harrisonvanderbyl / rwkv-cpp-accelerated

A torchless, c++ rwkv implementation using 8bit quantization, written in cuda/hip/vulkan for maximum compatibility and minimum dependencies
MIT License
303 stars 19 forks source link

Add Makefile #37

Closed Jipok closed 11 months ago

Jipok commented 11 months ago

The build scripts are a little broken, because for the correct execution of amd.sh or vulkan.sh you need a realese and vocab folder, which are created only in build.sh. It's not obvious, and of course running build.sh on non-nvidia systems causes an error. I wrote a makefile that fixes this problem and now the scripts are no longer needed.