cypress66

10 ms·

61.
by cypress66 3y ago
Supply is limited (for something like bitcoin), cheaper, faster settlement and less burocracy when sending money abroad (instead of the awful SWIFT)
62.
by cypress66 3y ago
I personally find stereo/3d video basically no more immersive than regular video. I wouldn't bother putting a headset just for that. It needs to be like VR 360 video where you can look around to be worth it.
63.
by cypress66 3y ago
Luckily Steam is a private company, controlled by gaben. If it were public it would have been enshitified years ago.
64.
by cypress66 3y ago
Even if bounds checking made the decoder 10x slower, would that even matter, outside of low end mobile devices? How many milliseconds are spent decoding images in your average website anyway?
65.
by cypress66 3y ago
This is not that uncommon. The miner signs a tx with that insane fee, but intentionally doesn't broadcast it (so nobody can include that tx but themselves). When he finds a block, that tx is included and he's basically paying hims
66.
by cypress66 3y ago
Instead of all this cookie mess, they could have simply required websites to respect the do not track header.
67.
by cypress66 3y ago
Ooof. I'd expect this to cost like 5 bucks on runpod using a single 3090. I use axolotl for training, I didn't check your notebook but axolotl likely comes with more optimized defaults for speed and vram than what you're doin
68.
by cypress66 3y ago
Llama 7B is quite dumb. Using the 13B you'd get significantly better results, and you can train a qlora on a single 3090 (I think even less is possible but not sure)
69.
by cypress66 3y ago
Is it using the correct prompt format for the different models? You should show exactly the string that was sent to the LLM.
70.
by cypress66 3y ago
You should add what version of the model you are testing For example you mention Jon Durbin Airoboros L2 70B But is it 1.4? 2.0? 2.1? Etc.
71.
by cypress66 3y ago
The comments summary needs to be bullet points. And both summaries should be shorter.
72.
by cypress66 3y ago
Llama2 chat uses RHLF. It's generally for making the models "safer", not smarter (in fact it usually makes them dumber)
73.
by cypress66 3y ago
8bit is virtually lossless. Not much point in running fp16
74.
by cypress66 3y ago
Not really. That's called few shot learner. It's basically unrelated to what happens during training, which is using gradients.
75.
by cypress66 3y ago
It's a bit amusing how people treat chinchilla scaling laws as a law of nature, when it's just about a certain architecture and dataset.
76.
by cypress66 3y ago
1.1B with 3T tokens will never be comparable to 7B with 2T tokens. And I'm not sure what you mean by inference latency being infeasible. Most people using thsss models at home don't even bother with the 7B and go straight to 13B b
77.
by cypress66 3y ago
Wow, vscode extensions having drm is insane
78.
by cypress66 3y ago
Wish I could get rid of all points of interest (hotels, restaurants, etc). Just show me the streets.
79.
by cypress66 3y ago
Just because other's are awful doesn't make it alright.
80.
by cypress66 3y ago
Just make all rooms private and build more rooms or hospitals. It's not rocket science. Relative to the already absurdly high health care costs, the construction costs should be pretty small.
81.
by cypress66 3y ago
I'd actually come to the opposite conclusion here. Unless you think this violation of privacy is OK?
82.
by cypress66 3y ago
Then use llama?
83.
by cypress66 3y ago
Side note, but Wikipedia editors decided to censor KiwiFarms url, which I think is very interesting.
84.
by cypress66 3y ago
Has the owner provided any proof of such interaction at the service center, other than his words? A simple video of the whole car would cast all doubts away.
85.
by cypress66 3y ago
Has the car owner shown a full video of the car? Are we sure we're not just looking at a crashed car? It'd be nice for the article to include a video like that so there's no doubt.
86.
by cypress66 3y ago
Would this actually work?
87.
by cypress66 3y ago
Well, those "actual tests" clearly don't reflect reality. This is obvious if you actually use whisper.
88.
by cypress66 3y ago
I am skeptical on many of those. Speech recognition is not even close to human level. Whisper, and whatever Google uses will make a lot of mistakes on audio files that are trivial to any native speaker.
89.
by cypress66 3y ago
Falcon isn't really comparable to the SLS, but yes the SLS is awful cost wise. Starship will almost surely be between one and two orders of magnitude cheaper per launch vs SLS.
90.
by cypress66 3y ago
Github wasn't abused to train LLMs. Whatever website existed that openly hosted OSS repositories would have been scrapped if it had such a large number of repos. You don't need to be Microsoft to train on github.

More ›