, I couldn't download some models and was instead incentivized to use Sage Maker (maybe only for finetuning? I don't recall exactly) recently.

I deeply hope that this partnership doesn't become a "to use models hosted by huggingface, you must use AWS". The fact that this announcement reeks of business speak ("Together, the two leaders aim to accelerate [...]") leaves a bad feeling in my mouth.

Disclaimer: I used to work for AWS ML

Buried in the press release is the announcement that there will be a push to bring up models on the AWS Trainium accelerator.

In the name of "ML democratization" this agreement could help shift tides away from the Nvidia HW + SW monopoly. Amazon (trn1) and Google (TPU) are little guys compared to Nvidia.

The book may already be written. Pytorch is heavily biased towards libraries that are customized for cuda/Nvidia (see FSDP). This leaves the other HW folks playing SW catch-up, with everyone rallying behind the open source XLA compiler.

It's too bad the press release didn't touch on these points and used cliche rhetoric instead.

I don't see any mention of what Amazon are actually doing for this partnership. other than providing the same services they provide for everyone.

This reads like "Amazon gave us a bunch of money for us to use and promote AWS."

Unless Google or MSFR buying huggingface upfront, this is going to eventually make sure AWS be the fist class citizen for future huggingface initiatives

Including all open source models

and that's almost certainly what's happened, lol.

This is correct. All of a sudden, we'll see "Amazon state-of-the-art Natural Understanding systems paving the way [blah blah blah]"... Umm... No. It's HuggingFace. And just like that, HuggingFace fades away; its well-earned recognition reassigned to Amazon, by Amazon. It's not a possibility; it's a certainty.

Shout out to Amazon though. People keep falling for the "partnership promise." The wax must be hella poetic.

That's unfortunate. I liked hugging face. Could have been the next github for ML.

Hope the push to the AWS walled garden doesn't eat them alive.

I don't really understand how this announcement will prevent them from becoming the next github for ML.

This is about making it easier to run huggingface models in AWS.

That's how it starts, isn't it?

>becoming the next github for ML.

Considering github digested all the open source code into a proprietary AI...

One AI to rule them all

(sorry)

Not an acquisition, just a product

... all the way to slaughter.

Same. Pretty bummed. I like HuggingFace, but there's no such thing as "partnership" when Amazon's involved.

Amazon has a clear reputation of "brand killer" (e.g., Zappos, IMDb, 6pm, WholeFoods, ComiXology, GoodReads, Roomba, Ring, Eufy, Woot, Sleep Jar, Fabric.com, Shopbop, Graphiq). Very much "if you can't beat'em, eat'em," although I sometimes wonder if it's negligence more than strategy. After all, their own services don't even work half the time (e.g., Prime Video and Music, Alexa), and its retail site is overrun with fake goods with fake reviews.

Good luck, HuggingFace. You'll need it, along with an active team of lawyers and a backup repo in Azure.

Edit: Adding additional sentiment below.

I desperately wish what I said above wasn't true. I am in no way faulting employees who work(-ed) on these projects with best of intentions. Persons who have endured understand, the problem is above, not necessarily below. Again I will strongly encourage HuggingFace to have backups and lawyers. Please don't give everything away; there is no good faith at present.

I believe the examples you’re highlighting were acquisitions, not partnerships. This seems more along the lines of https://stripe.com/newsroom/news/amazon-and-stripe

I will say this. Many acquisitions begin as partnerships.

Shall we begin to unearth the "partnership" with Stripe, and its impact to customers? Perhaps a necessary discussion worth continuing elsewhere, e.g., https://news.ycombinator.com/item?id=34863530...

I worked at Whole Foods (before and after the acquisition), and I have been a Prime Video customer for as long as I can remember.

I remain a happy customer of both. Whole Foods is still expensive, and there are some things I want to buy that they still don't sell. And so I tend to do the majority of my shopping at HEB instead. But I acknowledge that Whole Foods is still objectively better in this space -- in virtually all regards other than price.

Prime Video works well for me, and always has. Almost as good as AppleTV. My wife and I probably watch more stuff on Prime Video than all the other services combined.

I haven't used any of the other services you mentioned, so I can't speak for them. But these two have always worked very well for me, and so far as I can tell, will continue to do so for a long time.

What will happen with the HuggingFace partnership, I have no idea. But I do look forward to watching this partnership play out and what each can do for the other.

There's nothing to indicate this means they will be exclusive to Amazon. Only that they'll add some integrations or even just pricing consideration for AWS customers.

Funny how both OpenAI and now Hugging Face promise “openness” for their AI tools so that they’re “accessible to everyone” and then both partner with big tech companies who are trying to control and monetize as much as they can.

Huggingface started off as GitHub for machine learning and anyone can download the models there. OpenAI has everything behind a black box where you don’t exactly know what actually happening.

Then to monetize they host the models you want to use at a cost which would be a good investment for Amazon but you can still self host using their libraries.

Hugging Face internally has a lot of very strong believers (including the exec team) in open source + decentralized approach towards AI as the only safe and productive way forward. HF is ostensibly the only place to find deep+broad resources for open source DL and transformers/diffusers are the libraries du jour of open DL.

Compute is really important for AI, and having a cloud provider align themselves with an organization which is genuinely trying to be "Open" AI is I think a positive step forward.

