Training time

[vineet2104]

Thank you for the amazing work!
I am trying to train your model on Coco to replicate some results using the same hardware setup (4 A100 GPUs with batch size =16). Can you please tell me how many days of training did it take you?

[chenxi52]

You can refer to this.

[vineet2104]

Do you use any cuda optimizations to run the code faster? Or maybe training with FP16 or BF16?

[chenxi52]

Do you use any cuda optimizations to run the code faster? Or maybe training with FP16 or BF16?

We haven't utilized them in this repo, but you could experiment with them to observe their impact on training time and the final outcomes.