E2FGVI-HQ Inpainting engine cannot be used

[backspace696]

After installing everything (following the guide) the program seems to be unable to find GPU (RTX 4060 Ti) and always falls back to CPU
E2FGVI-HQ in this case can't be used and throws the error
«Failed to load model: module 'torch._C' has no attribute '_cuda_resetPeakMemoryStats'»

The console itself show this:
2025-12-11 13:59:53.326 | WARNING | src.demark_world.models.model.modules.feat_prop::14 - mmcv is not available, using a fallback implementation

i tried installing mmcv, and it is installed, yet it's still falling back to CPU only.

Is there something else needs to be installed, that doesn't mentioned in the readme?

[backspace696]

UPD
It seems i find the fix
Turns out that by default (?) uv sync installs everything needed, but it's going for the 2.9.1+cpu version of the PyTorch

I manually removed torch packages from the virtual environment (probably a janky solution) and then reinstall torch, but with the support for CUDA. After that it worked

«mmcv is not available, using a fallback implementation» error still showing up though

[linkedlist771]

Sorry about the late response. I have been a bit busy recently. I will check this issue when available.

[backspace696]

Sorry about the late response. I have been a bit busy recently. I will check this issue when available.

I did fix it: #2 (comment)
it just seems to download non-CUDA version of the PyTorch, if you replace it with the CUDA one it works

[linkedlist771]

As noted in the official uv PyTorch integration guide(https://docs.astral.sh/uv/guides/integration/pytorch/#installing-pytorch), the default Linux resolution is GPU-enabled. In my uv.lock, only Linux resolves to the GPU-enabled torch wheel, while Windows and macOS resolve to CPU-only wheels.(https://github.com/linkedlist771/DeMark-World/blob/main/uv.lock)

wheels = [
  { url = "https://files.pythonhosted.org/packages/0f/27/07c645c7673e73e53ded71705045d6cb5bae94c4b021b03aa8d03eee90ab/torch-2.9.1-cp312-cp312-manylinux_2_28_aarch64.whl", hash = "sha256:da5f6f4d7f4940a173e5572791af238cb0b9e21b1aab592bd8b26da4c99f1cd6", size = 104126592, "upload-time" = "2025-11-12T15:20:41.62Z" },
  { url = "https://files.pythonhosted.org/packages/19/17/e377a460603132b00760511299fceba4102bd95db1a0ee788da21298ccff/torch-2.9.1-cp312-cp312-manylinux_2_28_x86_64.whl", hash = "sha256:27331cd902fb4322252657f3902adf1c4f6acad9dcad81d8df3ae14c7c4f07c4", size = 899742281, "upload-time" = "2025-11-12T15:22:17.602Z" },
  { url = "https://files.pythonhosted.org/packages/b1/1a/64f5769025db846a82567fa5b7d21dba4558a7234ee631712ee4771c436c/torch-2.9.1-cp312-cp312-win_amd64.whl", hash = "sha256:81a285002d7b8cfd3fdf1b98aa8df138d41f1a8334fd9ea37511517cedf43083", size = 110940568, "upload-time" = "2025-11-12T15:21:18.689Z" },
  { url = "https://files.pythonhosted.org/packages/6e/ab/07739fd776618e5882661d04c43f5b5586323e2f6a2d7d84aac20d8f20bd/torch-2.9.1-cp312-none-macosx_11_0_arm64.whl", hash = "sha256:c0d25d1d8e531b8343bea0ed811d5d528958f1dcbd37e7245bc686273177ad7e", size = 74479191, "upload-time" = "2025-11-12T15:21:25.816Z" },
]

[linkedlist771]

UPD It seems i find the fix Turns out that by default (?) uv sync installs everything needed, but it's going for the 2.9.1+cpu version of the PyTorch

I manually removed torch packages from the virtual environment (probably a janky solution) and then reinstall torch, but with the support for CUDA. After that it worked

«mmcv is not available, using a fallback implementation» error still showing up though

As for the mmcv issue, I found this library is no longer actively maintained and doesn't have a pre-built version for torch 2+ versions. I have a manually implemented version with a warning log, which I suppose is OK, but it may not be that effective.

[sgtpepper67]

Could you update it so that the high fidelity model works on Windows?

[linkedlist771]

It should work on Windows if you have CUDA and the CUDA driver. Could you share what problem you need when use the high fidelity model?

[sgtpepper67]

Same error as OP: Failed to load model: module 'torch._C' has no attribute '_cuda_resetPeakMemoryStats'»

I have CUDA 13.1 installed and am using the default list of packages in the UV.lock file.