My graphics card is Radeon 6800xt, and my ROCm version is 6.3.0.
I tried to run LLM in Ubuntu 22.04 LTS with the environment of pytorch(Nightly)+ROCm6.3.0, but it seems to have a very low effciency, lower than my 4060(laptop)gpu and even 6800xt itself on windows with Runtime llama.cpp(ROCm)
I firstly install the wheel with the requirements.txt of the projects, and run "pip3 uninstall torch", than install pytorch(Nightly)+ROCm6.3.0 from pytorch.org
When I‘m doing inference I find some CPU thread is about 100%load, and my graphics card usage is also very high, It just work not slow.
And how do I fix it to make it work efficiently?