Leadership is fungible,everyone has a price, and words are innevitably wind.

I would consider this a solid step in the direction agaisnt why huggingface is actually useful.

Not disagreeing with your first point, but as I see it - this only enables HF to continue to spend money on making good open source tooling + doing research on open alternatives to closed source LLMs (something sorely needed), while opening them up to more enterprise customers who primarily use AWS.

FWIW I am an ML engineer there (maybe should have disclosed earlier) and I feel pretty optimistic about the opportunities this will enable for the open source community. Maybe with the visibility into the closed door discussions I have a more positive attitude, or maybe I'm being naive.

Time will tell!

>Not disagreeing with your first point, but as I see it - this only enables HF to continue to spend money on making good open source tooling + doing research on open alternatives to closed source LLMs (something sorely needed), while opening them up to more enterprise customers who primarily use AWS.

This is the response/ justification every time. The issue is that it never goes that way.

>Time will tell!

It always does. What its shown me is that one step towards the slippery slope is enough to abandon hope for the project. Every good and useful open source project I've seen go this way inevitably turns their back on their customer and long outlives their usefulness.

There's a world of difference between OpenAI and Hugging Face. OpenAI simply has the word "open" in their name, while there is nothing open about them. Hugging Face on the other hand deserves a ton of credit and praise in that regard. It's perfectly fine for a company of that nature to partner with a big player(whether that be Microsoft or AWS or Google or whatever): infrastructure is not free, nor is it cheap. Of course AWS and Microsoft have a lot to gain from this, which is also fine for that exact reason. So long as the community continues to exist and have the ability to use, extend and participate, I don't see anything wrong.

Moreover, Hugging Face has made it stupid simple for anyone to use. So Microsoft, AWS or Google, all I can say is bravo, Hugging Face and thanks for your work.

The world is at war- Amazon burning compute resources on hugging face to fight MSFT.

Same way as US doing with missiles, if you know what I mean.

Jokes aside, hugging face is the real open AI in the space and desperately needs this.

does AWS provide a google colab alike environment, colab is now sign-in then using it, AWS you have to jump so many hoops to start using it, could be easier to access.

Check out AWS Sage maker studio Lab

And when you delete the Sagemaker domain, remember to also delete everything associated with it under the Lifecycle configuration... that was an expensive weekend

isn't sagemaker studio lab fully free?

it's said colab is easier to use, but sagemaker studio lab provides better hardware underneath.

I agree that Google Colab is so much nicer to use, but if you already use AWS, have accounts and access control already setup, the SageMaker Studio is OK to use also.

It is usually easy to take existing Jupiter notebooks and run them on either platform, assuming data is easily accessible.

Thanks. Not aware of this until now. Just registered, seems cool.

Trying to compare it with colab on the freemium plan, I guess they must be close to each other(how long you can run a sesssion,how much cpu/gpu resource you can use for free,etc)

Lame. Everyone keep getting sold to the same shitpile at the end of the day.

Well, with the current climate it's not far fetched to try to milk something out of it and cash out before the ship starts sinking.

Ideally I'd like to see more work on decentralized AI/DL methods. Maybe that's out there (I should dig into it more than I have) but either way I think it needs more attention.

Next headline: HugingFace agrees to be acquired by Amazon.

Did I miss a memo somewhere that said "every single AI related announcement must pretend as if you are bringing peace to the world and solving all forms of inequality forever"? I mean come on, "democratize machine learning"? Even the langauge of "accessible" which is constantly used by AI people over and over is just absurd.

It feels like whoever wrote this mashed together Social Justice Words until they got something intelligable. Example:

>“The future of AI is here, but it’s not evenly distributed,” said Clement Delangue, CEO of Hugging Face. “Accessibility and transparency are the keys to sharing progress and creating tools to use these new capabilities wisely and responsibly. Amazon SageMaker and AWS-designed chips will enable our team and the larger machine learning community to convert the latest research into openly reproducible models that anyone can build on.”

I agree completely with this sentiment. I don’t want to create great new works of literature, I just wanna label some text with metadata and learn the bare minimum about some random tech I’m supposed to use next quarter.

I predict the next year will be companies promising AI will fix everything wrong about my life, and approximately zero will deliver. The companies who will do well will be the companies who don’t even advertise their AI use, but incorporate it into their existing value prop.

I get what you're saying but I'd put hugging face as an exception. When I read democratizing access, I see a company making models freely available to the masses who do not necessarily have the resources to train them themselves. Is it solving global poverty or hunger? No of course not. Is it helping to tear down the ivory tower of academia or fill in the moats surrounding megacorps? Yes.

But now they're just another cog in a megacorp..

> I mean come on, "democratize machine learning"?

That slogan made more sense back when Hugging Face first launched in 2019 and training models was prohibitively expensive and required sorting through poor model training codebases, but nowadays it's less applicable. (cost/efficiency of GPUs is much better and lots of higher level libraries, including Hugging Face's transformers, reduces a lot of the pain)

Huggingface has done more to democratize ML than any other entity in our field except maybe the folks making sklearn and other high profile tools. They get to use this kind of language, stop critisizing them for it as though they aren't really the only bastion of openness we have. They really are the only one's doing what they do.

Reply on news.ycombinator.com