RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!

[MengFoong]

Filtering the images containing characters which are not in opt.character Filtering the images whose label is longer than opt.batch_max_length

dataset_root: all_data opt.select_data: ['all_data'] opt.batch_ratio: ['1']

dataset_root: all_data dataset: all_data all_data/en_sample sub-directory: /en_sample num samples: 882 all_data/rec\test sub-directory: /rec\test num samples: 0 all_data/rec\train sub-directory: /rec\train num samples: 0 all_data/rec\val sub-directory: /rec\val num samples: 0 num total samples of all_data: 882 x 1.0 (total_data_usage_ratio) = 882 num samples of all_data per batch: 10 x 1.0 (batch_ratio) = 10

Total_batch_size: 10 = 10

dataset_root: all_data/en_sample dataset: / all_data/en_sample/ sub-directory: /. num samples: 882

...

continue to train, start_iter: 300000 training time: 11.559250354766846 Output is truncated. View as a scrollable element or open in a text editor. Adjust cell output settings...

RuntimeError Traceback (most recent call last) Cell In[6], line 2 1 opt = get_config("config_files/en_fine_tunning_config.yaml") ----> 2 train(opt, amp=False)

File c:\Users\mengfoong\Desktop\Train_Docling_2\EasyOCR-Trainer\train.py:233, in train(opt, show_number, amp) 230 model.eval() 231 with torch.no_grad(): 232 valid_loss, current_accuracy, current_norm_ED, preds, confidence_score, labels,
--> 233 infer_time, length_of_data = validation(model, criterion, valid_loader, converter, opt, device) 234 model.train() 235 print(infer_time, length_of_data)

File c:\Users\mengfoong\Desktop\Train_Docling_2\EasyOCR-Trainer\test_1.py:45, in validation(model, criterion, evaluation_loader, converter, opt, device) 43 preds_size = torch.IntTensor([preds.size(1)] * batch_size) 44 # permute 'preds' to use CTCloss format ---> 45 cost = criterion(preds.log_softmax(2).permute(1, 0, 2), text_for_loss, preds_size, length_for_loss) 47 if opt.decode == 'greedy': 48 # Select max probabilty (greedy decoding) then decode index to character 49 _, preds_index = preds.max(2)

File c:\Users\mengfoong\Desktop\Train_Docling_2\venv\Lib\site-packages\torch\nn\modules\module.py:1739, in Module._wrapped_call_impl(self, *args, **kwargs) 1737 return self._compiled_call_impl(*args, **kwargs) # type: ignore[misc] 1738 else: ... 3085 _Reduction.get_enum(reduction), 3086 zero_infinity, 3087 )

RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and CPU!

I'm facing this error when I run the cell, anyone can share how you resolved this ?

[SigawathRungrueng]

I'm facing the same problem. Have you fixed it?

[yashithalrao]

yes, same, im abel to run it on cpu, but not gpu

[VicentAvaria]

Has anyone solved this problem? Can you share it?

[mengfoong-dev]

it's been a long time, but I remember I solved by following instructions of using cuda in pytorch that supports GPU. rmb to install driver from cuda also.

It been a long time, I will try my best to help.

[ppaauuoo]

I also got this problem after training around an hour, super time wasting.

[maks1812]

But that problem is easy to solve, just move tensors to same device i.e.

# Calculate evaluation loss for CTC decoder.
            preds_size = torch.IntTensor([preds.size(1)] * batch_size).to(device)
            # permute 'preds' to use CTCloss format
            cost = criterion(preds.log_softmax(2).permute(1, 0, 2), text_for_loss.to(device), preds_size, length_for_loss.to(device))