Is it possible to release the performance benchmark on LLaDA dense model like LLaDA-1.5 or 8B-instruct? Thx~