Newer version of PyTorch enforces strict device consistency for all tensors and indices

[linhoangce]

PyTorch now throws an error for this code:

`def sample(model, starting_str,
len_generated_text=500,
scale_factor=1.0):

encoded_input = torch.tensor([char2int[s] for s in starting_str])
encoded_input = torch.reshape(encoded_input, (1, -1))

generated_str = starting_str

model.eval()
hidden, cell = model.init_hidden(1)
hidden = hidden.to(device)
cell = cell.to(device)
for c in range(len(starting_str)-1):
    _, hidden, cell = model(encoded_input[:, c].view(1), hidden, cell) 

last_char = encoded_input[:, -1]
for i in range(len_generated_text):
    logits, hidden, cell = model(last_char.view(1), hidden, cell) 
    logits = torch.squeeze(logits, 0)
    scaled_logits = logits * scale_factor
    m = Categorical(logits=scaled_logits)
    last_char = m.sample()
    generated_str += str(char_array[last_char])
    
return generated_str

torch.manual_seed(1)
model.to('cpu')
print(sample(model, starting_str='The island'))`

Issue comes from encoded_input[:, c].view(1) and last_char.view(1) being not on the same device and the model. Same for hidden and cell, but these have been corrected.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB