Tensorflow import Error

[MayaSakaguchi617]

Issue Type: Bug

Tensorflow is already installed.

import tensorflow as tf

run the next cell
Error: Direct kernel connection broken

Extension version: 2020.12.414227025
VS Code version: Code 1.52.1 (ea3859d4ba2f3e577a159bc91e3074c5d85c0523, 2020-12-16T16:30:02.420Z)
OS version: Darwin x64 20.3.0

System Info

Item	Value
CPUs	Apple M1 (8 x 2400)
GPU Status	2d_canvas: enabled flash_3d: enabled flash_stage3d: enabled flash_stage3d_baseline: enabled gpu_compositing: enabled metal: disabled_off multiple_raster_threads: enabled_on oop_rasterization: enabled opengl: enabled_on protected_video_decode: unavailable_off rasterization: enabled skia_renderer: disabled_off_ok video_decode: enabled webgl: enabled webgl2: enabled
Load (avg)	2, 2, 2
Memory (System)	8.00GB (0.03GB free)
Process Argv	--crash-reporter-id 6d5cec51-b17c-464e-9bbd-a5aad282620e
Screen Reader	no
VM	0%

[rchiodo]

Thanks for the bug. This sounds like importing tensorflow caused the kernel to crash.

Are you using pytorch? That might be the issue if you don't install it according to pytorch's directions?
https://pytorch.org/get-started/locally/

Not really an expert on tensorflow or pytorch though.

[DonJayamanne]

@MayaSakaguchi617
In addition to what @rchiodo has suggested, Please could you also enable logging & provide the output when you get the above error (please try this first before re-installing pytorch).

Please could you enable logging as follows:

• Open settings & go into Jupyter->Logging
• Change the value to debug
• Reload VS Code,
• Attempt to repro the issue & then copy the output from the Jupyter output panel.

@rchiodo would be good to see what errors are returned by kernel, this way when it happens in the future we can proactively ask user from extension to re-install pytorch with some instructions (ie. prevent user from having to report an issue - i.e. provide a way for them to resolve it at their end then and there...).

[MayaSakaguchi617]

I really appreciate your advice.

Since I couldn't find Jupiter->Logging,
I tried to debug by following below.
https://code.visualstudio.com/docs/python/jupyter-support

---Result---
CODE
[ ]:import tensorflow as tf
[ ]:print('hello')

OUTPUT
(base) % source /Users/{username}/opt/anaconda3/bin/activate
(base) % conda activate base
(base) % /Users/{username}/opt/anaconda3/bin/python "{filename}.py"
zsh: illegal hardware instruction /Users/{username}/opt/anaconda3/bin/python
(base) %

[DonJayamanne]

Since I couldn't find Jupiter->Logging,

Open your settings from the command palette & select Preferences: Open Settings UI

The reload VS Code after changing it to debug.
Run your cell & wait for the error Direct Kernel connection broken.
Go into your output panel Jupyter and copy the output & paste it here.
See here

[rchiodo]

Your python example indicates that tensorflow crashes without using VS code.

Sounds like a tensorflow bug. I believe this is the sort of thing that happens if you don't install pytorch using the directions (I had something similar happen to me). Or it could be that tensorflow isn't supported on your chip yet either.

See this issue here:
tensorflow/tensorflow#44999

[MayaSakaguchi617]

@DonJayamanne Thanks for kind instruction.

I've just run the first line: import tensorflow as tf
The output was a little long, so I will attach a text file.

Log_1.txt

[greazer]

Please reopen this bug (or file in the tensorflow repo) if the above advice doesn't resolve your issue. Thanks!