Mert

No worries! It's great that you're reporting on this so we know about it and can fix it. I think the best option would be to open an issue in...

According to OP, this command is problematic: ``` ffmpeg -hwaccel rkmpp -hwaccel_output_format drm_prime -afbc rga -noautorotate -i upload/library/Kirk/2024/2024-11-09/PXL_20241109_133516985.mp4 -y -c:v hevc_rkmpp -c:a aac -movflags faststart -fps_mode passthrough -map 0:2 -map...

Thanks for the report, but I don't think this is the same issue. The issue in question should be specific to RKMPP and (I think) fixed by [this](https://github.com/jellyfin/jellyfin-ffmpeg/pull/498) PR. Please...

No, there are a few more things that would need to happen to let you switch without resetting everything. But it's a big step toward that.

Do you know how to make the PR label validation check happy?

Just FYI this is saying it couldn't allocate the VRAM it requested. If your GPU doesn't have enough free VRAM, this can also be the cause of the error.

I think the issue you raise is valid (the edited image has incorrect metadata), but it's different from the topic of the issue. You can make a separate issue for...

I was under the impression that it actually is a JPEG file, but the metadata is from the original PNG file. On second thought, exiftool is probably right that this...

Hmm, the line `torch.backends.mha.set_fastpath_enabled(False)` should avoid this error. Also, make sure you’re using the Conda/Mamba environment in that folder to have the right dependency versions.

Glad to hear it worked! The fast path is used in eval mode when gradients are disabled. This path isn't supported by PyTorch's ONNX exporter, so it needs to be...