Was hoping it was so easy :) But I probably need to look into it some more.
llama_model_load: error loading model: error loading model architecture: unknown model architecture: 'gemma4'
llama_model_load_from_file_impl: failed to load model
Edit: @below, I used `nix-shell -p llama-cpp` so not brew related. Could indeed be an older version indeed! I'll check.
As it has been discussed in a few recent threads on HN, whenever a new model is released, running it successfully may need changes in the inference backends, such as llama.cpp.
There are 2 main reasons. One is the tokenizer, where new tokenizer definitions may be mishandled by the older tokenizer parsers.
The second reason is that each model may implement differently the tool invocations, e.g. by using different delimiter tokens and different text layouts for describing the parameters of a tool invocation.
Therefore running the Gemma-4 models encountered various problems during the first days after their release, especially for the dense 31B model.
Solving these problems required both a new version of llama.cpp (also for other inference backends) and updates in the model chat template and tokenizer configuration files.
So anyone who wants to use Gemma-4 should update to the latest version of llama.cpp and to the latest models from Huggingface, because the latest updates have been a couple of days ago.
I just hit that error a few minutes ago. I build my llama.cpp from source because I use CUDA on Linux. So I made the mistake of trying to run Gemma4 on an older version I had and I got the same error. It’s possible brew installs an older version which doens’t support Gemma4 yet.
And that's exactly why llama.cpp is not usable by casual users. They follow the "move fast and break things" model. With ollama, you just have to make sure you're getting/building the latest version.
Its not possible to run the latest model architectures without 'moving fast'. The only thing broken here is that they are trying to use an old version with a new model.
I'm a bit unsure what that has to do with someone running an outdated version of the program while trying to use a model that is supported in the latest release.
Part of this is also our culture that somehow decided kids need their own bed, and it's easy to get baby formula.
So the kids are not sleeping in our beds, where they feel 100% secure, getting to the breast whenever they want (and they quickly will want it at a lesser frequency). The woman will feel this, but hardly has to wake up, me... I slept right through all that. Fwiw, we had a bed for the baby that attached to ours.
In our time everybody advised us: Give the bady a load of milk at 23:00 just before you go to bed! We never did, just stuck to about did 20:00, or just when baby cried, both babies took about 2 months to sleep for about 12 hours straight (although soon after the second one developed reflux which had me watch Rick and Morty in its entirety somewhere between 2 and 4 for some time).
Anyway, not saying everybody is that lucky, just saying sometimes it's good to questions things that are given in one's culture. Worst advice imho is "let the baby cry" which was common on our days. How nice to let a baby cry alone in a room, not understanding anything about what happens...
I coslept, but I had shit milk production. Without formula my kid would be dead. My friend breastfeeds, but is an active danger when asleep, so without a crib, she'd have crushed her child.
It turns out that safe sleep rules and the availability of formula exist for a reason. Safe sleep rules exist in the west because pur beds are fundamentally different (and more dangerous) than in places here cosleeping is more common. Tp cosleep you need a certain situation that many people are not prepared to deal with.
There's literally nothing you can do about low supply at all. It's not a matter of trying for me. My body never made more than an ounce even with weeks of attempts. This is even setting aside that some people would like assistance so they can sleep and breastfeeding means dad can't take on night feeds, which is what another friend is experiencing and the child is having a bad time from her severe sleep deprivation.
And even more complications of small child. It's not as simple as "let's go back to the old days". The great days when kids died at much higher rates remember.
Sorry, I did not mean it like that, formula is fine. We also hated the breastfeeding mob.
And you're right about the rules they exist for a reason, but I think we should as parents take our space to try what works for us and our kids and what feels right.
Sometimes I just can't grasp how cruel nature is. Imagine no formula -> baby starving and you just have to accept it. When the tribe grows in numbers the baby is safe. When the tribe grows in numbers like mankind, you get cheats like formula and polio vaccine.
We're just replaying the life game on easier mode.
That is good to know. My original post was a bit reductionist, I do generally feel that people try to do good and are much more co-operative than adversarial. It is just that we can be dragged into a defensive/attack position easily if we do not keep an eye on that.
I suspect that despite all the issues of the world, we are slowly heading to a future that will be much more peaceful.
The book gives a lot of examples of where and how the beautiful sides op human nature surface in difficult times. It also argues that at least some of the things we hold as true (ie civilized behaviour is a thin layer of veneer) are not true.
One example is that we often think that as in "Lord of the Flies", a book that shaped us in a way, civilization slowly drains from the group as they become more primal when hunger sets in and oversight is lost. But Rutger finds an example of a real group of stranded kids, who thrived and generally showed admirable behaviour. That's just one of the examples.
Fwiw, yesterday I was messing about with GitLab backups. One of the options for direct off-site is S3. But I want that S3 bucket to live on a box on my own Tailnet.
So I too just want simple S3. Minio used to be the no-brainer. I'll checkout RustFS as well.
It does not sound hard (although it is hard for me!). It sounds like it should be some LinuxServer io container. Doesn't it? At this point S3 is just as standard as WebDav or anything right?
Nothing changed, it's new ground, we are searching it with a search light. From some vantage points our view on things may feel quite complete, even insightful. Then we look at if differently and feel lost. It's a process we are in together.
