bitnet-5B-v1

I redesigned mergekit so that I could merge the bitnet 3B model (unlike version v0, where I did merge via PyTorch).

In theory, this is compatible with quantization via llama.cpp , but tests are needed.