Viewing profile — sznio
sznio
HN member- Joined
- Fri, Aug 05, 2022, 12:36 AM UTC
- HN karma
- 884
- Public activity
- 285 items
- HN profile
- View on Hacker News ↗
About sznio
Socials: - t.me/szniok
Interests: Cybersecurity, Programming, Web Development
---
Recent public activity
-
comment
Comment #49243146
chop off the spine and you can use an automated scanner. to keep the book intact you'll need a flatbed scanner and an operator.
-
comment
Comment #49243114
no, because that's cost and everything needs to be the cheapest possible. only valid spend is money going to the shareholders. and chopping off the spine and putting the stack of p…
-
comment
Comment #49194146
for me, the non-committal nature of online friends is nice. the best online friends are the people I've known for 15 years, whom I just message two/three times a year for a deep co…
-
comment
Comment #49194078
I've been in a similar position 10 years ago when I was 16. I wanted to meet a bunch of friends from Arma 3, in Wrocław as well. My parents didn't allow me to go. Years later I sti…
-
comment
Comment #49179582
Parameter count is not everything. 20b parameters * 1.5 bits per parameter is just 30 billion bits, about 3.75gb a full 20b fp16 is about 40GB. I find it weird how a smaller model …
-
comment
Comment #49179542
my android phone still sends multiple SMS instead of an MMS. it might make sense if you're on a plan where an MMS costs less than 2xSMS
-
comment
Comment #49152357
use a quantized version. since it's MoE, what matters is that the 3b parameters that are used for every token fit in gpu vram, the rest can stay in system ram. really great if you …
-
comment
Comment #49152339
quantized + offload I have an RX 6700 XT with 12gb vram and 64gb system ram. running dense models like 27b is difficult, but i can run IQ4/IQ5 qwen 122b-a10b or 35b-a3b at ~20tok/s…
-
comment
Comment #49112362
why does newer Claude reject this
-
comment
Comment #49108944
>bursts so it's bursting now?
-
comment
Comment #49107146
at that much ram you can just load it outright without tricks. it will be much faster even if it ends up swapping.
-
comment
Comment #49107140
Nice. Gemma feels nice to write with but every time I use it for coding it struggles with tool calling significantly.
-
comment
Comment #49107126
distill. start with a general model, then create a smaller model trained only on relevant coding examples.
-
comment
Comment #49055835
Let's just settle it with pigeons, chickens and turkeys. Maybe too early for the turkeys right now...
-
comment
Comment #49055822
I as a random Pole with a lefty euroenthusiast stance would gladly accept the Pantheon or Notre Damme. I'd prefer Colosseum and Eiffel tower though. The blandness comes from the fa…
-
comment
Comment #49055743
I don't think any of these buildings has the recognizability potential. I could recognize the Commission too since I see it on TV every day, but others are generic offices. The Whi…
-
comment
Comment #49055727
they could put the ECB on the 500€ since a money factory is a decent image to put on a banknote, and follows the original bridge series theme of ancient-modern. but find some actua…
-
comment
Comment #49055662
oh attempting to put anyone from what now is ex-Jugoslavia on a banknote is a recipe for yet another genocide
-
comment
Comment #49055646
I think the whole original bridge and doorway was representing the whole idea of the Euro. The bridge between various nations, the doorway into a larger market. As the EU is slowly…
-
comment
Comment #49047513
to train models to detect and remove porn from their platforms thread closed
-
comment
Comment #49044733
If it's "read as an article" or something then yeah it's crap and Fable's current style isn't actually better than older models. For flowery speech old models are perfectly fine. F…
-
comment
Comment #49031599
near sure, i live in a city and people have lines right next to their homes, but never right above them because that's what causes issues.
-
comment
Comment #49028296
you use more memory?
-
comment
Comment #49024091
yep, theft but not plagiarism.
-
comment
Comment #49002571
i assume openai is trying to beat anthropic at any cost, and made a training regiment that makes agents manic