I started my client's site on Hugo, withing 2 days I was editing something for them every 30 mins (slight exaggeration). They wanted something they could edit, they don't do Markdown, they don't manually write URLs they want to drag images into their posts and pages.
Eleventy + Decap CMS + gitlab has been working well for our marketing team for the past 3+ years. Also, with Claude/Gemini/Copilot it is a lot easier to edit templates and add new pages when all of the files are right there on your filesystem.
Normal people refuse to learn markdown - they want RTF editor.
I could understand someone might refuse learning LaTex but markdown is so simple.
RTF editing sucks badly if you have to include it in your project. No one wants to specifically pay for implementing it but they also expect it to be there.
Nowadays you can just use one of those WYSIWYG markdown editors that come with a toolbar (incl. buttons for inserting images) and hide the formatting syntax by default.
But each editor also has its issues with generating "backend html" doing bunch of wonky stuff so it works for this editor — using that text in various other places and when you have multiple rich text fields and not single big one to edit whole document is always a major PITA.
Then you get whatever they write to PDF report so for example you have to render HTML correctly there. Amount of ways it can break is basically infinite, getting paragraphs page breaks is non trivial amount of work, especially when customer wants their own layout for the report and not generic looking or just broken layout. So problem is mix and match display of whatever they write in different places.
Not to mention, everyone wants prefilled templates, so they don't start from scratch, oh and your templates need to have dynamically filled in placeholders, now you have to put something like tags that will be updated by your back end.
Maybe you need to send it via API and all kind of companies have WAF on incoming/outgoing data then you have to strip tags.
Yes you can encode/decode, limit options, white list allowed tags, and I was doing that for years now, but amount of things that break is still big and another customer wants you to enable lists when you wanted to
support just italic, bold, then you have whole blast radius and feature creep is real.
ROI is just not there, as I mentioned no one wants to specifically pay for all that, we have a really good run telling customers to just use plain text, amount of regressions to be tested, amount of expectations of things to work out of the box once you go with rich text is really high.
Bar to jump to is basically re-implement MS Word — oh did I mention everyone will expect copy pasting from Word to work perfectly - just imagine how much time your customer support has to spend explaining you limited options in that field to be just bold and italic.
RTF editor may be part of WYSIWYG solution but it also might be used to edit text that on publication will be looking entirely differently or will be used in various places.
Ohh, I think I get it now – is this about literally using RTF as an intermediate format? I really don’t think I’ve heard about it being used in the context of static sites, but I see how it might make sense as a way of storing, well, rich text.
This is generally solvable unless they wanted radical layout changes. There are headless CMS solutions, just ensure that no texts are hardcoded and instead put in a config. Such configs can be made editable in the CMS.
I consider myself a Power User, use of Windows is not friction free :)
Over the years I've come to believe that there is only one thing important: What you are used to. The friction is in the change process. Not in the destination.
As an independent, I have several customers on MS365, you know what my super power is? FireFox cookie containers. One for each org, and I switch with 0 effort between the orgs. No need for Windows in that workflow at all. In fact, using Windows and the native apps would probably give me a lot more friction.
Yes, sometimes I have issues. I.e. yesterday Word kept deleting my last 1-2 sentences for some reason, even though hitting ctrl-s tells everytime: "I should not worry". but in general it's fine.
My business is on Proton, and I love that MS365 AND Google workspace calender invites go right into my agenda with no effort. There is nice stuff out there. Especially now we have Proton Meet, I can take some ownership over videocalls in Teams and Google Meet finally.
Absolutely. I've given using a tablet (with keyboard) as an alternative to a laptop when traveling and it sort of frustrates me for a lot of things. But talking to people I know who have largely switched over, my conclusion is that, in general, I probably mostly just haven't put the effort and commitment to make it worth it for me. And I'm not sure, not spending nearly as much time on planes as I used to, it's worth it relative to getting a laptop that is even lighter than the combination.
As part of the human species, which has conquered our planet's poles, its deserts and its jungles, I believe we are in a unique position to adapt to many -if not most- circumstances thrown our way, and flourish.
You hopefully can adapt to what you need to. That's not the same things as switching to something you find awkward and you don't find to have a particular, if any advantage.
You know what is nice? If you have clients that get automatically switched to "the new Outlook" and loose all imap connections (and they don't work anymore, period).
Took me so long to learn that the fix was to switch back to the old Outlook.
IMAP works in outlook. Its just horrible to set up and half broken. Click "Add account". Then type in your email address, click "Choose provider", select IMAP, then click "Sync directly with IMAP" (dark pattern hidden button). If you don't click that last button, outlook uploads your IMAP email credentials to their own MS Cloud instance, and that proxies all your emails via microsoft's cloud servers. Do they read your email messages for advertising? Nobody knows!
In my testing, the local IMAP client implementation quite frequently launches a DoS attack against your IMAP server. It'll send the same query requesting new mail messages in a tight loop, limited by the round-trip latency. But luckily, almost nobody uses IMAP via outlook because its so difficult to set up.
> If you don't click that last button, outlook uploads your IMAP email credentials to their own MS Cloud instance, and that proxies all your emails via microsoft's cloud servers. Do they read your email messages for advertising? Nobody knows!
I've seen cases where people have it set up like that and it's so awfully slow. Minutes to display a single new message. That cloud brings absolutely zero user-benefit.
reply