switch to ZeroGPU: transformers pipeline, drop llama-cpp-python

model update

update

first