fool
fool
dim(e.g., 1024) is the final output dimsion of multi attention module. dim_head = dim // head_num, when head_num = 1, dim_head is equal to dim. b: batch size n: num_patches...
# 3. Build pytorch Custom CUDA Extensions ./setup.sh
Thank you very much
sorry, I modified the code and converted python2 to python3. But the code still exists many issues, So I give up tring the code.