Not unload a model from VRAM anymore.

[berkut1]

The last version 0.1.66, not unload model from VRAM.
I thought the problem was with oobabooga/text-generation-webui oobabooga/text-generation-webui#2920, but after digging into the code from projects, I think the problem is with this library.
I think there

llama-cpp-python/llama_cpp/llama.py

Line 1439 in 442213b

def __del__(self):

should be added a new llama_free_model function.

Sorry if I'm wrong, I don't work with python.

[abetlen]

@berkut1 thank you for caching that, will have publish a new version shortly

[iactix]

It still only "deletes" the model though, VRAM still ends up a lot higher than before loading the model.
#223