koboldcpp/README.md

llama-for-kobold

A hacky little script from Concedo that exposes llama.cpp function bindings, allowing it to be used via a simulated Kobold API endpoint.

It's not very usable as there is a fundamental flaw with llama.cpp, which causes generation delay to scale linearly with original prompt length. Nobody knows why or really cares much, so I'm just going to publish whatever I have at this point.

If you care, please contribute to this discussion which, if resolved, will actually make this viable.

Considerations

• Don't want to use pybind11 due to dependencies on MSVCC
• ZERO or MINIMAL changes as possible to main.cpp - do not move their function declarations elsewhere!
• Leave main.cpp UNTOUCHED, We want to be able to update the repo and pull any changes automatically.
• No dynamic memory allocation! Setup structs with FIXED (known) shapes and sizes for ALL output fields. Python will ALWAYS provide the memory, we just write to it.
• No external libraries or dependencies. That means no Flask, Pybind and whatever. All You Need Is Python.

Usage

• Windows binaries are provided in the form of llamacpp.dll but if you feel worried go ahead and rebuild it yourself.
• Weights are not included, you can use the llama.cpp quantize.exe to generate them from your official weight files (or download them from...places).
• To run, simply clone the repo and run llama_for_kobold.py [ggml_quant_model.bin] [port], and then connect with Kobold or Kobold Lite.
• By default, you can connect to http://localhost:5001 (you can also use https://lite.koboldai.net/?local=1&port=5001).

License

• The original GGML library and llama.cpp by ggerganov are licensed under the MIT License
• However, Kobold Lite is licensed under the AGPL v3.0 License
• The provided python ctypes bindings in llamacpp.dll are also under the AGPL v3.0 License