Question about training hardware (GPU and VRAM)

[adabrh]

Thank you for making this excellent work and its implementation publicly available. I really enjoyed reading the paper and appreciate the effort you've put into open-sourcing the project.

I have a quick question regarding the training setup. Could you please share:

• GPU model(s)
• Number of GPUs
• VRAM per GPU
• Approximate training time (if available)

I'm planning to reproduce and extend your work, so this information would be very helpful for estimating the required computational resources.

Thank you!