How to improve Latency

[ukemamaster]

Hi,
In monologue streaming generation, the model generates 80ms audio in about 95ms time using NVIDIA A30 GPU, with bf16 true.

1. Is there a way to improve the latency to be able to generate real time?
2. What is the Realtime factor on L20 GPU?
3. Does the model support VLLM or NVIDIA TensorRT-LLM serving?