Closed JosephKJ closed 1 year ago
Hi @JosephKJ,
Thank you for showing interest in our work!
Your comments are addressed below:
How many GPUs were used? We train MaPLe on each dataset with a batch size of 4 using a single NVIDIA A100 GPU.
On average, how long did the training take? Training MaPle on ImageNet for 5 epochs takes 1 hour for a single seed. So results for 3 seeds takes around 3 hours. For all remaining 10 datasets, it combinedly takes around 4 hours (for all 3 seeds) on a single A100 GPU. To ease reproduction of MaPLe results, we have provided training logs for all datasets.
Please let us know incase there are any other queries. Thank you.
Thank you very much for your swift response. Really appreciate your clarification.
Thank you for your amazing works! It would be great if you could comment on the compute requirement for people who are planning to replicate the results.