Open meng-x opened 2 months ago
it seems that train_syncnet_sam.py is not using GPU for training the SyncNet.
torch successfully detected GPU and actually other program work great with the GPU on the same PC.
using Python 3.7 on WSL2(Ubuntu 22) on Windows 11.
using 3s long video clips for training.
using syncnet_batch_size=64 (default) -> shall I make this a larger number to make use of larger GRAM or something?
You can load the model in GPU. in the code in train_syncnet.py