Is this similar to fastllm?
https://github.com/ztxz16/fastllm
fastllm targets the GPU, while colibri uses CPU inference only
fastllm targets the GPU, while colibri uses CPU inference only