This article briefly summarizes the process of Jiutian Bisheng from creating a project to building an environment.
1. Upload dataset
Arrive at the following page, just wait
2. Create an instance
After creation, jump to the "Single-machine training" page, and the project will start automatically, showing "Starting"
Wait for the startup, after the startup is complete, as shown in the figure below
Click Jupyter to enter the project (I choose Jupyter, or VSCcode, but I haven't used it yet)
3. Project environment configuration
enter project
Click on the right panel, Terminal in Tther, and you can download various packages (of course, downloading in the jpynb file is also fine, the simple pip is fine, you can refer to the official document for details, that is, the one that appears after creating the project ReadMe document)
Enter the terminal:
Here, I want to configure pytorch
First create a virtual environment, and then download pytorch in this environment
(1) Create a virtual environment
conda create -n bertsum python=3.7
(2) Download pytorch
Check version nvcc -V
Then find the corresponding torch download code according to the version
Available here: Previous PyTorch Versions | PyTorch