Since the graph processing library is all about cl-waffe2, I feel that building the entire backend from scratch is reinventing the wheel. I don't know which is the best, but the following list is the choice;
[ ] GGML Wrapper and Interop
[ ] oneDNN (Most Promising for CPU)
[ ] with the power of oneDNN, we can provide bfloat16 training and uint8 inference
Since the graph processing library is all about cl-waffe2, I feel that building the entire backend from scratch is reinventing the wheel. I don't know which is the best, but the following list is the choice;