Fine-tune training

[LiuChunlin-eng]

Hello author! Thank you for your contribution. I noticed that you answered before that you need at least 8 or 16 GPUs for training, if we want to try the simplest Finetune VGGT, I would like to ask you what is the minimum configuration? Please tell us about it in general or in detail, it will help us a lot, thank you！

[jytime]

I would suggest using 8 GPUs for fine-tuning

[LanesraL]

Hi, @jytime! Thanks for your contribution! I would like to attempt to reproduce your training results to learn about the large model training process. When processing the dataset, I noticed that the dataset code you provided incorporates data processing into the get_data() —unlike other methods, which typically process the data first and then use a unified data loading method.

I would like to ask two questions: First, could this approach cause the CPU processing speed to fail to keep up with the GPU speed? Second, did you also use this method during your initial training?

Thank you in advance for your time and response.