| ▲ | Hetzner is working on LLM Inference(sliplane.io) |
| 127 points by jonas_scholz 6 hours ago | 53 comments |
| |
|
| ▲ | swiftcoder 4 hours ago | parent | next [-] |
| It would certainly be interesting to have a highly respected EU-native inference provider, if only to make the regulatory gods happy |
| |
|
| ▲ | NetOpWibby 23 minutes ago | parent | prev | next [-] |
| This is interesting because I thought Hetzner was anti-crypto? LLMs aren't the same but they're often lumped in with crypto as "things no one wants." |
| |
| ▲ | satvikpendem 10 minutes ago | parent | next [-] | | Who lumps LLMs with crypto? The former has actual usage and purpose, the latter doesn't. Anyone who does must think any new technology is all the same. | |
| ▲ | cousinbryce 10 minutes ago | parent | prev [-] | | Crypto DAU: 10’s of k
LLMaaS DAU: 100’s of k |
|
|
| ▲ | ano-ther 5 hours ago | parent | prev | next [-] |
| Good to see more developments in this space. I quite like this service, which is a little further than Hetzner and has several models to choose from: https://www.infomaniak.com/en/hosting/ai-services |
| |
| ▲ | drcongo 3 hours ago | parent | next [-] | | Important note on Infomaniak's offering - when they first launched it, I gave it a try but despite claims that it's OpenAI API compatible, even basic things like sending a base64 encoded image were broken. I reached out to their support, who replied with (verbatim): "Unfortunately, we do not provide support for our AI Service, as the solution is highly unmanaged and uses our API." "highly unmanaged" didn't fill me with confidence, and "uses our API" is a very weird reason to give for not offering support. I wrote off ever using them for anything beyond email after that reply. | |
| ▲ | jonas_scholz 5 hours ago | parent | prev | next [-] | | Interesting, didn’t know about them! Weird model selection though, no glm or deepseek? | |
| ▲ | archerx 5 hours ago | parent | prev [-] | | Infomaniak is such a shitty company, I had to use them on a previous job I worked at and dealing with them was awful. | | |
| ▲ | sureglymop 2 hours ago | parent | next [-] | | I highly agree. I once registered a domain through them that happened to be my last name but coincidentally also the name of a company. They restricted me from using the domain and never refunded me the yearly fee for it (citing trademark reasons, although under Swiss trademark law, someones legal name is not a trademark violation, especially as it was just my personal website and I wasn't working on a competing product/service). Only weeks later someone else successfully registered it without such issues. Worse even, in my account panel I still had control over the e-mail service of the domain, so there was most definitely a pretty bad security critical bug there. | |
| ▲ | MASNeo 4 hours ago | parent | prev [-] | | That makes it sound like dealing with Hetzner is easy. Is the case? | | |
| ▲ | sam_lowry_ 3 hours ago | parent | next [-] | | Depends on your mindset. If you are an Engineer that does not need hand-holding and you accept terse answers from support with the gratitude to the human on the other side, then go for it. You will at least not need months of training and tough exams to be an expert in Hetzner Cloud. It is simple, but that's the point. | | |
| ▲ | jonas_scholz 3 hours ago | parent [-] | | I mostly use Hetzner baremetal servers and not cloud, but the support there is in my experience very competent, but also VERY german and direct. I can imagine if youre neither german or super technical that this can be intimidating or considered unfriendly | | |
| ▲ | drcongo 3 hours ago | parent [-] | | Being slightly on the spectrum, I find Hetzner's "VERY german and direct" support perfect. |
|
| |
| ▲ | Saris 3 hours ago | parent | prev [-] | | I only have a couple of VPS plans with them, but my experience is good. |
|
|
|
|
| ▲ | danlitt 12 minutes ago | parent | prev | next [-] |
| Excited to see the price of every other product they offer triple for no reason. |
|
| ▲ | mark_l_watson 4 hours ago | parent | prev | next [-] |
| This seems like a smart move, given their ability to host efficiently. I approve of efforts to make the cost of inference for smaller useful models slowly approach 'close to zero' and there are many good paths for getting there. It is useful for companies to get fast hosting for the class of smaller models they may end up hosting in house. |
| |
| ▲ | jonas_scholz 4 hours ago | parent | next [-] | | I really hope they dont stop at the small models though! The bigger ones that dont fit on a single GPU are more interesting I think | |
| ▲ | mips_avatar an hour ago | parent | prev [-] | | Problem is right now the biggest GPU boxes they have is single rtx pro 6000s. |
|
|
| ▲ | perelin an hour ago | parent | prev | next [-] |
| Whats definitely missing: a solid (non Mistral) GDPR compliant coding plan / subscription. All offerings are either US or China based. With the newest open weights models this became really interesting imo. |
|
| ▲ | embedding-shape 5 hours ago | parent | prev | next [-] |
| > The enable_thinking option is worth mentioning. Without it, the model can spend a surprising amount of the completion budget reasoning before it returns a visible answer. Straight up the opposite, which the name makes abundantly clear, with the option it does reasoning, without it it doesn't... |
| |
| ▲ | jonas_scholz 4 hours ago | parent [-] | | My writing wasnt clear here I think, without the option it defaults to reasoning enabled. With "without it" I meant without enable_thinking=False! | | |
