Combining multiple post-training techniques to achieve most efficient quantized LLMs