Difference between revisions of "Known problems"
(Update with information about IBM WMLA conda channel)
|Line 1:||Line 1:|
== The bhist command is not working ==
== The bhist command is not working ==
Revision as of 15:28, 17 June 2020
- 1 Where did Tensorflow go?
- 2 The bhist command is not working
- 3 Where is my job output file?
- 4 Jupyter notebooks or other programs fail trying to access a /run directory
- 5 Browser fails to connect to Jupyter Notebooks
- 6 Cannot Install PyTorch dependencies
- 7 Cannot use Caffe on login node or compute nodes without GPUs
- 8 Cannot see GPUs in an LSF job
- 9 Nested anaconda environments may cause strange behaviour
Where did Tensorflow go?
On June 26 we will update the GPU compute nodes to a new version of IBM Watson Machine Learning Accelerator. This will change the way you access deep learning packages like Tensorflow and Pytorch. Instead of "activating" these packages, you will be able to install new versions directly in your anaconda environment.
We are actively updating the wiki documentation to explain the new method of accessing deep learning packages. Please bear with us during these updates as some documentation may still refer to the old method of "activating" deep learning packages
The bhist command is not working
We are aware of a problem causing the bhist command to not find the job log file necessary to print information about completed jobs. Instead, the output is always "no matching job found" even if the user specifies the "-a" option to print information about all running, completed, and failed jobs.
While we work on solving this issue you can manually specify the location of the log file using the -f option. For example, the following command will print all running, completed, and failed jobs for the current user:
bhist -f /lsfshare/lsfswg/lsf/work/DeepSenseLSFCluster/logdir/lsb.events -a
Where is my job output file?
Output may not be written to the specified file immediately when using the
-o <filename> or
-oo <filename> options. There are two workarounds for this problem:
- You can use the bpeek <jobid> command to view the output of a currently running job.
- You can send your output to a file with the typical unix output specifications such as
> <filename>with your executed programs or by specifying output files in programs that support such options.
Jupyter notebooks or other programs fail trying to access a /run directory
The default login shell is BASH. Make sure the following parameter is in your .bashrc file in your home directory, as it prevents a problem where some types of jobs fail when run through the LSF queue. This should be done automatically the first time you log onto DeepSense.
echo 'unset XDG_RUNTIME_DIR' >> ~/.bashrc
This line has been added to the default .bashrc file for new users but older user accounts may need this step to be done manually.
Browser fails to connect to Jupyter Notebooks
On our MacBook Pros, Jupyter notebooks work in Chrome, but don't work in safari. Unfortunately, no error is given. Safari just fails to connect. Please let us know if you have issues with any other browsers, and we can add that info here.
Cannot Install PyTorch dependencies
UnsatisfiableError: The following specifications were found to be in conflict: - powerai-pytorch-prereqs=0.4.1_12295.5cb3523
You may see this error when attempting to install the pytorch dependencies in a local anaconda environment. This error indicates that some of your installed python packages are not compatible with the pytorch prequisites. In particular, we see this error when conda has been updated to version 4.6 (which may sometimes happen when installing the tensorflow dependencies first).
To resolve this problem, create a new environment with a 4.5.x conda version and then install the pytorch dependencies in that environment.
Cannot use Caffe on login node or compute nodes without GPUs
Cuda number of devices: -579579216 Current device id: -579579216 Current device name: [==========] Running 2207 tests from 293 test cases. [----------] Global test environment set-up. [----------] 9 tests from AccuracyLayerTest/0, where TypeParam = caffe::CPUDevice<float> [ RUN ] AccuracyLayerTest/0.TestSetup E0206 15:59:26.604874 7990 common.cpp:121] Cannot create Cublas handle. Cublas won't be available. E0206 15:59:26.611477 7990 common.cpp:128] Cannot create Curand generator. Curand won't be available. F0206 15:59:26.611616 7990 syncedmem.cpp:500] Check failed: error == cudaSuccess (30 vs. 0) unknown error *** Check failure stack trace: ***
You may see this error when attempting to use Caffe on a node without GPUs or a GPU node without specifically requesting a GPU.
To resolve this problem, use a GPU node and request a GPU. Caffe cannot run without an available GPU.
Cannot see GPUs in an LSF job
$ nvidia-smi No devices were found
GPUs must be requested with the
-gpu - option to bsub. See LSF#GPU_Computation for more information.
Nested anaconda environments may cause strange behaviour
Some users have experienced strange behaviour when activating an anaconda environment within another environment. This may include permission errors, loading incorrect versions of software, or strange conflicts when attempting to install packages. If you encounter problems with a nested anaconda environment then first try deactivating all anaconda environments and activating just the desired environment.