Operator is not supported in mobile module.

[nnick14]

🐛 Bug

When converting a TorchScript model to mobile, if the model contains unsupported mobile ops, a vague error is returned. It'd be nice if there was a more clear description of the exact op code error.

To Reproduce

Steps to reproduce the behavior:

Run the following code with an optimized_scripted_module containing a non-supported mobile operator.

optimized_scripted_module._save_for_lite_interpreter("<output_name>.ptl")

Expected behavior

The RuntimeError is not exactly clear. It would be helpful to either include a link to the FORALL_OPCODES or return the specific OPCODE information. Maybe something like:

RuntimeError: ENTER (enter scope of a contextmanager) is not supported in mobile module.

Instead of what's currently returned:

RuntimeError: ENTER is not supported in mobile module.

Environment

• PyTorch version: 1.9.1+cu111
• OS: Ubuntu 18.04.6 LTS (x86_64)
• Python version: 3.8.0, pip
• CUDA: 11.1

[zabir-nabil]

Same issue with pytorch torch-1.11.0.dev20220108+cpu (nightly) on windows.

RecursiveScriptModule(original_name=MainModel)
Traceback (most recent call last):
  File "melspec_test.py", line 90, in <module>
    optimized_model._save_for_lite_interpreter("vgg_sre.ptl")
  File "C:\ProgramData\Anaconda3\envs\py38\lib\site-packages\torch\jit\_script.py", line 707, in _save_for_lite_interpreter
    return self._c._save_for_mobile(*args, **kwargs)
RuntimeError: ENTER is not supported in mobile module.

[VasenkovArtem]

hi
i don't know if this issue is still actual for you but it can be helpful for others who has the same problem
i got this message
Traceback (most recent call last): File "C:/Users/Артём/Downloads/image_captioning_translate/checking_new.py", line 67, in <module> optimized_scripted_module_full._save_for_lite_interpreter("full_model_29-05.ptl") File "C:\Program Files\Python39\lib\site-packages\torch\jit\_script.py", line 707, in _save_for_lite_interpreter return self._c._save_for_mobile(*args, **kwargs) RuntimeError: ENTER is not supported in mobile module.
solution was to delete strings with torch.no_grad(): so i moved the code from these blocks to the main block:
trg_tensor = torch.tensor(trg_indexes).unsqueeze(0).to(self.device)
trg_mask = self.translater.make_trg_mask(trg_tensor)
with torch.no_grad():
output = self.translater.decoder(trg_tensor, enc_src, trg_mask, src_mask)
pred_token = output.argmax(2)[:,-1].item()
trg_indexes.append(pred_token)
became
trg_tensor = torch.tensor(trg_indexes).unsqueeze(0).to(self.device)
trg_mask = self.translater.make_trg_mask(trg_tensor)
output = self.translater.decoder(trg_tensor, enc_src, trg_mask, src_mask)
pred_token = output.argmax(2)[:,-1].item()
trg_indexes.append(pred_token)
this action helped me not to get such problem again
sorry for my english )