| ▲ | cyanydeez 4 hours ago | parent [-] | | llamacpp has a reasoning-budget and reasoning-message setting that can both be a global or per header setting. Using it, you can stop it's reasoning token count and insert a message at the point you stopped it. This allows both the client and server to customize it. I typically use a message that tells it to compact the conversation and use subagents. I find the reasoning gets bloated when it's failed to do whatever task it's doing and often times it either has too little context (subagents) or its context is bloated (compact). This works fairly well to get it to extend workable life up to ~1M on a local model. |
|
|
|
| ▲ | rebelde 3 hours ago | parent | prev | next [-] |
| Hetzner is very efficient hosting servers Will this be the new division of labor? Americans - best proprietary models Chinese - best open weight models Europeans - best / most efficient inference service |
| |
| ▲ | andsoitis 3 hours ago | parent | next [-] | | We need frontier open weight models from the West. Maybe Nvidia's Nemotron can get there. https://research.nvidia.com/labs/nemotron/Nemotron-3-Ultra/ | | |
| ▲ | 9dev 2 hours ago | parent | next [-] | | There is no "West" anymore, at least not in the sense it was used in the last 50–70 years. There is now Europe/Canada/Australia on the one side, and the USA as an adversary on the other, and that won't change in the foreseeable time. What we need is frontier open weight models in the EU, Canada, and Australia. | | |
| ▲ | slashdev 2 hours ago | parent [-] | | Give it two years more then it changes again. | | |
| ▲ | 9dev 27 minutes ago | parent [-] | | It won't. That's the point many people don't seem to get: The people who wanted everything that is happening now don't just go away. There is no guarantee they won't vote for a narcissistic grifter again, and thus any long-term decisions building on mutual trust and stability can't be made. How could you justify investing into a factory that relies on a 20 year ROI window when tariffs could be changing every day? Why would you want to use an AI model that could be restricted to US citizens or taken away entirely without notice? Big, slow wheels have turned everywhere around the world. They won't turn back so easily. |
|
| |
| ▲ | ForHackernews 3 hours ago | parent | prev [-] | | why? | | |
| ▲ | andsoitis 2 hours ago | parent [-] | | Reasons include, but are not limited to: a) broad competition is good b) jurisdiction diversification (you don't want to be dependent on the regulatory winds of a single jurisdiction) | | |
| ▲ | ForHackernews 2 hours ago | parent [-] | | ...but you're not dependent on anyone? If I'm running a Kimi model on my local cluster, the CCP can't shut it off. It's not like the Finnish government has veto power over Linux. | | |
| ▲ | andsoitis 2 hours ago | parent [-] | | You're not necessarily wrong for now, but that framing might be too narrow. It is not beyond imagination that a government can limit availability to future frontier versions of the open weight models produced within their jurisdiction. It pays to be a little paranoid and have a diversified portfolio of options to choose from. Anything that can have major geopolitical consequence is especially susceptible to such structural risk. This has nothing to do with any particular government/country/jurisdiction except insofar as said country is a leading power and hence has immense leverage across a very wide range of dimensions. Said more plainly, it is safe to assume that a country with immense power is unlikely to be willing to cede or distribute such power to others. If you're hung up on my choice of "the West", consider that I mean it geopolitically and economically, so broadly includes North America, Europe, Australia, New Zealand, Japan, and South Korea. I could have also said the Global South, but I think it is a fair statement that none of those countries have the means, but I'd be totally cool and happy if I'm proven wrong! |
|
|
|
| |
| ▲ | frangonf 3 hours ago | parent | prev [-] | | I read that data centers in Frankfurt use 40% of city energy, and seems that some fossil fuel based workarounds are being pushed. I hope we do better than the US in the energy and data center gold rush. |
|
|
| ▲ | Havoc 3 hours ago | parent | prev | next [-] |
| Interesting. I could see them perhaps coming in competitive for models that fit into single cards? Less so playing in the big model serving league...climbing into that esp right now would be madness |
| |
| ▲ | jonas_scholz 3 hours ago | parent [-] | | Why do you think this would be madness? It seems like, at least in the EU, there is barely competition for the big open-weight models? |
|
|
| ▲ | _pdp_ 37 minutes ago | parent | prev | next [-] |
| I mean yah... host glm and kimi and I am game. |
|
| ▲ | nubg 3 hours ago | parent | prev [-] |
| Potentially interesting article ruined by AI slop hallucinations like > For now, the API is fast, free, and fun to try. The next hardware announcement will tell us much more than another small model would. |
| |
| ▲ | jonas_scholz 3 hours ago | parent [-] | | what is the hallucination here? it is fast, free and fun to try. And I genuinely think that the hardware decision (if they get bigger gpus) decides if this will be a banger product or not? | | |
| ▲ | nubg 3 hours ago | parent [-] | | nobody writes like that. the article is based on a prompt, which i would much rather read, instead of the ironed out version that the llm produced | | |
| ▲ | jonas_scholz 3 hours ago | parent | next [-] | | fine, i agree that this doesnt sound like me. But the content is correct and not a hallucination! | | |
| ▲ | nubg 2 hours ago | parent [-] | | so did you write this post with the assistance of LLMs? |
| |
| ▲ | petesergeant an hour ago | parent | prev [-] | | > the article is based on a prompt, which i would much rather read I would much rather read someone's carefully written and LLM-assissted article than shallow and lazy dismissals like this one. What, precisely, do you feel you've added to the conversation here? |
|
|
|
