canpan

5 ms·

31.
by canpan 5mo ago
Recent models (Qwen 3.6 and Gemma) can really do coding locally. Feels like SOTA from maybe a year ago? But you would want about 32-40GB total memory. 24GB is just a bit short of that. A gaming PC with 16GB graphics card and 32GB RAM brings
32.
by canpan 5mo ago
I am sure ads can work on me, and the HN crowd, if I was targeted. But there are much easier targets that generate more income. So companies spend more on gambling etc ads, where they can make more money.
33.
by canpan 5mo ago
Ads target a specific sub group. And I think many HNlers are not part of that group. My guess is that ads online target people with more impulsive buying. Even if you showed me the perfect ad, I probably would not buy it. Because if I need
34.
by canpan 5mo ago
It's not great, but basically if your lock is better than the lock on the bicycle next to yours, they will most likely not steal yours..
35.
by canpan 5mo ago
I feel bad for the people working on it, but looking at the selected metric "60% faster", I don't think it is enough to turn the tides. The correct metric would be something like "opens within one frame at 60fps". B
36.
by canpan 5mo ago
I wondered how is training data balanced? If you put in to much Wikipedia, and your model sounds like a walking encyclopedia? After doing the Karpathy tutorials I tried to train my AI on tiny stories dataset. Soon I noticed that my AI was a
37.
by canpan 5mo ago
Just ran llama-bench at home with the similar priced AMD AI PRO R9700 32G. The phoronix numbers look extremely low? Probably I misunderstand their test bench. Anyway, here are some numbers. Maybe someone with access to a B70 can post a comp
38.
by canpan 5mo ago
More RAM running slower is still true today. With AM5 you probably cannot enable EXPO with four ram slots filled vs two. The gap is not that extreme though. https://www.corsair.com/us/en/explorer/diy-builder&#
39.
by canpan 5mo ago
In Tokyo many bicycle lanes are pretty useless for this reason. Cars are parking every 20m making them absolutely inaccessible. Then there is the bicycle lane between Asakusa and Ueno, which is separated from the street, but made like some
40.
by canpan 6mo ago
While that would be cool, something like the 7400 series, is already pretty close to scratching that itch. And a lot less dangerous.
41.
by canpan 6mo ago
I would love to buy one and know a bunch of others. Wonder if shipping to Japan is in the works?
42.
by canpan 6mo ago
You don't need it, but as someone who has been there: For me making a 3D engine is a lot of fun! But then I never finish the actual game. So if you actually want to ship a game, I recommend using an engine. Personally I prefer Unreal.
43.
by canpan 6mo ago
122b would be awesome. It is the largest size you can kinda run with a beefy consumer PC. I wondered about gemma stopping in the 30b category, it is already very strong. 122b might have been too close to being really useful.
44.
by canpan 6mo ago
Not OP, but I ran 122b successfully with normal RAM offloading. You dont need all that much VRAM, which is super expensive. I used 96gb ram + 16gb vram gpu. But it's not very fast in that setup, maybe 15 token per second. Still, you ca
45.
by canpan 6mo ago
Any good gaming pc can run the 35b-a3 model. Llama cpp with ram offloading. A high end gaming PC can run it at higher speeds. For your 122b, you need a lot of memory, which is expensive now. And it will be much slower as you need to use mos
46.
by canpan 6mo ago
Here in Japan they started to forward me to the app page when ordering. So you are forced to use the app with a mobile browser. Even though the website could do it perfectly fine in the past. I do not go often, but if I do I prefer to sit,
47.
by canpan 6mo ago
Thanks for the answer. Brings my hope up. Looking in my local shops, I can get 3 cards for the price of one 5090. May I ask, what kind of tok/s you are getting with the r9700? I assume you got it fully in vram?
48.
by canpan 6mo ago
I was thinking to get 2x r9700 for a home workstation (mostly inference). It is much cheaper than a similar nvidia build. But still not sure if good value or more trouble.
49.
by canpan 6mo ago
The Keyboard Speciality shop you linked is great. It is always surprisingly crowded for such a niche topic. The have good test boards with different key switches. Note that in the area are quite a few other good shops. Tsukumo has various h
50.
by canpan 6mo ago
I wonder if the winning game becomes your own boss and tiny companies. I want to do the jump, but lack of courage, good ideas, sales skills and a very good salary still holding me back (open for suggestions). But if the very good salary wou
51.
by canpan 6mo ago
Can you share some other signs you think may indicate it rising as a powerhouse? Living in Japan, I am interested what others see. Regarding immigration, Japan is actually making it a lot stricter now. Not sure how that will play out.
52.
by canpan 6mo ago
I am playing around with this at home right now. I think a lot of the latest improvements came with the harness, instead of AI. The part I am working on is to have better tools and data to search over. Curated for my needs. Similar to the K
53.
by canpan 6mo ago
I went to a chain Family Restaurant recently here in Japan. The food is brought by a robot for a while now. Recently you get your seat selected at a touchscreen. You can pay at your table's tablet using PayPay. There is still some wait
54.
by canpan 6mo ago
Things are slow, aren't they? I feel there was a lot of less lag in old operating systems and software. I use two editors now. VS Code as full IDE when I want to code heavily. And a homemade FLTK based editor with just basic syntax col
55.
by canpan 6mo ago
Same. Looking at your username, guessing, you are accessing from Japan too.
56.
by canpan 6mo ago
Patent free video is in a strange space. I recently looked into just using old formats. To be super safe. With audio we have mp3 which is good enough for a lot of cases and seems to be patent free. But for mpeg2 (used on DVD) even though it
57.
by canpan 6mo ago
Wasn't there also patent issues with early texture compression schemes? DXT1? Adding a delay of 20 or so years.
58.
by canpan 6mo ago
I am missing this part too. I can't really say ever having a problem upgrading go to the latest version. Now with "go fix", a lot of features are even improved automatically.
59.
by canpan 6mo ago
I did not think of that. Maybe for some industries, it might make sense. But if I want a regression test, I would probably set it up as automated test. In the case I mentioned above it was the only test beside my own for a new service.
60.
by canpan 6mo ago
Yes, QA is important. My code will always "work" in that everything I tested is bug free. But having someone other test, especially someone who knows the service is gold. But there is also bad QA: The most worthless QA I was force

More ›