dajonker

6 ms·

31.
by dajonker 9mo ago
You are equally understating past performance as you are overstating current performance. One year ago I already ran qwen2.5-coder 7B locally for pretty decent autocomplete. And I still use it today as I haven't found anything better,
32.
by dajonker 9mo ago
These LLM benchmarks are like interviews for software engineers. They get drilled on advanced algorithms for distributed computing and they ace the questions. But then it turns out that the job is to add a button the user interface and it u
33.
by dajonker 9mo ago
Yes! This update works great. Seems to be pretty good at first glance. I'll have to setup an interesting task and see how different models approach the problem.
34.
by dajonker 9mo ago
I use llama.vim with llama.cpp and the qwen2.5-coder 7B model. Easily fits on a 16 GB GPU and is fast even on a tiny RTX 2000 card with 70 watts of power. Quality of completions is good enough for me, if I want something more sophisticated
35.
by dajonker 9mo ago
Update: I'm experiencing issues with OpenCode and this model. I have built the latest llama.cpp and followed the Unsloth guide, but it's not usable at the moment because of: - Tool calling doesn't work properly with OpenCode
36.
by dajonker 9mo ago
I don't know whether it just doesn't work well in GGUF / llama.cpp + OpenCode but I can't get anything useful out of Devstal 2 24B running locally. Probably a skill issue on my end, but I'm not very impressed. Bench
37.
by dajonker 9mo ago
Yes I usually run Unsloth models, however you are linking to the big model now (355B-A32B), which I can't run on my consumer hardware. The flash model in this thread is more than 10x smaller (30B).
38.
by dajonker 9mo ago
Great, I've been experimenting with OpenCode and running local 30B-A3B models on llama.cpp (4 bit) on a 32 GB GPU so there's plenty of VRAM left for 128k context. So far Qwen3-coder gives the me best results. Nemotron 3 Nano is su
39.
by dajonker 10mo ago
Of course you can subnet ipv6, in fact I run several ipv6 subnets at home. You have to delegate a different prefix to each subnet.
40.
by dajonker 10mo ago
At least CPython and CRuby (MRI), the most common implementations of each language, ignore all type hints and they are not able to use them for anything during compile or runtime. So the performance argument is complete nonsense for at leas
41.
by dajonker 10mo ago
Well said. There are many problems you have to deal with when writing code and type annotations only solve one particular kind. And even type annotations can be wrong: when you're dealing with data from external sources, dynamic langua
42.
by dajonker 10mo ago
I don't have a great code example at hand unfortunately, but I found that people often tend to write more "nominally" typed code (expecting explicitly named classes) rather than taking advantage of duck typing (interfaces, st
43.
by dajonker 10mo ago
In my honest opinion, if you can't live without static typing, Ruby just isn't for you. Adding static typing to a dynamic language mostly gives you the disadvantages of both, without a lot of benefits. It's better to stick to
44.
by dajonker 1y ago
Even if the hardware is really good, the software should be even better if they want to succeed. Support for operating systems, compilers, programming languages, etc. This is why a Raspberry Pi is still so popular even though there are a lo
45.
by dajonker 1y ago
Try MinerU 2.5 with two-step parsing. It gives good results with bounding boxes per block. Not sure if you can get it to do more detailed such as word or character level.
46.
by dajonker 1y ago
Implementing a doubly linked list without either unsafe or some very confusing code that could arguably win an obfuscation contest.
47.
by dajonker 1y ago
If I understand the article correctly it's for those cases where you want memory safety (i.e. not using "unsafe") but where the borrow checker is really hard to work with such as a doubly linked list, where nodes can point to
48.
by dajonker 1y ago
I believe patents play a big role here as well. Anything new must be careful to not (accidentally) violate any active patent, so there might be some tricks that can't currently be used for AV1/AV2
49.
by dajonker 1y ago
That graphic from the Swiss Railways is absolutely wonderful https://sma-partner.com/storage/app/media/Dokumente/Netzgraf...
50.
by dajonker 1y ago
A dynamic language with type hints has all of the disadvantages of a dynamic language with all the disadvantages of a static language on top of it.
51.
by dajonker 1y ago
I haven't experienced any price increase on the cheapest Postmark tier over the past 3 years or so? In any case they deliver excellent service and as a business earning money and sending emails per transaction it's almost free.
52.
by dajonker 1y ago
The edit prediction model is open source https://huggingface.co/zed-industries/zeta It's based on qwen 2.5 coder 7B and you should be able to run it locally quite easily since it would only require about 8 GB of V
53.
by dajonker 1y ago
Exactly, makes it feel almost like an advertorial for Anthropic, who likely need most customers to pay 1000 bucks a month to break even.
54.
by dajonker 1y ago
I guess that depends on your definition of "decent". For the smaller models that can run on a 16/24/32 GB nvidia card, the chip is anywhere between 3x and 10x slower compared to say a 4080 super or a 3090 which are relat
55.
by dajonker 1y ago
Would this be preventable by setting strict alignment for SPF? i.e. aspf=s in the DMARC DNS record.
56.
by dajonker 1y ago
It is. To some it is more fun to reinvent the wheel than to study history
57.
by dajonker 1y ago
A lot of us programmer folk are indefinitely in search of that one thing that will finally let us write the perfect, bug-free, high performance software. We take these concepts to the extreme and convince ourselves that it will absolutely w
58.
by dajonker 1y ago
Gemini 2.5 pro is, in my usage, quite superior for high quality transcriptions of phone calls, in Dutch in my case. As long as you upload the audio to GCS there you can easily process conversations of over an hour. It correctly identified a
59.
by dajonker 2y ago
It's a necessary evil, isn't it? You're not programming your application in pure HTML or SQL, you write it in some general language that can do much more. But the browser and database don't speak that language natively.
60.
by dajonker 2y ago
The Ryzen APU could be interesting for running local AI models and with 128 GB of RAM you can fit quite a large model. Plus it should be relatively energy efficient compared to a full size desktop build with separate GPU. Lack of PCIe 5.0 i

More ›