Problem about quantization using PTSQ: RuntimeError: "rsqrt_cpu" not implemented for "Half"

float16 operations are mostly not supported on CPU which is why the error is raised.
I’m not familiar with your use case, but you might want to use the GPU if you are interested in using mixed-precision training with float16.
PS: you can post code snippets by wrapping them into three backticks ```, which would make debugging easier as the images are quite hard to read.