nodja

6 ms·

61.
by nodja 11mo ago
All they have to do is market the fact you don't have to pay for online. PS5 + 3 years of PS Plus = $740 Steam Machine = $700 Add/remove more years of PS Plus if the SM turns out to be more/less expensive. If you add the fact
62.
by nodja 11mo ago
Any model that does thinking inside <think></think> style tokens before it answers. This can be done with finetuning/RL using an existing pre-formatted dataset, or format based RL where the model is rewarded for both a
63.
by nodja 11mo ago
I've never seen a $300 one but I've seen $70 ones. I don't think they're nefarious in that sense, but these boxes are usually scams. They come preloaded with a pirate iptv service that only works for 1-2 months then they
64.
by nodja 1y ago
It's on the huggingface readme https://huggingface.co/utter-project/EuroLLM-9B#results https://huggingface.co/utter-project/EuroLLM-9B#english
65.
by nodja 1y ago
GPT3 existed 5 years ago, and the trajectory was set with the transformers paper. Everything from the transformer paper to GPT3 was pretty much speculated in the paper, it just took people spending the effort and compute to make it reality.
66.
by nodja 1y ago
Why create the markov text server side? If the bots are running javascript just have their client generate it.
67.
by nodja 1y ago
I think another easy improvement to this diffusion model would be for the logprobs to also affect the chance of a token being turned into a mask. So higher confidence tokens should have less of a chance to be pruned, should converge faster.
68.
by nodja 1y ago
That's not true, you could just have looked at the first gif animation in the OP and seen that tokens disappear, the only part that stays untouched is the prompt, adding noise is part of the diffusion process and the code that does it
69.
by nodja 1y ago
The article is cherry picking data points to make a clickbait headline. Why is this being posted here?
70.
by nodja 1y ago
Couldn't you kept it in and labeled it as bots? i.e. using stacked barcharts and such.
71.
by nodja 1y ago
For people reading this that are worried, .com and .net domains are price capped and while the price may rise, it's regulated directly by the ICANN. If you're paying more than that then either your registrar is not following ICANN
72.
by nodja 1y ago
Yeah from memory on-prem was always cheaper, it just removed a lot of logistic obstacles and made everything convenient under one bill. IIRC the wisdom of the time cloud started becoming popular was to always be on-prem and use cloud to sca
73.
by nodja 1y ago
I don't think Hetzner provides locations in SF. Those 100GBit connections don't do much if they need to connect outside the city the rest of the equipment is in, but maybe peering has gotten better and my views are outdated.
74.
by nodja 1y ago
Did you not look at the link I provided? I stopped watching him completely around the time of the intel dGPU release. He would show leaked roadmaps of intel's dGPU launch with Celestial and Druid on there, but the video would be him ba
75.
by nodja 1y ago
Intel Arc - Intel's dedicated GPUs, each GPU generation has a name in alphabetical order, names are taken from nerd culture. Alchemist - First gen GPUs A310 GPUs are the low end, A770 are the high end. Powerful hardware for cheap, very
76.
by nodja 1y ago
They're a market entry point. CUDA became popular not because it was good, but because it was accessible. If you need to spend $10k minimum on hardware just to test the waters of what you're trying to do, that's a lot to thin
77.
by nodja 1y ago
They had videos saying intel was gonna cancel the dGPU division and focus on datacenter pretty much since the intel cards came out. Amongst many other things they've said. I used to follow them too, but they speak with too much confide
78.
by nodja 1y ago
Glad to see that there are laptops that don't suffer like this. But I think the combo of having a steam deck + business laptop beats buying a gaming laptop. Assuming you already own a gaming rig at home.
79.
by nodja 1y ago
I've owned 2 gaming laptops in my lifetime and both had similar issues that were never fixed. One was the first gen Alienware M17 with two GTX 270M GPUs (yes two) and an onboard nvidia GPU whose specific model I can't remember. Th
80.
by nodja 1y ago
For my homelab: portable state. I don't use this image specifically but I use many others. I put docker-compose files in ~/configdata/_docker/ The docker-compose files always mount volumes inside the ~/configdata&#x
81.
by nodja 1y ago
There's a whole subsection of app devs that will just stop making apps for android. Getting graphene or a chinese phone with android won't mean anything because all you will be running is old version of apps since there will be ve
82.
by nodja 1y ago
Fixing a bad habit is very hard, and I clearly stated it that outputting is very helpful, but you need to be constantly corrected or you'll develop bad habits that are very hard to fix. I'm not a native english speaker and I'
83.
by nodja 1y ago
I've used several LLMs to do translations and they're very hit/miss, specially in very high context languages like japanese. I'm not sure recommending their usage for a beginner is good advice, it's better than noth
84.
by nodja 1y ago
This is actually NOT recommended for a beginner. Writing and speaking are effective at establishing long term memories, it's why we do it for other things, but a language learning beginner has no idea if what they're writing makes
85.
by nodja 1y ago
It's much more than that. It's an app that patches apks and has a series of community patches for specific apps. For youtube it's the usual ad blocking and sponsorblock, etc. but it can apply patches for all apps, one of the
86.
by nodja 1y ago
Go back in time and post with em—dashes.
87.
by nodja 1y ago
Loading here refers to loading from VRAM to the GPUs core cache, loading from VRAM is extremely slow in terms of GPU time that GPU cores end up idle most of the time just waiting for more data to come in.
88.
by nodja 1y ago
Yeah chatgpt pretty much nailed it.
89.
by nodja 1y ago
This is the real answer, I don't know what people above are even discussing when batching is the biggest reduction in costs. If it costs say $50k to serve one request, with batching is also costs $50k to serve 100 at the same time with
90.
by nodja 1y ago
Back in the GPT3 days people said that prompt engineering was going to be dead due to prompt tuning. And here we are 2 major versions later and I've yet to see it in production. I thought it would be useful not only to prevent leaks li

More ›