bytedance / GR-1

Code for "Unleashing Large-Scale Video Generative Pre-training for Visual Robot Manipulation"
Apache License 2.0
95 stars 3 forks source link

How long does it take to train the model? #2

Closed StarCycle closed 4 months ago

StarCycle commented 4 months ago

Hello @bdrhtw,

Thank you for making it open source!

What kind of GPU did you use to train this model? How many GPU hours did you use? I find that you used 8M frames from Ego4D but I am not sure about the actual training cost.

Best, StarCycle

bdrhtw commented 4 months ago

Hi @StarCycle , we used 32 V100 to pretrain and finetune GR-1. Pretraining takes about 4 days; finetuning takes about 1 day.

StarCycle commented 4 months ago

Thank you for your response!

StarCycle commented 4 months ago

@bdrhtw Do you use V100 32G or V100 16G?

bdrhtw commented 4 months ago

We were using V100 32G.