In case you are looking for a cutting edge addition #23
Description
If you want to leapfrog over other c# toolkits. Beyond just inference. The path forward would be to c# the LoRA/QLoRA training toolchain. This is better way than RAG. Can be done on moderate capable computers. Avoids needing an A/H100 for a true fine-tune phase. Avoids needing a GPU cloud. Can load these lora's into your base model which you can specialize/augment it without altering the base model. no need to have the weights. Currently it all has to be done within a python environment/swift/oobabooga which has all the problems we are solving using a c# environment in the first place.