I want to promote the use of local/small infrastructures but the tools to define them in sofware stack are quite heavy and bad and not quick and lean.
For testing, debugging etc. of this type of project I'd recommend pi, as it seems to be only of these agentic clli's not being slop by itself (was developed by industry seasoned game programmer).
For testing, debugging etc. of this type of project I'd recommend pi, as it seems to be only of these agentic clli's not being slop by itself (was developed by industry seasoned game programmer).
0
0
0
Jarkko Sakkinen
jarkko
Edited 23 days ago
I'm developing my own ostree and BuildRoot based operatingy system:
- https://codeberg.org/puu
- There will be puu-desktop and puu-llm builds.
- puu-desktop uses ext4.
- puu-llm uses ext4 + xfs for the models.
- puu-llm hosts k3s, llm-d and generally very latest of stack for running local LLM modesl.
- GNOME build is tough and I'm working on it but I have plan how to make in happen on BuildRoot.
- Given being BR2_EXTERNAL, builds and configures with beloved kbuild :-)
Despite having puu-llm build this project disregards AI contributions.
That said. I don't see it as a crime that you generate some initial code with LLM.
I do that sometimes too especially when I need to get an idea of e.g. alien projec. However, usually I do also spend week or few applying hard manual labor to make the actual code change I want to do despite having some random slop that functionality fits the bill.
This is what I do accept but it is generally best just to say plain no in README.md as most people will likely not be able to do such weighting.
[The logo image artwork is copyrighted by me (drawn with Inkscape) and licensed with CA BY-SA 4.0]
- https://codeberg.org/puu
- There will be puu-desktop and puu-llm builds.
- puu-desktop uses ext4.
- puu-llm uses ext4 + xfs for the models.
- puu-llm hosts k3s, llm-d and generally very latest of stack for running local LLM modesl.
- GNOME build is tough and I'm working on it but I have plan how to make in happen on BuildRoot.
- Given being BR2_EXTERNAL, builds and configures with beloved kbuild :-)
Despite having puu-llm build this project disregards AI contributions.
That said. I don't see it as a crime that you generate some initial code with LLM.
I do that sometimes too especially when I need to get an idea of e.g. alien projec. However, usually I do also spend week or few applying hard manual labor to make the actual code change I want to do despite having some random slop that functionality fits the bill.
This is what I do accept but it is generally best just to say plain no in README.md as most people will likely not be able to do such weighting.
[The logo image artwork is copyrighted by me (drawn with Inkscape) and licensed with CA BY-SA 4.0]
1
0
2
Jarkko Sakkinen
jarkko.gitignore I always specify always specify for repositories that are cloned as part of directory tree (such as my root home directory):
/*Jarkko Sakkinen
jarkko
Something I will take into use at instant:
1. https://github.com/btford/write-good
2. https://github.com/ubaldot/vim-writegood
1. https://github.com/btford/write-good
2. https://github.com/ubaldot/vim-writegood
0
0
0
Jarkko Sakkinen
jarkko
I tried Suno Studio so that I can spare you from doing it. Here's a longer review.
Disclaimer. I don't have any problems Suno Studio existing but what can I say is that it piece of shit DAW by any reasonable standards 🤷
Suno Studio probably works if you want to make a shitty wedding song with your and your spouses name mentioned in the song lyrics but making a song with it is a complete suicide mission :D IF you know how to compose, arrange and no what you're after i.e, do a stuff that musicians generally do.
Obviously the actual DAW functions are complete garbage. When comparing to the staples of web DAWs such as Bandlab, Suno Studio is just plain ridiculous.
The AI functionality serves no purpose. I don't have any use of bad quality generated audio clips, as Splice already serves me if I want to pick up off-the-shelf samples.
E.g., a useful AI functionality would be to match the MIDI edits to "similar script" but this is actually out of scope for Suno Studio because it has no MIDI clips in the first place except as exports. And I never convert audio to MIDI because I can do that already with my ear if I hear something nice :-)
This is a wider issue of AI products because they are just generally awful and nobody wants to pay for them.
Suno Studio is a complete joke but here's some other web options that can do the job:
1. https://www.audiotool.com/
2. https://www.bandlab.com/
Disclaimer. I don't have any problems Suno Studio existing but what can I say is that it piece of shit DAW by any reasonable standards 🤷
Suno Studio probably works if you want to make a shitty wedding song with your and your spouses name mentioned in the song lyrics but making a song with it is a complete suicide mission :D IF you know how to compose, arrange and no what you're after i.e, do a stuff that musicians generally do.
Obviously the actual DAW functions are complete garbage. When comparing to the staples of web DAWs such as Bandlab, Suno Studio is just plain ridiculous.
The AI functionality serves no purpose. I don't have any use of bad quality generated audio clips, as Splice already serves me if I want to pick up off-the-shelf samples.
