MemoryFree | Arduino MemoryFree library

I am using Google Colaboratory to train an image recognition algorithm, using TensorFlow 1.15. I have uploaded all needed files into Google Drive, and have gotten the code to run until the shuffle buffer finishes running. However, I get a "^C" in the dialog box, and cannot figure out what is going on.

Note: I have previously tried to train the algorithm on my PC, and did not delete the checkpoint files that were generated from the previous training session. Could that perhaps be the problem?

Code:

I have been following this stack overflow article : Accurate calculation of CPU usage given in percentage in Linux?

It is written in different language so I decided to follow the logic and convert it to C#.

I am running a Convnet on colab Pro GPU. I have selected GPU in my runtime and can confirm that GPU is available. I am running exactly the same network as yesterday evening, but it is taking about 2 hours per epoch... last night it took about 3 minutes per epoch... nothing has changed at all. I have a feeling colab may have restricted my GPU usage but I can't work out how to tell if this is the issue. Does GPU speed fluctuate much depending on time of day etc? Here are some diagnostics which I have printed, does anyone know how I can investigate deeper what the root cause of this slow behaviour is?

I also tried changing to accelerator in colab to 'None', and my network was the same speed as with 'GPU' selected, implying that for some reason i am no longer training on GPU, or resources have been severely limited. I am using Tensorflow 2.1.

I'm running this code on my Arduino Uno:

update: this question is related to Google Colab's "Notebook settings: Hardware accelerator: GPU". This question was written before the "TPU" option was added.

Reading multiple excited announcements about Google Colaboratory providing free Tesla K80 GPU, I tried to run fast.ai lesson on it for it to never complete - quickly running out of memory. I started investigating of why.

The bottom line is that “free Tesla K80” is not "free" for all - for some only a small slice of it is "free".

I connect to Google Colab from West Coast Canada and I get only 0.5GB of what supposed to be a 24GB GPU RAM. Other users get access to 11GB of GPU RAM.

Clearly 0.5GB GPU RAM is insufficient for most ML/DL work.

If you're not sure what you get, here is little debug function I scraped together (only works with the GPU setting of the notebook):

I programmed rather long Arduino application and when I tried to run it on my Arduino Nano(ATmega328P CPU),I found it crashed when it turns to "Serial.println(F("ESP8266_PHP_CONNECTED"));"(Line 462), just stopped there and printed abnormal thing such as "ESP8266_PHP_CONNECTBB", I don't know why did it happened even if the SRAM was enough.this is what I read from monitor

I'm using Arduino mega with MD5 library and ESP8266 with SoftwareSerial. The problem is that after 370 loops, Arduino auto-restart due to lack of memory. I used FreeMemory to troubleshoot, I noticed the problem is the decreasing of available memory with each loop. It is weird behavior, because it only appears when I use AT commands along with MD5, however if I separate the sketch in two sketches they work properly without memory problems. My original sketch is a quite more complex, but I reduced to the essential code the example showed bellow in order to be more clear, the behavior is the same, so if i fix it i will be able to fix the my original sketch