Minimal version of a gpu monitor for the deep-gpu-* machines.
Uses the tool from https://github.com/rossumai/nvgpu with an additional column for memory usage, added as in this PR.
Updates /data/theory/robustopt/gpu_monitor/gpu_monitor.html which is served on http://data.csail.mit.edu/madrylab_deep_gpu/gpu_monitor.html
On each machine run FLASK_APP=nvgpu.webapp nohup flask run --host 0.0.0.0 --port 1080 >> flask-deep-gpu-<n>.log 2>&1 &
from /data/theory/robustopt/gpu_monitor/flask_logs
Then run (on some mahcine) nohup python nvgpu_combine.py >> flask_logs/flask-master.log 2>&1 & from /data/theory/robustopt/gpu_monitor/ (the original combination using flask from the repo didn't work for me, hence this combine script).