How To: Dynamic Batch, Low Bit Quantization Calibration, FP16 Inference, Training via TVM

[kalcohol]

Most wanted:
Dynamic Batch
Low Bit Quantization Calibration
FP16 Inference
Training via TVM

Low level:
Adapt to a new CPU/GPU Architecture, such as MIPS, Specialization RISC-V(with special SIMD implementation).

[kalcohol]

The more C++ courses, the less injuries caused.

[kalcohol]

add neural architecture search to the list?

[yidawang]

It depends on the latest advance of the area. We will cover some on-going research topics at the end.