E.g., a useful AI functionality would be to match the MIDI edits to "similar script" but this is actually out of scope for Suno Studio because it has no MIDI clips in the first place except as exports. And I never convert audio to MIDI because I can do that already with my ear if I hear something nice :-)
This is a wider issue of AI products because they are just generally awful and nobody wants to pay for them.
Suno Studio is a complete joke but here's some other web options that can do the job:
1. https://www.audiotool.com/
2. https://www.bandlab.com/
0
1
1
Jarkko Sakkinen
jarkko
@pauldoo @inffy @ikkeT
This got me hooked. I have my sandbox "arch distro". I'm hammering it into container and pre-populate most of GNOME apps from Flatpak. And for developement enviroments I add distrobox to the core image.
I don't have anything from AUR (except bootc needs to be hooked up from there) and my command-line tools are like "20 years the same and still ongoing" type of selection for the most part :-) (e.g., I even favor regular vim over nvim).
This got me hooked. I have my sandbox "arch distro". I'm hammering it into container and pre-populate most of GNOME apps from Flatpak. And for developement enviroments I add distrobox to the core image.
I don't have anything from AUR (except bootc needs to be hooked up from there) and my command-line tools are like "20 years the same and still ongoing" type of selection for the most part :-) (e.g., I even favor regular vim over nvim).
1
0
2
@pid_eins Do you think this could make sense for systemd? Just poking a bit before I do decide to do anything with it.
It's derived works of https://lwn.net/Articles/723561/ with the difference that there are no container objects, there's just a new namespace. Empty root does not categorize as a FS namespace as FS namespaces are all non-empty filesystem hierarchies (enforced by BUG_ON actually).
Have been in and between whether to send this work some day or not, as anything namespaces related feature-wise are a somewhat sensitive topics to begin with...
It's derived works of https://lwn.net/Articles/723561/ with the difference that there are no container objects, there's just a new namespace. Empty root does not categorize as a FS namespace as FS namespaces are all non-empty filesystem hierarchies (enforced by BUG_ON actually).
Have been in and between whether to send this work some day or not, as anything namespaces related feature-wise are a somewhat sensitive topics to begin with...
0
0
0
1
0
0
Jarkko Sakkinen
jarkko
Edited 27 days ago
I'm now ready to send my rootns patch set that I've had branch for some time at git.kernel.org.
I got the story now in my head why we need it (in universal sense). It has been hard to finalize the code when there is no mission to accomplish.
Now there is a mission to accomplish, which makes it easy to cover letter the code.
It's exactly Claude Mythos and similar LLM powered agentic security scanners. the window that pivot_root leaves grows too big of a risk.
We cannot rely on to stuff like this anymore.
I got the story now in my head why we need it (in universal sense). It has been hard to finalize the code when there is no mission to accomplish.
Now there is a mission to accomplish, which makes it easy to cover letter the code.
It's exactly Claude Mythos and similar LLM powered agentic security scanners. the window that pivot_root leaves grows too big of a risk.
We cannot rely on to stuff like this anymore.
1
0
0
Jarkko Sakkinen
jarkko
0
0
0
Jarkko Sakkinen
jarkko
0
0
1
Jarkko Sakkinen
jarkko
I find rpm-ostree really cool technology. Have been playing with it my X390 Thinkpad, which is like my "testing ground for new stuff".
Has this been applied to other distributions so far similarly as in Silverblue?
I'm particularly interested could I redo this with arch baseline?
Has this been applied to other distributions so far similarly as in Silverblue?
I'm particularly interested could I redo this with arch baseline?
3
0
1
Jarkko Sakkinen
jarkkoFINALLY, figured this one out:
# Apply Linux kernel patch.
macro index L "<pipe-message>b4 shazam -l -s ''<enter>" "b4 shazam"Pay attention to ''. I don’t why it fixes everything but after hours of poking on the problem I had that by accident there and suddenly it all came together 🤷
0
0
1
Jarkko Sakkinen
jarkko
Edited 1 month ago
Despite it will take still a bit of time to publish CCD, but I'll release libttk.so much sooner.
It's my retained mode single-threaded terminal UI ...
... i.e, it has jobs but they yield and multi-tasking is co-operative :-) Scenegraphs are thread-safe (or may thread-local?) in the sense that they don't shared local state.
Running fm in single-thread neither does prevent parallel processing thanks to POSIX and non-blocking I/O, and in Linux you have uring.
This is how I look at performance, when I visit the topic. I.e. not thread-bombing but only taking what you need and "stay out of the way bothering" :-)
It's my retained mode single-threaded terminal UI ...
... i.e, it has jobs but they yield and multi-tasking is co-operative :-) Scenegraphs are thread-safe (or may thread-local?) in the sense that they don't shared local state.
Running fm in single-thread neither does prevent parallel processing thanks to POSIX and non-blocking I/O, and in Linux you have uring.
This is how I look at performance, when I visit the topic. I.e. not thread-bombing but only taking what you need and "stay out of the way bothering" :-)
0
0
1