[Performance] When using per-channel quantization for models with large weight sizes, the process can be extremely slow

[Performance] When using per-channel quantization for models with large weight sizes, the process can be extremely slow #5175