Context params with the given window and batch size. nCtx = 0 uses the model's training length.
A llama_context that frees itself on destruction.
See Source File