tfenv | Terraform version manager | Infrastructure Automation library

I found the tools tfenv and tfswitch to easily and quickly change the terraform and terragrunt version - and I guess the tools work cause people use it, but for some reason they don't chagne the versions on my computer, after I run let's say

I'm trying to create a symlink and I'm not able to solve this error . Please suggest me a solution on how to solve this error

Code: Creating a symlink for /usr/local/bin/terraform-env/bin/* in folder /usr/local/bin I tried with /usr/local/bin/ (with and without slash)

I am building the tiny yolo cnn model in Keras. I get a strange error. Anyone can help?

Imports

I am using sequence generator in keras to parallelly fetch data from the disk but I am getting a very odd error.

So, this is my code for the sequence generator

I train on a GPU cluster with 8 NVIDIA Tesla P100 GPUs. The script code is based on the TensorFlow Tutorial Convolutional Neural Networks. As training data I create binary files based on the Cifar10 data set, which contain only 5000 of the 50,000 images. I use only one of these files per training session.

Here are some basic data about the training:

• Epoch: 100,000
• Batch size: 128
• Number of training data: 5,000
• Init learning rate: 0.1
• Learning rate decay factor: 0.1
• Number of Epochs per decay 350.0
• No exponential moving average
• ...
• If you need more information please comment

The problem is, if I do the training with more than one GPU, I get a NaN error sooner or later. The loss value then explodes within a few epochs from ~0.4 to values above ~1.e+26 to infinity I think and then comes the NaN error.

So far I have tried the following to identify the origin of the NaN bug, so that I can fix it.

• I added tf.check_numerics() after almost every return of a tensor. (Error messages are below)
• I added tf.add_check_numerics_ops(). The error messages are as incomprehensible to me as those of tf.check_numerics() :-)
• I checked the input data for NaN, data are okay.
• I have reduced the learning rate, so the error comes later.
  • Init learning rate: 0.01
  • Learning rate decay factor: 0.01
• Run simpleP2P (Cuda Examples) and passed the test. (Output are below)
• Modified the TensorFlow Tutorial Code only so far that the tf.train.string_input_producer() gets only 1 file instead of 5 files. filename_queue = tf.train.string_input_producer([/path/traindata.bin]) And set number of training data to 5,000.
• To save the variables instead of cpu:0 (see TensorFlow Tutorial here and here) on gpu:0 and to train only with gpu: 1-7. But that was so slow it wasn't an option and I broke it off. (Maybe I did it wrong, too)
• Only train with 4 GPUs, (0-3 or 4-7), but the NaN error came later.

• Linux kernel: 4.4.72-18.12-default x86_64
• 8x NVIDIA Tesla P100-PCIE-16GB
• Cuda 8.0 - V8.0.61
• TenserFlow 1.4.1
• Python3

tf.check_numerics() error message:

I have problem with activating virtualenv. I'm working on the server and using SSH secure shell.

My final goal is to activate virtualenv and run the latest version of tensorflow

The following is the command lines:

I've worked through a few other errors that've been thrown, but I've never seen this error before and even after doing some research I'm still not sure exactly what the issue is here or how to solve it.

I'm guessing reshaping the data at some point is required, but what i don't get is why this is an issue, or what the sizes of [1,2] and [1,1] actually mean.

The data input into the script is [128 x 128 x 128 ndarray, binary label]

The code I'm using is:

I'm using 128 x 128 x 128 ndarrays as input in to a cnn using:

I'm trying to feed a single 300 x 300 x 300 tensor into a placeholder in tensorflow, and although I'm using imgs = tf.placeholder(tf.float32, [1, 300, 300, 300, 3]) I'm getting the ValueError: