cztomsik

9 ms·

61.
by cztomsik 3y ago
Hm, thanks for pointing that out. I need nginx for SSL anyway, so I was not even thinking about going raw :) I'm wondering where exactly is the problem in the std.http.Server because it might be easy to fix. Maybe it's just that n
62.
by cztomsik 3y ago
It's using comptime inline for, see https://github.com/cztomsik/tokamak/blob/main/src/injector.z...
63.
by cztomsik 3y ago
same thing, done in a different way. the only notable difference (except of writing less code) is that zig does not type-check the code which you don't use, so the genericity goes a bit further than what you can do in rust.
64.
by cztomsik 3y ago
Oh, I didn't know std.http.Server is not supposed to be used. I guess I could just use your project (as dependency) and that would fix that perf issue? And do a thread-pool and keep-alive, of course. TBH perf was not my priority yet, I
65.
by cztomsik 3y ago
I think there are multiple reasons why Zig is great fit for this: - zig has powerful meta-programming, it's a bit like functional programming with structs. I would consider that high-level concept, at least comparable to traits and con
66.
by cztomsik 3y ago
thanks :) yeah, ava is my pet project, but there was very little activity lately because I was tight on deadline with another project. I hope to get back to it around next week.
67.
Show HN: Tokamak – Server-side framework for Zig(github.com)
109 points by cztomsik 3y ago|26 comments
68.
Ava-SQLite: Zig wrapper for SQLite with automatic migrations(github.com)
4 points by cztomsik 3y ago|1 comments
69.
by cztomsik 3y ago
FYI The migration script is not my idea, it's based on this wonderful article https://david.rothlis.net/declarative-schema-migration-for-s...
70.
by cztomsik 3y ago
AFAIK mamba is continuation of the SSM research, which is basically something called long-convolution. Instead of doing quadratic attention (computing how much each token attends to every other token) you just "somehow" compute a
71.
by cztomsik 3y ago
I only heard good things about fortran :)
72.
by cztomsik 3y ago
It's small, all-in-one binary. You only need to download model, and you can use builtin downloader to do that.
73.
by cztomsik 3y ago
How long did you wait, depending on the model size, it can take a while. Can you check logs? It's in the Settings -> System tab.
74.
by cztomsik 3y ago
LM Studio is great :) Also Faraday is great. But neither of them is open-source and this is going in a little different direction too (QuickTools, Workflows, there will be RAG and Vision support)
75.
by cztomsik 3y ago
everything what llama.cpp supports
76.
by cztomsik 3y ago
Yes, it's at localhost:3002/api/ (and api/*.zig files respectively) The Workflows screen is feature-flagged for now and only visible in the DEV build. Yes, that's the whole idea behind QuickTools - to make repetitiv
77.
by cztomsik 3y ago
This reminds me naked objects. The idea was that you only define Java classes with some annotations + theme and the whole GUI (or web frontend) application will be generated for you. Very cool idea but it never worked out (to my knowledge).
78.
by cztomsik 3y ago
it's ok, you can put the link back I don't mind the competition :) yes, I know it definitely works, so it has to be some stupid mistake but it's impossible to reproduce it without actual intel mac :-/ I had enough troubl
79.
by cztomsik 3y ago
Intel works in Ava, it's just slow, for some reason.
80.
by cztomsik 3y ago
It goes a bit against the original idea, this should be 100% air-gapped. No phoning home, no data-collection, everything is private.
81.
by cztomsik 3y ago
It should work fine if you're on Apple silicon, everything else is unfortunately second-class citizen, because I don't have any other devices and it's currently just me working on it. But if you'd like to maintain the wi
82.
by cztomsik 3y ago
Actually, it was, here's the initial thread: https://news.ycombinator.com/item?id=37561085 The big news here is that it's now open-source: https://github.com/cztomsik/ava
83.
Ava PLS: open-source application for running language models locally(avapls.com)
88 points by cztomsik 3y ago|24 comments
84.
Show HN: All-in-one desktop app for running LLMs locally(github.com)
1 points by cztomsik 3y ago|0 comments
85.
by cztomsik 3y ago
Yes, I know, the point was not to say that it's safe to depend on this universally, but rather why it's safe in JS and why it's not elsewhere -> other languages use hash maps simply because authors were either lazy or unaw
86.
by cztomsik 3y ago
I don't know what SQLite does, but in JS the order is actually defined. JS is not Java, object is not HashMap, if anything, it's closer to LinkedHashMap, but even that is not correct because there are numeric slots which always go
87.
by cztomsik 3y ago
BS=1 also means you can only decode one token. This is not always what you want, and then you're back to matrix-matrix multiply again. Most of the from-scratch llama implementations start with matrix-vector because it's simpler &a
88.
by cztomsik 3y ago
Also, if you don't know what all the toggles are for, this is a simpler attempt by me: https://www.avapls.com/
89.
by cztomsik 3y ago
Why exactly is tile-based a bad idea? Most rendering engines are tile-aware already.
90.
by cztomsik 3y ago
Functions in Zig can be called both in runtime and in compile-time. You can force some expression to be called during comptime using a keyword, and sometimes the comptime is implied (like when you define top-level const) If a function is ca

More ›