I tried to finetune codellama model using CURlora. As the architecture is having Q, K, V attention, so i did some changes. But in “select_indices_with_replacement(probs,k) function, i face the errors as it returns numpy array, and numpy array cannot be used with DTensor. The wieght metrics A is a DTensor. Could you please help in finding the solution for this.
I tried to finetune codellama model using CURlora. As the architecture is having Q, K, V attention, so i did some changes. But in “select_indices_with_replacement(probs,k) function, i face the errors as it returns numpy array, and numpy array cannot be used with DTensor. The wieght metrics A is a DTensor. Could you please help in finding the solution for this.