Open-Sora - Linux (WSL Install)
Added 2024-07-22 00:03:38 +0000 UTCFREE for all Prairie Dog members!
This installer script is for those of you who feel comfortable working in a Linux (WSL) environment and will get you going within minutes! Since this is installed in your WSL (rather than a windows environment like our Win/WSL install script), it makes getting your videos out into windows as a bit trickers. However, we have added tools so that you can easily see and delete your videos as you make them!
Can be downloaded here:
https://www.patreon.com/posts/open-sora-linux-108568930
For those of you who would prefer to avoid Linux, we still have our advanced Windows/Linux install script that removes all that and sets you up in a Windows explorer. This is an advanced install and is best for avoiding Linux problems:
https://www.patreon.com/posts/open-sora-ease-108408537
Note: this software is heavily resource intensive. Expect to work in 360P most of the time with a single GPU
Comments
Been fighting with and it always fails. I have your script and it fails too. Where it fails right now is this "torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 160.00 MiB. GPU 0 has a total capacity of 4.00 GiB of which 0 bytes is free. Including non-PyTorch memory, this process has 17179869184.00 GiB memory in use. Of the allocated memory 5.12 GiB is allocated by PyTorch, and 124.74 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)" I've been beating my head for hours in how to resolve this to reduce the batch file size. I didn't write the app.py file and am clueless as to what is causing it. Either figure out how to fix the OOM problem or just not load the damn models. Are they really necessary? I've tried changing this line "export PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:32" to a lower number. The lowest it will go is 20 and it still fails. Increasing it still fails.
Blacknoon
2024-09-14 11:16:30 +0000 UTC
