7 ms·
Grim prediction: "Quantized models are no longer permitted on the Hugging Face platform. Reduced precision models are a safety hazard and a violation of our TO
by randusername 20d ago
Grim prediction:
"Quantized models are no longer permitted on the Hugging Face platform. Reduced precision models are a safety hazard and a violation of our TOS. Click here to speak with a sales representative about our many exciting cloud hosting offerings or enterprise GPU packages."
- shadowgovt 20d agoYeah. I want to be optimistic, but this feels a lot like how there used to be multiple video hosting sites and now there's just YouTube (and its copyright filters).
- dd8601fn 20d agoNow I’m going to get a little nostalgic about Vimeo. I feel like they really tried to be a better, more civilized video platform.
- JV00 20d agoIn that case, somebody will open a new platform to upload the quantized models and hf will lose market share
- dd8601fn 20d agoOne would assume. Just out of curiosity, how did Hugging Face have the money to run that operation?
- peezd 20d agoVC funding (much of it from big tech). They have a freemium model and offer some pay for enterprise features, really just private repos, permissioning, etc. but not sure if it is meaningful
- imhoguy 20d agoWhy platform? Bitorrent is the best for all these read-only model files laying on SSDs. Agreed some tiny DHT model search engine would be nice.
- vonneumannstan 20d agoJevons Paradox says no. Even more "efficient" models will increase overall GPU demand and be good for NVIDIA.
- kevinthedigger 20d agoSounds like there’d be space to build another company.
- cyanydeez 20d agoProbably start with the abliterated ones first. But selling more product is still on thr table until cloud AI starts the real cash drain, then therell be google like shenanigans.
- armcat 20d agoDon't see this happening. Both Nemo and TensorRT from Nvidia are heavily invested in low precision formats and higher inference throughput as well as low memory use. It's more about what happens in relation to GGUF, MLX and non CUDA-native frameworks/formats.
- belzebub 20d agoBingo.
- Tenoke 20d agoToo base cynicism. I'd be willing to bet you my $200 to your $100 that doesn't happen.
- tclancy 20d agoNatural log cynicism and any cynicism about public corporations in our late-stage capitalism world is likely to look tame a year later. Or possibly too.
- randusername 20d agoYou're probably right. It was mostly hyperbole, but the older I get the more I think hyperbole does eventually come to pass in situations like this. It's gradual, though, not all at once.
- digitaltrees 20d agoThe history of acquisitions mothballing the acquired assets or moving them off their original mission or letting them atrophy is too stark to support your optimism. Especially when nvidia has a direct incentive to to steer the ecosystem. For example are they going to highlight and surface alternative chip designs like grok? Or will model search mysteriously not find related models?
- flmontpetit 20d agoNvidia doesn't exactly have Oracle's reputation when it comes to acquiring companies, but given its overall hostility towards competition and open source I can't blame people for being anxious about the news. The most important resource for self-hosting LLMs is now under the control of a company that has very markedly kept the specialized hardware outside of the broader public's hands.
- willmadden 20d agoUnlikely, but say goodbye to uncensored and abliterated models. These acquisitions are not healthy. They stifle competition and make it easier for governments to censor open weight models by choking off distribution points.
- ALLTaken 20d agoWill just help EU/ASIA to fill that gap
- mcv 20d agoGoodbye? Or maybe hello to a new platform.
- willmadden 20d agoHello to the new app store and "Nvidia model approval process".
- vkaku 20d agoWe really need an artifactory for GGUFs
- nunez 20d agoJFrog is trying to create something like that iirc
- echelon 20d agoNvidia's LLM hyperscaler partners are actively developing custom, non-Nvidia chips. Nvidia has a deep interest in open source models to counter the LLM hyperscalers. Nvidia also has a big bet on robotics and local AI, which requires model efficiency. I'd worry far more about Dario and Sam regulatory capturing the US market via a "model safety" scare.
- afpx 20d agoThis acquisition and the Stripe acquisition of OpenRouter are puzzling. Both HF and OR, from a technological standpoint, don't appear to have much IP that cannot be easily replicated. I wonder how much it's worth to Nvidia just to obtain a list of who's downloading which models and for what purpose. Similarly with the Stripe acquisition of OpenRouter. In the past, it would have been valuable to obtain an audience or market share or even a productive team. However, in this new age of the surveillance state ... telemetry, orchestration, and financial
- JoeOfTexas 20d agoBrand name is hard to beat. HuggingFace is the defacto leader in open models.
- PaulHoule 20d ago... and there is some of that two-sided market effect that (1) people who want to publish a model will publish on HuggingFace and (2) people who look for models will look at HuggingFace. Even if HuggingFace is poorly run in the future it will be hard to dislodge.
- velominati 20d agoIP is easy. Tractionis hard.
- LastTrain 20d agoI've just about stopped trying to think about things in terms of fundamentals.
- DebtDeflation 20d agoNvidia is hedging their bets against becoming too dependent on OpenAI, Anthropic, and the hyperscalers as their only customers. If on-prem deployments of open models starts taking off in corporate America, this acquisition positions them well.
- Insanity 20d ago
- colingauvin 20d agoNvidia literally releases quantization tools.
- logancbrown 20d agoAlso very silly prediction given this would dramatically lower the amount of models hosted on HuggingFace and move users elsewhere to platforms that do support quantized models (not what Nvidia wants).
- cpburns2009 20d agoThat would be if Anthropic purchased Hugging Face, not Nvidia.
- PhunkyPhil 20d agoThe new era of thepiratebay/silk road but for LLMs will be an exciting time In all seriousness, isn't it pretty straightforward to quantize a model locally? I can do it using mlx in one command.
- HarHarVeryFunny 20d agoI doubt it. I see this as NVidia placing 50% of their bets on local models as the future, which it seems may be more profitable to them than cloud (sell 100/whatever local cards, which see 40hr/week max utilization, vs one cloud card seeing 24x7 utilization). NVidia seem like a smart company - if they want to promote this direction they are not going to cripple it by trying to limit quantized models, even though you might expect them to promote benchmarks showing the benefits of larger and less quantized ones. It's like the smart Intel of old - support CPUs all all price points, while also promoting CPU hogging applications like OpenCV.
- m3kw9 20d agoAdvertisment "HuggingLegs", it's hugging face for hugging face.
- ericd 20d agoI really, really doubt that, and I'm a bit surprised (not too surprised, HN has gotten really cynical) that this is the top comment. I think the future nvidia most doesn't want is a small number of closed labs that run away with it, and gain power to steer a large part of the hardware spending towards nvidia's competitors, as leverage in pricing negotiations for big hardware buys. Their ideal would seem to be lots of competition in the model space, driving lots of innovation in lots of areas, at low margins for the model makers, driving use and demand for hardware way up. They want the portion of the economy working on this to go up, and that's not going to happen if there's just a few companies that can afford to build models.
- hector_vasquez 20d agoEx-NVIDIA employee here who worked on inference software. When it comes to inference software and models, NVIDIA's strategy is advancing the state of the art in the open model realm. They are building software and services business for those markets that need to be competitive on their own merits, and quantization and precision tuning are techniques they use themselves constantly.