Support for W4A16 and W4A8 Quantization in TensorRT Model Optimizer

[david-PHR]

Hello NVIDIA team,​

I noticed that the TensorRT Model Optimizer currently supports W4A16 and W4A8 quantization configurations, as detailed in your quantization configuration documentation.​
nvidia.github.io

However, according to the Best Practices for Choosing Quantization Methods, these configurations are currently deployable only via TensorRT-LLM.​
nvidia.github.io

I would like to inquire if there are plans to extend support for W4A16 and W4A8 quantization to the standard TensorRT backend, beyond just TensorRT-LLM.​

Such support would be highly beneficial for deploying models in environments where only TensorRT is used and not TensorRT-LLM
Thank you for your continued efforts in optimizing model deployment workflows.!