Questions about how to improve model training speed about itransformer HOT 2 CLOSED

Thank you very much for your guidance! We will take a closer look at the environment configuration. I wish you better paper output! Thank you again for taking the time to reply, thank you very much! :)

from itransformer.

Hello, thank you for your feedback:

• Regarding the first question: Using Mixed Precision Training will not significantly affect performance.
• The second question: It seems the training time you encountered far exceeds on the Traffic dataset in our paper. In fact, iTransformer does not require multiple GPU training to reach a considerable speed (compared to models such as PatchTST that use CI). Please check if your system environment supports CUDA and normal GPU training, and if there are any issues with your GPU device.
• The third issue: GPU memory bit width may need to be prioritized, and CPU performance requirements may only be related to dataset size.

We are not professional machine learning engineers, the above suggestions are for reference only. Thank you again for your feedback!

from itransformer.