search

JingyunLiang / VRT

VRT: A Video Restoration Transformer (official repository)
https://arxiv.org/abs/2201.12288
Other
1.37k stars 130 forks source link

Request training setting recommendation of ×4 VSR #25

Closed Luciennnnnnn closed 2 years ago

Luciennnnnnn commented 2 years ago

If I only have 2/4 3090s and want to train a model for ×4 VSR, how can I set training parameters effectively? That is no OOM, no large performance drop, mild training time.

For example, there are two parameters of using checkpoint to save Cuda memory, use_checkpoint_attn and use_checkpoint_ffn, which one is the most influence one for training time/memory consumption?

Looking forward to your reply, thank you.

JingyunLiang commented 2 years ago

Using torch.checkpoint will produce exactly the same results. use_checkpoint_attn (recommended if oom) can save more memory than use_checkpoint_ffn.

Luciennnnnnn commented 2 years ago

Do I have to modify channels or depth? If I keep parameters unchanged and use_checkpoint_attn, how long it will take for training on 2 3090s?

JingyunLiang commented 2 years ago

Reducing channels or depth may lead to worse performance. I have no idea how long it takes on 2 3090s.