Introducing multi-backends (TRT-LLM, vLLM) support for Text Generation Inference