I have deployed Meta-Llama-3-8b model on my server using Xinference. Everything is working good, however the inference time is about 16 seconds even for simple prompts, also it generating extra and additional information (garbage) which is not required. I would like to have some help in this regard that how to reduce inference and avoid generation of extra information. I have thrice checked the compatibility my hardware, it’s fine to run the model smoothly. I believe that, there must be something I have to change in configuration files, but have no idea how to do it. Please lemme know if someone have any solution about this problem.