First, while the app works without the newer Flash Attention (attached in Releases), I guess it should work faster with it, right?
But, how does one actually install it without breaking the app? Sure, I'm not the best in a Python environment, but some instructions wouldn't hurt.
First, I had to download/install the app 3 times (each install weighing 40GB!), because downloading files from HuggingFace couldn't be completed for some weird reason, and now after doing "pip install" on that newer Flash Attention 2.7.4 WHL file, I can't even run the app because of some nasty dependency conflicts, so I'm downloading it for the 4th time :(
Btw, in the middle of the pip update, there was a message to try it with "--force-reinstall", which again resulted in some more errors.
I'm using Pinokio on Windows 11. Any pointers on how I should install it, and potentially speed up the generation process? Right now, "by default", it needs 181s for a 90-second-long demo song on my RTX 4080 - so is that slow or fast, even without the newer Flash Attention?
First, while the app works without the newer Flash Attention (attached in Releases), I guess it should work faster with it, right?
But, how does one actually install it without breaking the app? Sure, I'm not the best in a Python environment, but some instructions wouldn't hurt.
First, I had to download/install the app 3 times (each install weighing 40GB!), because downloading files from HuggingFace couldn't be completed for some weird reason, and now after doing "pip install" on that newer Flash Attention 2.7.4 WHL file, I can't even run the app because of some nasty dependency conflicts, so I'm downloading it for the 4th time :(
Btw, in the middle of the pip update, there was a message to try it with "--force-reinstall", which again resulted in some more errors.
I'm using Pinokio on Windows 11. Any pointers on how I should install it, and potentially speed up the generation process? Right now, "by default", it needs 181s for a 90-second-long demo song on my RTX 4080 - so is that slow or fast, even without the newer Flash Attention?