V0.2.5.0 Seems don't support to use with cpu anymore.

[k79e]

It seems it used a function that not support to run with cpu only.
pytorch/pytorch#74625

Error "slow_conv2d_cpu" not implemented for 'Half'
        Tile 1/12
Traceback (most recent call last):
  File "inference_realesrgan.py", line 124, in <module>
    main()
  File "inference_realesrgan.py", line 105, in main
    output, _ = upsampler.enhance(img, outscale=args.outscale)
  File "D:\Program1\Python38\lib\site-packages\torch\autograd\grad_mode.py", lin
e 27, in decorate_context
    return func(*args, **kwargs)
  File "E:\Real-ESRGAN-master\realesrgan\utils.py", line 188, in enhance
    self.tile_process()
  File "E:\Real-ESRGAN-master\realesrgan\utils.py", line 146, in tile_process
    output_start_x:output_end_x] = output_tile[:, :, output_start_y_tile:output_
end_y_tile,
UnboundLocalError: local variable 'output_tile' referenced before assignment

[xinntao]

please add --fp32 option to your command

[k79e]

It works!!

[mudit514]

Yes, it worked.

[nmunisa]

please add --fp32 option to your command

how can I add this? please help me

learn.lr_find() --fp32
learn.recorder.plot()

#output:
LR Finder is complete, type {learner_name}.recorder.plot() to see the graph.

RuntimeError Traceback (most recent call last)
in ()
----> 1 learn.lr_find()
2 learn.recorder.plot()

10 frames
/usr/local/lib/python3.7/dist-packages/torch/nn/modules/conv.py in _conv_forward(self, input, weight, bias)
442 _pair(0), self.dilation, self.groups)
443 return F.conv2d(input, weight, bias, self.stride,
--> 444 self.padding, self.dilation, self.groups)
445
446 def forward(self, input: Tensor) -> Tensor:

RuntimeError: "slow_conv2d_cpu" not implemented for 'Half'

[WASasquatch]

How can this be implemented for something like MindsEye Beta? I feel this should be default behavior, to fallback to CPU, not require manual input.

[RifkyA911]

Author

it will push CPU only, how to work with GPU too?