Congratulations on achieving such outstanding results. I have tried to train the model, unfortunately, it seems that our training server has not been updated for a long time. Each GPU has only 12GB of memory.
Obviously model that cannot be trained in this hardware. At the same time, we hope to run the model on a single card. We can accept a decrease in performance while minimizing memory usage as much as possible.
Can you provide a configuration file for a lightweight model? Any suggestions are welcome.
Congratulations on achieving such outstanding results. I have tried to train the model, unfortunately, it seems that our training server has not been updated for a long time. Each GPU has only 12GB of memory. Obviously model that cannot be trained in this hardware. At the same time, we hope to run the model on a single card. We can accept a decrease in performance while minimizing memory usage as much as possible. Can you provide a configuration file for a lightweight model? Any suggestions are welcome.