Replies: 3 comments 8 replies
-
Related to #68 |
Beta Was this translation helpful? Give feedback.
-
To be fair, if the model was on an unfragmented HDD, it's quite possible that it wouldn't always be the bottleneck, HDDs can get like 300mb/s on large files. And for now there hasn't been any hot-swapping with hdd, only with RAM, so shouldn't be too worrysome. |
Beta Was this translation helpful? Give feedback.
-
The thing about "managing" these files is that a) there are about 5+ different methods files (weights) are acquired, and it's not always proper from license standpoint to say "copy" everything to HuggingFace. For example, until recently, some of the model downloads were hosted on Facebook's own servers. What's important to note is also that if you just wanted a "clean install" then generally you would only need to install the core code/dependencies without all of the model weights. Basically that's why the default settings (if you can call them that) are a bit flaky. They have benefits and drawbacks, but this way, it's less likely that you have gigabytes of unneeded files. Finally, to answer the question - how to "start over" - assuming everyone's implementations are sane (SD/text generation etc etc) you can just delete and redownload the caches. It should not break the stuff. |
Beta Was this translation helpful? Give feedback.
-
Like said before, I'm a noob and not that handy with all commands to install certain things using pip etc.
My Windows had some weird issues and since it automatically creates an image of C, I decided to put an image back from a week before.
That was before I actually installed the TTS Generation WebUI which is basically located on drive G but also has a lot of files on C as I noticed. I tried to get them back from yesterdays image but there must be some directories/files somewhere it misses since it no longer works. I get a lot of things like below when I press Generate in in the Bark tab.
Totally my problem but I think it would be nice if the script actually would ask where you would want to put all extra files. So basically at the end you've one directory which has all so it's also clear which files belong to what and I don't ruin a Stable Diffusion thing as well ;)
A lot of files are stored in %appdata%/Python/Python310 I think and all models are in the .cache directory under the user. Especially the .cache directory uses a lot of space which I would like to have on G, together with the other files. That will make loading models much slower for sure but my NVme has a limit capacity on C.
Having all things at one location would make it clear what you can delete if you want to start over (like me) but also it will be easier to move things around. F.e. I also moved my Stable Diffusion to G but that had a few lines in a config file which you could change so you could move the models etc. along with the rest.
Maybe an idea to have some things implemented so it's easier to see what will end up where and change it afterwards or, maybe better, on install.
I tried already installing things again but think it still uses old files somewhere it shouldn't use or it misses some. The Install buttons mentioned at my earlier discussion post don't seem to work in last version and I saw the install button for Demucs Demo was called 'MusicGen', not a biggie ofcourse.
For now all exclamation marks in the tabs are gone and only Bark doesn't seem to work. And, man, how bizar is that Demucs, that's impressive!
Beta Was this translation helpful? Give feedback.
All reactions