Spaces:
Running
on
CPU Upgrade
Determining Minimum GPU Memory and Input Text Length Calculation in Model Training
Hello there,
I was wondering if it would be possible to consider the result obtained from this tool as an estimation of the minimum GPU memory required for model training when the batch_size is set to 1. Additionally, I would appreciate it if you could provide some information regarding the length of the input text used to calculate this result.
Thank you kindly.
I am also concerned about it.
As noted in the description, the values that are shown are the minimum recommended vRAM needed for a model is denoted as the size of the “largest layer”, and training of a model is roughly 4x its size (for Adam). This tool is still under development and we will add more variables (batch_size, ...) in the future. To have a better understanding on how much VRAM is needed, I suggest you to read this article: https://blog.eleuther.ai/transformer-math/