Most of the parameters you would include in ollama's ModelFile are things you would pass to llama.cpp using command line flags: https://github.com/ggml-org/lla

Most of the parameters you would include in ollama's ModelFile are things you would pass to llama.cpp using command line flags: https://github.com/ggml-org/llama.cpp/blob/master/examples/main/README.md https://github.com/ggml-org/llama.cpp/blob/master/examples/m... If you only ever have one set of configuration parameters per model (same temp, top_p, system prompt...), then I guess you can put them in a gguf file (as the format is extensible). But what if you want two different sets? You still need to keep them somewhere. That could be a shell script for llama.cpp, or a ModelFile for ollama. (Assuming you don't want to create a new (massive) gguf file for each permutation of parameters.)