Ran into issue with running the demo due to unsupported autocast device_type 'cuda:0' ! Can you tell me how to resolve this ?

PHOENIXFURY007 commented 4 months ago

Hello, Congrats on this work ! I am new to this field and ran into this error. Can you tell me how to resolve this issue ?

(roma) [s20225561@pleiades3 demo]$ python demo_fundamental.py xFormers not available xFormers not available /home2/s20225561/.conda/envs/roma/lib/python3.10/site-packages/torchvision/models/_utils.py:208: UserWarning: The parameter 'pretrained' is deprecated since 0.13 and may be removed in the future, please use 'weights' instead. warnings.warn( /home2/s20225561/.conda/envs/roma/lib/python3.10/site-packages/torchvision/models/_utils.py:223: UserWarning: Arguments other than a weight enum orNonefor 'weights' are deprecated since 0.13 and may be removed in the future. The current behavior is equivalent to passingweights=None. warnings.warn(msg) Using coarse resolution (560, 560), and upsample res (864, 864) Traceback (most recent call last): File "/home4/playground/SiggraphAsia/RoMa/demo/demo_fundamental.py", line 27, in <module> warp, certainty = roma_model.match(im1_path, im2_path, device=device) File "/home2/s20225561/.conda/envs/roma/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context return func(*args, **kwargs) File "/home4/playground/SiggraphAsia/RoMa/romatch/models/matcher.py", line 655, in match corresps = self.forward_symmetric(batch) File "/home4/playground/SiggraphAsia/RoMa/romatch/models/matcher.py", line 516, in forward_symmetric feature_pyramid = self.extract_backbone_features(batch, batched = batched, upsample = upsample) File "/home4/playground/SiggraphAsia/RoMa/romatch/models/matcher.py", line 462, in extract_backbone_features feature_pyramid = self.encoder(X, upsample = upsample) File "/home2/s20225561/.conda/envs/roma/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl return self._call_impl(*args, **kwargs) File "/home2/s20225561/.conda/envs/roma/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl return forward_call(*args, **kwargs) File "/home4/playground/SiggraphAsia/RoMa/romatch/models/encoders.py", line 112, in forward feature_pyramid = self.cnn(x) File "/home2/s20225561/.conda/envs/roma/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl return self._call_impl(*args, **kwargs) File "/home2/s20225561/.conda/envs/roma/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl return forward_call(*args, **kwargs) File "/home4/playground/SiggraphAsia/RoMa/romatch/models/encoders.py", line 70, in forward with torch.autocast(device_type=autocast_device, enabled=autocast_enabled, dtype = autocast_dtype): File "/home2/s20225561/.conda/envs/roma/lib/python3.10/site-packages/torch/amp/autocast_mode.py", line 241, in __init__ raise RuntimeError( RuntimeError: User specified an unsupported autocast device_type 'cuda:0'

+---------------------------------------------------------------------------------------+ | Processes: | | GPU GI CI PID Type Process name GPU Memory | | ID ID Usage | |=======================================================================================| | No running processes found | +---------------------------------------------------------------------------------------+`

Parskatt commented 4 months ago

Hi,

We recently had an pr for autocast, perhaps it introduced something. I'll have a look.

Parskatt commented 4 months ago

@PHOENIXFURY007 Should now be fixed. Could you try pulling latest changes and rerunning?

Parskatt / RoMa

Ran into issue with running the demo due to unsupported autocast device_type 'cuda:0' ! Can you tell me how to resolve this ? #56