HackerTrans
トップ新着トレンドコメント過去質問紹介求人

kbumsik

no profile record

投稿

Ray Tracer in SQL

github.com
58 ポイント·投稿者 kbumsik·10 日前·22 コメント

Ray Tracer in SQL

github.com
3 ポイント·投稿者 kbumsik·11 日前·1 コメント

Python JIT project was asked to pause development

discuss.python.org
182 ポイント·投稿者 kbumsik·先月·101 コメント

Python JIT project was asked to pause development

discuss.python.org
6 ポイント·投稿者 kbumsik·先月·1 コメント

Kimi K2.6

huggingface.co
11 ポイント·投稿者 kbumsik·3 か月前·1 コメント

Logitech Options+ and G Hub macOS Certificate Issue

support.logi.com
3 ポイント·投稿者 kbumsik·6 か月前·0 コメント

コメント

kbumsik
·27 日前·議論
So no multimodal support yet?
kbumsik
·2 か月前·議論
So the draft model's performance is directly linked to the overall speed. Thank you for the explanation!

By the way, can it be slower than without speculative decoding in worst case then?
kbumsik
·2 か月前·議論
> performance often degrades under different chat templates, long-context inputs, or out-of-distribution system prompts.

I heard that speculative decoding doesn't affect performance (I meant accuracy). Am I wrong about it?
kbumsik
·4 か月前·議論
Right? Even enterprise routers, e.g. Cisco, are not produced in USA.
kbumsik
·7 か月前·議論
We switched from Celery to Temporal. Temporal is such a great piece of distributed system.
kbumsik
·9 か月前·議論
> realistic written documents?

Just get a DEF 14A (Annual meeting) filing of a company from SEC EDGAR.

I have seen so many mistakes when looking at the result closely.

Here is a DEF 14A filing from Salseforce. You can print it to a PDF and then try converting.

https://www.sec.gov/Archives/edgar/data/1108524/000110852425...
kbumsik
·9 か月前·議論
Right, I wouldn't use full table detection to VLM model because they tend to mistake with numbers in table...
kbumsik
·9 か月前·議論
> My impression is that OCR is basically solved at this point.

Not really in practice to me. Especially they still struggle with Table format detection.
kbumsik
·10 か月前·議論
Well, that has nothing to do with JS itself though.
kbumsik
·10 か月前·議論
In case if you don't know, Auth.js is not a frontend-only framework. It uses a backend server to make it secure.

So it basically has no difference from the alternatives you mentioned.
kbumsik
·10 か月前·議論
It is really easy to tell the difference btw. You will always see "4" or "7" in the middle.
kbumsik
·4 年前·議論
Really? I have more than 1000k files and I have never faced issues for more than 7 years.
kbumsik
·6 年前·議論
Depending on your motherboard you might need two GPUs. In my case I use Proxmox and control VMs through is web GUI.
kbumsik
·6 年前·議論
I am using Windows VM with GPU passthrough too. If you want it out of the box (well, kinda) I would recommend Proxmox. Proxmox is amazing hypervisor OS. Otherwise you need to study tricky things a lot: QEMU, Kernel configuration for VFIO.

Also you may need hardware, especially motherboard that supports VFIO groups well.

Fortunatly there are quite big communities like Reddit in this topic. Just google "VFIO"
kbumsik
·7 年前·議論
Have you heard of AWG Gov tier?
kbumsik
·7 年前·議論
I'm not an expert in C, but then what's the issue with strncpy() or any "n" functions? It prevents overflow AFAIK. Also what is the alternative (memcpy?) and why?
kbumsik
·7 年前·議論
Because of possible string buffer overflow?
kbumsik
·7 年前·議論
So is it fully migrated to Gtk3 now?
kbumsik
·7 年前·議論
Most CBOR implementations for MCUs tend to be a part of RTOSs, integrated and optimized for their own memory allocation libraries. But many of them are based on Intel's TinyCBOR: https://github.com/intel/tinycbor
kbumsik
·7 年前·議論
It it also weird that "network byte order" is considered as BE. Is there any advantage of BE when it is transferred over the network?