Xla: Can't import torch_xla in colab

[superuser992002]

❓ Questions and Help

VERSION = "1.5" #@param ["1.5" , "20200325", "nightly"]
!curl https://raw.githubusercontent.com/pytorch/xla/master/contrib/scripts/env-setup.py -o pytorch-xla-env-setup.py
!python pytorch-xla-env-setup.py --version $VERSION

import torch_xla
import torch_xla.core.xla_model as xm
device = xm.xla_device()
print(device)

However the colab cant import torch_xla and return this error

ImportError Traceback (most recent call last)
in ()
----> 1 import torch_xla
2 import torch_xla.core.xla_model as xm
3 device = xm.xla_device()
4 print(device)

/usr/local/lib/python3.6/dist-packages/torch_xla/__init__.py in ()
39 import torch
40 from .version import __version__
---> 41 import _XLAC
42
43 _XLAC._initialize_aten_bindings()

ImportError: /usr/local/lib/python3.6/dist-packages/_XLAC.cpython-36m-x86_64-linux-gnu.so: undefined symbol: _ZN2at6native6einsumENSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEEN3c108ArrayRefINS_6TensorEEE

[dlibenzi]

Must be some local issue, as it works for me.
Did you try nightly or some other version?

[btomtom5]

This error happens when there is a mismatch between torch and torch_xla. Are you reinstalling torch again after this script runs?

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[xcdmp21]

I had this error too. I think this happens because after you install it, it doesn't go straight into the notebook environment. I fixed it by restarting the runtime and not running the install again.

[biki321]

I have the same issue , same error while import torch_xla.

[harsh244]

Similar error here
undefined symbol: _ZN2at19slow_conv_dilated3dERKNS_6TensorES2_N3c108ArrayRefIlEES2_S5_S5_S5_

[jysohn23]

@harsh244, looks like you had a version mismatch where torch==1.7.0 but torch_xla==1.6. We've just updated our colab notebooks: https://github.com/pytorch/xla/pull/2596 can you try again installing torch_xla==1.7 like we updated on those notebooks?

[harsh244]

@harsh244, looks like you had a version mismatch where torch==1.7.0 but torch_xla==1.6. We've just updated our colab notebooks: #2596 can you try again installing torch_xla==1.7 like we updated on those notebooks?

@jysohn23 Another error popped up after running the new updated notebook

WARNING:root:TPU has started up successfully with version pytorch-1.6
---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
<ipython-input-14-ebe519c076f6> in <module>()
      3 
      4 # imports the torch_xla package
----> 5 import torch_xla
      6 import torch_xla.core.xla_model as xm

2 frames
/usr/local/lib/python3.6/dist-packages/torch_xla/_patched_functions.py in _patch(fn, newfn)
     11   xfingerprint = inspect.signature(fn)
     12   fingerprint = inspect.signature(newfn)
---> 13   if xfingerprint != fingerprint:
     14     raise RuntimeError(
     15         'Unable to patch {}, signature mismatch: {} vs {}'.format(

RuntimeError: Unable to patch <function clip_grad_norm_ at 0x7f88ba201f28>, signature mismatch: (parameters:Union[torch.Tensor, Iterable[torch.Tensor]], max_norm:float, norm_type:float=2.0) -> torch.Tensor vs (parameters, max_norm, norm_type=2)

Am i missing something here?

[jysohn23]

Hmm that doesn't look right. In your logs it still says:

WARNING:root:TPU has started up successfully with version pytorch-1.6

Can you try restart your runtime and also double check version with something like:

print(torch_xla.__version__)

[harsh244]

Thanks. It works with torch==1.7 . Seems like it was running on torch==1.6 earlier

[bharath5673]

!pip uninstall torch -y
!pip install torch==1.6 

if it appears like.. "WARNING:root:TPU has started up successfully with version pytorch-1.6"

enjoy ...!!