MegCC is a deep-learning model compiler with the following features:
Extremely Lightweight Runtime: Only keep the required computation kernel in your binary. e.g., 81KB runtime for MobileNet v1
High Performance: Every operation is carefully optimized by experts
Portable: generate nothing but computation code, easy to compile and use on Linux, Android, TEE, BareMetal
Low Memory Usage while Boot Instantly: Model optimization and memory planning are generated at compile time. Get State-of-the-art level memory usage and spend no extra CPU during inference
这个框架和TVM框架相比起来,有哪些优劣之处?