A very beautiful place on earth: one friend who lives there enjoys road biking in the surrounding mountains. Another friend does cross-country skiing in his lunch break (in winter times).
I can totally see that. I imagine it's great for hiking and food. But if/when this company has an issue you'll have to move again or start working at the local Sparkasse or something.
I live nearby, its an amazing area, one of the sunniest in Germany with perfect landscape. Not sure about the talent pool though but there is a university.
You are completely ignorant. This is one of the best places to live and work on earth. Nonetheless, they also have positions in San Francisco for what is worth.
Usually Justwatch should be able to tell you where something is available. However I will note that it's remarkably common for something slightly old to just not be available anywhere. You cannot watch Das Boot in my region on any service so far as I can see. So I guess you loan from the library or buy DVDs off of ebay or something?
This critique has accumulated some issues over time. Modern approaches like GTAO are way less intense than before. They essentially only add shadows to crevices. A normal corner like this will barely look different. More importantly, techniques like SSGI will correctly take color into consideration to darken black corners and brighten white ones.
In my experience, yes. A bit more reliable than gemma for me. I mostly use A3B (35B, mix of experts) though, because it‘s faster, and in the same ballpark intelligence wise as the dense 27B, so it’s the sweetspot for me. I want to try cohere‘s mini code model next, but worried the runtimes aren‘t optimized for that yet.
Worth knowing that Unsloth have just put out another Gemma 4 release from Google's upstream updates which should improve reliability. Bugs in the chat template affecting tool calling and other issues, apparently. https://www.reddit.com/r/unsloth/s/MpC6Hzs4Wj
Yeah, it's only a chat template change. So if you don't fancy re-downloading the whole model and can just hack the new chat template into your process, it's a lightweight test to do.
The best model you can run locally is Kimi K3, as long as you have the hardware. If "what model I can still run on a something resembling something I can put on desktop without separate electricity and cooling water inputs", then it is probably GLM 5.2 (can be run on e.g. Nvidia DGX Station workstation). As long as you have about $100k-$150k.
I have been running 3.6 27b on a dual AMD r9700 setup using Opencode and Matt Pocock's skills workflow for writing Golang CLIs. It's decent, but won't win any awards on code architecture. I guess you can try to AGENTS.md the deficits but I am just exploring its raw Opencode experience right now. Much slower than an API but still 3x times faster than I can read. Tuning it in with a community chat template and a specific penalty for repeats was the sauce needed to get it to work. I can probably start loop daddying it now over the tickets Matt's flow creates.
So yeah, it's the best local model I've seen. I am going to try the Qwopus 3.6 fine tune soon with the same spec and tickets and compare the output of both.
Would you mind sharing more please? I literally just finished the same set up, with a 9950X CPU and 192G RAM at 4,800 MT/s. I used lemonade with Vulkan and the UD-Q8_x_x model from HF. 256k context. I have about 8G VRAM free, and use the iGPU for my desktop/monitor on Arch. What options do you give llama-cpp or whatever you run please? What other models have you found fit nicely in the 2xR9700? Thanks!!
I actually have long discussions with Gemini about this and have wound up download a bunch of different models for different things. There is no best, just fast but worse, slow but better, agentic or not, reasoning or not great at large contexts, better world knowledge, uncensored, etc…. It’s a bit daunting actually since there isn’t really a one size fits all model that you can just use for everything.
Yes it's between this and Gemma 4 31B which is much slower, but looks like it won't ever get an upgrade. I have to conclude that the MoE variants are unreliable, and MTP sometimes just can't get tricky formatting right.
The whole series had an upgrade a couple of days ago actually — they have addressed embedded tool calling (and hopefully the MTP formatting stuff though I gave up running the Gemma MTP because it's often slower than not-MTP)
Not tried it yet but I've seen tests that suggest they've properly fixed the tool calling issues.
For whatever reason prefill (on my DGX Spark) is faster with the Gemma models than Qwen 3.6 models of similar size. On vLLM anyways. Likely just deeply tuned code contributed to vLLM by Google?
vLLM gives me ~7000+ tok/sec with Gemma 4's MoE model. Vs ~6000 tok/sec for Qwen 3.6 MoE.
I find the 4-bit QAT with MTP to be entirely usable speed on both my boxes (Strix Halo and a desktop with two V620 GPUs, which are slightly faster than the Strix Halo).
I don't necessarily share this view, but I've heard it enough times to steelman it: It's about dignity. I don't want to feel like someone owns me or can tell me what to do just because they lose some pocket change in my direction. You know the notion of the "deserving poor" that deserves our aid and pity? There's a corresponding idea of a deserving customer. I'd do my best impression of a dog for you if you ask nicely, but if you offer me $100 for it then I'll be insulted. If you offer me $10000 I'll do it but I'll resent you for it.
I get the point, but I assume the people keeping the museum open at 2 AM for some rich dude are the same people who do exactly that kind of job in normal hours.
In other words, they are asked to do overtime. So, as long as they are getting adequately paid for that overtime and they are not forced to do so (but I assume you can always find at least two people out of 20 who could do that), I don't think it's a big deal.
A better comparison - you're a member of the https://en.wikipedia.org/wiki/Actors%27_Equity_Association, the part which you auditioned for says "dog", and the script & stage directions are within general AEA guidelines for animal parts.
I've only volunteered in museum work - but there are obvious social rules about interacting with wealthy benefactors, and I'd assume that any competent employer in this space is carefully screening for comfort and skill with those rules.
(BTW, "wealthy benefactor" is very much a pay-to-win system, with tier after tier of "just how many $thousands did you graciously gift us with this year?" reward levels.)
If someone is doing it because of the arm twisting done by the economy and personal financial circumstances then of course it will not land well. I will feel exploited.
On the other hand if I am coming from a place of joy to be an usher at an odd midnight hour to a more sic recital, where I am paid peanuts or nothing at all, but I get to see the show and participate in the collective enjoyment of wonder it would land very very differently.
People get paid to sell their organs to tread water, I doubt they feel happy about it.
I think that kind of breathless excited tone is justified by the subject matter. Unlike the usual LinkedIn posts this is actually about a pretty dire situation.
There are areas where the bureaucratic hurdles to changing anything and the incentives for changing anything work out to nothing ever changing. I assume in 20 years most of Berlin is still going to have 50mbit/s max. I hear residents of New York have completely given up and are using 5G modems because putting up new cables just isn't practical. On the other hand, these cities do have a significant minority of flats with gigabit internet, so if you care you can pick a modern building with modern cabling. Maybe the segment who both live in old apartments and also are willing to pay for fast internet is too small to bother with.
I bet it's being organized by project rather than product. Conway's law ensures such an org will create code around projects, not products, and that always ends horribly.
reply