nyrikki

10 ms·

91.
by nyrikki 7mo ago
Can you please explain the use here? I tried the demo, and cat, cp, echo, etc... seem to do the exact same thing without the cost. Their demo even says: `Paste any code or text below. Our model will produce an AI-generated, byte-for-b
92.
by nyrikki 7mo ago
I started using the ports and adapters pattern and protocol for any packages that have replacements or concerns. Basically treating HTTP requests as an orthogonal, or cross-cutting concern. It is sometimes hard to tell if these upstream pac
93.
by nyrikki 7mo ago
That’s why I either started with the ports and adapters pattern or quickly refactored into it on spikes. You don’t have to choose what flavor of DDD/Clean/… you want to drink, just use some method that keeps domains and use cases
94.
by nyrikki 7mo ago
I think there is a bit of the map territory relation here. > First, it's not a question of decidability but of tractability The question of decidability is a form of many-to-one, reduction. In fact RE-complete is defined by many-to
95.
by nyrikki 7mo ago
> However, as the name “batch-invariant” suggests, the technique is currently limited to handling variations related only to the batch dimension, making it robust to continuous batching and other batch-size–related changes, but not to ot
96.
by nyrikki 7mo ago
For binary operations, NaN values compare as unordered. The IEEE 754 Specification requires that >,<,= evaluate to False. Saying that two incomparable objects become comparable let alone gain equally would break things. We use specifi
97.
by nyrikki 7mo ago
Even with classic compilation, it is only the semantic behavior that is preserved. What the Church–Rosser property/confluence is in term rewriting in lambda calculus is a possible lens. To have a formally verified spec, one has to use
98.
by nyrikki 7mo ago
> Can't they run an example prompt and verify they get the exact same output token probabilities for all prompts? You don’t even get that with GPUs in general, or really floating point in general. The Art of Computer Programming. Vo
99.
by nyrikki 8mo ago
> Prop vs Decidable - I vaguely understand the distinction, but can use more examples. Also would like to know does this relate to the noncomputable property. My path to this subject was tortured, so sorry if I don’t account for Polysemy
100.
by nyrikki 8mo ago
This is exactly why I moved to self hosted code in 2017. While I couldn’t have predicted the future, even classic data mining posed a risk. It is just reality that if you give a third party access to your data, you should expect them to use
101.
by nyrikki 8mo ago
The memory bandwidth of the Laptop CPU is better for fine tuning, but MoE really works well for inference. I won’t use a public model for my secret sauce, no reason to help the foundation models on my secret sauce. Even an old 1080ti works
102.
by nyrikki 8mo ago
I can't answer for the OP but it works fine under llama.cpp's container.
103.
by nyrikki 8mo ago
It is crazy to me that it is that slow, 4 bit quants don't lose much with Qwen3 coder next and unsloth/Qwen3-Coder-Next-UD-Q4_K_XL gets 32 tps with a 3090 (24gb) as a VM with 256k context size with llama.cpp Same with unsloth&#x2F
104.
by nyrikki 8mo ago
Docker Sandboxes are microVMs. Basically due to many reasons, ld_preload, various containers standards, open desktop, current init systems, widespread behavior from containers images from projects, LSM limitations etc… It is impossible to m
105.
by nyrikki 9mo ago
Warning, this post is not covering the ‘systems thinking’ that most of you will be expected to know in staff level jobs, it is using the term for up front design. In a more typical modern sense systems thinking is more about relationships a
106.
by nyrikki 9mo ago
Also note you can count to 12 on one hand and 60 with the other. That is why the ancient Sumerians used it. Base 10 was added to Roman abacus but they still kept the uncia (12) for some functions. IIRC that wasn’t droop until the renaissa
107.
by nyrikki 9mo ago
Had more to do with punch cards and flexowriter tapes and octal, which predates large word sizes or even mainframes. Note the following from the MIDAS macro assembler [0] Fortran predates this and was a different lineage than IBM, but not h
108.
by nyrikki 9mo ago
OCI Container Runtimes like OCI's runc are "container runtimes", so the runtime spec[2] Basically, docker started using lxc, but wanted a go native option, and wrote runc. If you look at [0] you can see how it actually instan
109.
by nyrikki 9mo ago
An ISDN BRI (basic copper) actually had 2 64kbps b channels, for pots dialup as an ISP you typically had a PRI with 23 b, and 1 d channel. 56k only allowed one ad/da from provider to customer. When I was troubleshooting clients, the pr
110.
by nyrikki 9mo ago
Unfortunately even podman etc.. are still limited by OCIs decision to copy the Docker model. Crun just stamp couples security profiles as an example, so everything in the shared kernel that is namespace incompatible is enabled. This is why
111.
Systemd, Vsock, & OpenSSH-Server(blog.nsrun.io)
2 points by nyrikki 9mo ago|2 comments
112.
by nyrikki 9mo ago
I will try to go more in-depth in later posts, but many users, especially in a k8s context probably have a socket activated sshd listener on vsock, that may pose a serious risk and possibly violate your security assumptions.
113.
by nyrikki 9mo ago
I was looking at the PDF [0] and [1] and [0] calls out: > In addition to sampling error, question wording and practical difficulties in conducting surveys can introduce error or bias into the findings of public opinion polls. ``QN5:Perso
114.
by nyrikki 9mo ago
I don't think [0] is showing what you think it does. > % Very satisfied with the way things are going in personal life That Dropped from 65% in 2020 to 44% in 2025 > Record-Low 44% of Americans Are 'Very Satisfied' With
115.
by nyrikki 9mo ago
The old efnet #unix sex chart debunks some of these claims. I know I learned more from having ops in #unixhelp than stack overflow. But the 90s were different, and it really depended on the channels. I remember going into #nanog to finally
116.
by nyrikki 9mo ago
I should add, the o200 was a 4u chassis with a desk side skin option. If you popped off the plastic and had rails you could convert them. The o200 had stonith (shoot the other node in the head) ports like the o2000 for clustering etc..
117.
by nyrikki 9mo ago
I actually forgot that they made a low end ‘Indy’ and never saw one. We called the Indigo2 Indy There were some oddities with ad dollars and prestige clients in the 90s, where systems would be upgraded, avoiding swapping out the serialized,
118.
by nyrikki 9mo ago
There is a lot to blame on the OS side, but Docker/OCI are also to blame, not allowing for permission bounds and forcing everything to the end user. Open desktop is also problematic, but the issue is more about user land passing the bu
119.
by nyrikki 9mo ago
It must have been more complex then that, I was at the skywalker ranch when onyx was being replaced by o200/2000 and never saw a purple onyx at Kerner(ILM), I had a purple Indy impact as my home machine and was looking for purple. I ha
120.
by nyrikki 9mo ago
IIRC the desk side onyx had the royal purple stripes and only accepted one CPU board, the rack mount version were that blue-purple color, more indigo (color)

More ›