Does this support this original Meta LLaMA2?

[realhaik]

I see in the docs that you use Hugging Face llama2 model.
Unfortunately HF model is broken, and produces bad results.
I am looking for a way to improve the parallel performance of the original meta llama2 model, that you download from their servers.
Is there a way to use this software for my needs?

Currently when I run the model using the llama , the parallel performance does not exist. Even if I run the requests in parallel, my 4090 stays below 30% load.