Failed to download pytorch zip
veniamin-ilmer opened this issue · comments
After I run in Windows 10:
export TORCH_CUDA_VERSION=cu113
cargo run --release --bin sample burn SDv1-4 7.5 20 "An ancient mossy stone." img
I get this error:
Compiling torch-sys v0.13.0
error: failed to run custom build command for `torch-sys v0.13.0`
Caused by:
process didn't exit successfully: `D:\dev\stable-diffusion-burn\target\release\build\torch-sys-9321d07214d78619\build-script-build` (exit code: 1)
--- stdout
cargo:rerun-if-env-changed=LIBTORCH_USE_PYTORCH
cargo:rerun-if-env-changed=LIBTORCH
cargo:rerun-if-env-changed=TORCH_CUDA_VERSION
--- stderr
Error: https://download.pytorch.org/libtorch/cu113/libtorch-win-shared-with-deps-2.0.0%2Bcu113.zip: status code 403
When I try to visit that URL, I get AccessDenied error.
Torch is a pain to get setup and I'd rather use wgpu by default as it's much easier to work with. Hopefully in a few days I'll have stable diffusion working with wgpu. Then no one will need to suffer torch.
@veniamin-ilmer would you be kind and refile your issue with tch-rs team? It seems there is no download binary for Windows. This is a tch-rs issue.
Here is the link: https://github.com/LaurentMazare/tch-rs/issues
You can use your original description and title.
@antimora Done - LaurentMazare/tch-rs#781
@Gadersd I attempted to run it via the wgpu feature flag, and disabled the default with torch. After maxing out my RAM (over 10 GB), it ran out of memory. I couldn't figure out how to decrease the amount of RAM it could use.
The binary for cuda 11.3 is available here (https://download.pytorch.org/libtorch/cu113/libtorch-win-shared-with-deps-1.12.1%2Bcu113.zip), but I don't know about torch backwards compatibility regarding older cuda version. Cuda 11.3 was last supported in torch 1.12.1.