Same, I thought it was used commonly too.
Same, I thought it was used commonly too.
It isn’t misusing metric, it just simply isn’t metric at all.
Stop giving them ideas!
If you have ever seen a police interregation, you may notice the detectives ask a question and then, after either no answer or insufficient answer, they will just look at the suspect expectantly. This is done to put phsycological pressure on the suspect to answer the question. Given this info, I would say so, at least in a face to face situation.
Online, I am not so sure. How many posts did you scroll past in the last week on Lemmy that ask a question that you did not answer? How many did you answer? Even if you answered most, you would be in the minority, as if you were not, we would expect far higher engagement rates on posts.
If I’m being honest, it is fairly slow. It takes a good few seconds to respond on a 6800XT using the medium vram option. But that is the price to pay to running ai locally. Of course, a cluster should drastically improve the speed of the model.
You can run llms on text-generation-ui such as open llama and gpt2. It is very similar to the stable diffusion web ui.
It is just how I prefer to do my computing. I tend to live on the command line and pipe programs together to get complex behavior. If you don’t like that, then my approach is not for you and that’s fine. As for your analogy, I see it more as “instead of driving down the road in a car, I like to put my own car together using prefabs”.
Option 4: levy existing tools such as gpg and git using something like pass. That way, you are keeping things simple but it requires more technical knowledge. Depending on your threat model, you may want to invest in a hardware security key such as a yubikey which works well with both gpg and ssh.
I am aware what fine tuning is. It is available from the train tab while the base checkpoint is loaded in both cases.
Yes, definitely. My biggest use is transparent filesystem compression, so I completely agree!
Always trust user input’); DROP TABLE users;
There are some uis that allow for fine tuning (assuming you have an extremely high end rig designed for ml). For example ChatGPT alternative and DALLE alternative.
That sounds pretty nice, especially if you added onions to that! Might have to give that a go at some point.
Well when using zstd, you tar first, something like tar -I zstd -cf my_tar.tar.zst my_files/*
. You almost never call zstd directly and always use some kind of wrapper.
That threw me for a loop!
I’ve found that they commonly mislabel apps as FLOSS when they are actually proprietary. That being said, I only ever used it for more niche programs, so maybe common programs are more accurate.
Can confirm, strawberry is great!
Honestly, I think biting the bullet and trying your hand at coding will be worth your time. Visual scripting typically fall into two camps: code but it is visual (in this case it is just slower, more cumbersome and harder to read) or limiting (this may be fine depending on your needs, but you may also outgrow it). A middleground could be coding where the vast majority is done for you. For example in Godot, there are many nodes that are fully built and just need your custom settings. There are even freely available nodes in the asset store if you need more. Then, if you need some behaviour that does not yet exist, you just code that little part, which will be a great learning experience in of itself.
My biggest tip though, regardless of the approach you take, is keep it simple. Your first game should be ridiculously simple. For example, the first game I made was a 2d scroller spaceship shooter where there were only asteroids as “enemies”. I could then add onto that to test my coding skills, and eventually it was fairly fun, even if it had simple roots.
I used to have a similar setup but had it stream directly into MPV using the ytdl hook. Do you have it download the videos into a cache automatically and then load from file later and if so, how did you set that up?