Python bindings?

[richardr1126]

Now that Koboldcpp is one of the best ways to run non-llama models with CuBlas, it is so frustrating having to use ctransformers for Python bindings, running on just CPU. Will we ever see a kobold-cpp-python? Or do you think ctransformers will get CUDA support sooner than that?

[LostRuins]

The python bindings already exist and are usable - although they're more intended for internal use rather than downstream external apps (which are encouraged to use the webapi instead). You can take a look at the koboldcpp.py file inside the repo to see how they are being used from the dll.