问题
I'm a new man in Deep learning and Google Colab. I'm trying to run this repo on Google Colab https://github.com/harvitronix/five-video-classification-methods But when I running extract_features.py, after 30 minutes, google Colab suddenly extreme slow, from ~2s/it to ~ 30s/it. As you can see in the picture below. It's my fault or it's a problem of GG Colab and how to fix it?
回答1:
For starters I would like to mention basic troubleshooting steps like check you internet connection and make sure you are on a stable Wi-Fi or wired internet connection.
Also if someone else is also on your network and they are downloading a heavy file (say a movie) ask them to pause/stop it for a while.
Apart from these basic steps I would recommend you to follow the below mentioned if you haven't already:
- Go to Runtime in Menu Bar.
- Select Change runtime type
- Under Hardware accelerator drop down menu select GPU.
- Click on Save
The notebook will buffer for a second or so while showing Allocating on right side of the Menu Bar. This should solve your issue. Apart from that I would like you to go through the basic introduction of using colab by following this link -> Colaboratory Welcome Link
回答2:
If you are using google drive space to load the dataset, firstly download it to local colab notebook runtime space, I simply unzip the dataset into local runtime space and it resulted in quicker executions.
This trick can be applied if you are already using a GPU runtime and still want faster executions.
回答3:
I figure out my problem, it's Colab Problem, K80 GPU Memory decrease from 11.5GB to 500mb (randomly), it depends on Colab and we can't fix it!
回答4:
Google Colab will run slow if you use too much runtime, try having small sessions and also terminate all sessions not in use to speed up. Use a TPU or GPU based on your project.
来源：https://stackoverflow.com/questions/52391657/google-colab-suddenly-slow